diff --git a/src/diffusers/pipelines/alt_diffusion/pipeline_alt_diffusion.py b/src/diffusers/pipelines/alt_diffusion/pipeline_alt_diffusion.py
index 5a4746d24e06..507c082d9363 100644
--- a/src/diffusers/pipelines/alt_diffusion/pipeline_alt_diffusion.py
+++ b/src/diffusers/pipelines/alt_diffusion/pipeline_alt_diffusion.py
@@ -226,31 +226,6 @@ def disable_vae_tiling(self):
         """
         self.vae.disable_tiling()
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        Note that offloading happens on a submodule basis. Memory savings are higher than with
-        `enable_model_cpu_offload`, but performance is lower.
-        """
-        if is_accelerate_available() and is_accelerate_version(">=", "0.14.0"):
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("`enable_sequential_cpu_offload` requires `accelerate v0.14.0` or higher")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        if self.device.type != "cpu":
-            self.to("cpu", silence_dtype_warnings=True)
-            torch.cuda.empty_cache()  # otherwise we don't see the memory savings (but they probably exist)
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae]:
-            cpu_offload(cpu_offloaded_model, device)
-
-        if self.safety_checker is not None:
-            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
-
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
         Offloads all models to CPU using accelerate, reducing memory usage with a low impact on performance. Compared
@@ -279,24 +254,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     def _encode_prompt(
         self,
         prompt,
diff --git a/src/diffusers/pipelines/alt_diffusion/pipeline_alt_diffusion_img2img.py b/src/diffusers/pipelines/alt_diffusion/pipeline_alt_diffusion_img2img.py
index 21c1f0591a44..d6f297122ba4 100644
--- a/src/diffusers/pipelines/alt_diffusion/pipeline_alt_diffusion_img2img.py
+++ b/src/diffusers/pipelines/alt_diffusion/pipeline_alt_diffusion_img2img.py
@@ -224,31 +224,6 @@ def __init__(
         self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
         self.register_to_config(requires_safety_checker=requires_safety_checker)
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        Note that offloading happens on a submodule basis. Memory savings are higher than with
-        `enable_model_cpu_offload`, but performance is lower.
-        """
-        if is_accelerate_available() and is_accelerate_version(">=", "0.14.0"):
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("`enable_sequential_cpu_offload` requires `accelerate v0.14.0` or higher")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        if self.device.type != "cpu":
-            self.to("cpu", silence_dtype_warnings=True)
-            torch.cuda.empty_cache()  # otherwise we don't see the memory savings (but they probably exist)
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae]:
-            cpu_offload(cpu_offloaded_model, device)
-
-        if self.safety_checker is not None:
-            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
-
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
         Offloads all models to CPU using accelerate, reducing memory usage with a low impact on performance. Compared
@@ -277,24 +252,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     def _encode_prompt(
         self,
         prompt,
diff --git a/src/diffusers/pipelines/audioldm/pipeline_audioldm.py b/src/diffusers/pipelines/audioldm/pipeline_audioldm.py
index 88bfd8fe73e2..6da8e809103e 100644
--- a/src/diffusers/pipelines/audioldm/pipeline_audioldm.py
+++ b/src/diffusers/pipelines/audioldm/pipeline_audioldm.py
@@ -22,7 +22,7 @@
 
 from ...models import AutoencoderKL, UNet2DConditionModel
 from ...schedulers import KarrasDiffusionSchedulers
-from ...utils import is_accelerate_available, logging, randn_tensor, replace_example_docstring
+from ...utils import logging, randn_tensor, replace_example_docstring
 from ..pipeline_utils import AudioPipelineOutput, DiffusionPipeline
 
 
@@ -108,41 +108,6 @@ def disable_vae_slicing(self):
         """
         self.vae.disable_slicing()
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and vocoder have their state dicts saved to CPU and then are moved to a `torch.device('meta')
-        and loaded to GPU only when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae, self.vocoder]:
-            cpu_offload(cpu_offloaded_model, device)
-
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     def _encode_prompt(
         self,
         prompt,
diff --git a/src/diffusers/pipelines/consistency_models/pipeline_consistency_models.py b/src/diffusers/pipelines/consistency_models/pipeline_consistency_models.py
index 4e72e3fdbafe..ec4af7afe5ad 100644
--- a/src/diffusers/pipelines/consistency_models/pipeline_consistency_models.py
+++ b/src/diffusers/pipelines/consistency_models/pipeline_consistency_models.py
@@ -76,31 +76,6 @@ def __init__(self, unet: UNet2DModel, scheduler: CMStochasticIterativeScheduler)
 
         self.safety_checker = None
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        Note that offloading happens on a submodule basis. Memory savings are higher than with
-        `enable_model_cpu_offload`, but performance is lower.
-        """
-        if is_accelerate_available() and is_accelerate_version(">=", "0.14.0"):
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("`enable_sequential_cpu_offload` requires `accelerate v0.14.0` or higher")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        if self.device.type != "cpu":
-            self.to("cpu", silence_dtype_warnings=True)
-            torch.cuda.empty_cache()  # otherwise we don't see the memory savings (but they probably exist)
-
-        for cpu_offloaded_model in [self.unet]:
-            cpu_offload(cpu_offloaded_model, device)
-
-        if self.safety_checker is not None:
-            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
-
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
         Offloads all models to CPU using accelerate, reducing memory usage with a low impact on performance. Compared
@@ -129,25 +104,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     def prepare_latents(self, batch_size, num_channels, height, width, dtype, device, generator, latents=None):
         shape = (batch_size, num_channels, height, width)
         if isinstance(generator, list) and len(generator) != batch_size:
diff --git a/src/diffusers/pipelines/controlnet/pipeline_controlnet.py b/src/diffusers/pipelines/controlnet/pipeline_controlnet.py
index c266e8b20e74..dc6e22a7d8a8 100644
--- a/src/diffusers/pipelines/controlnet/pipeline_controlnet.py
+++ b/src/diffusers/pipelines/controlnet/pipeline_controlnet.py
@@ -212,27 +212,6 @@ def disable_vae_tiling(self):
         """
         self.vae.disable_tiling()
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae, controlnet, and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        Note that offloading happens on a submodule basis. Memory savings are higher than with
-        `enable_model_cpu_offload`, but performance is lower.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae, self.controlnet]:
-            cpu_offload(cpu_offloaded_model, device)
-
-        if self.safety_checker is not None:
-            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
-
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
         Offloads all models to CPU using accelerate, reducing memory usage with a low impact on performance. Compared
@@ -261,25 +240,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._encode_prompt
     def _encode_prompt(
         self,
diff --git a/src/diffusers/pipelines/controlnet/pipeline_controlnet_img2img.py b/src/diffusers/pipelines/controlnet/pipeline_controlnet_img2img.py
index fd013c4974f1..30eba4fa8329 100644
--- a/src/diffusers/pipelines/controlnet/pipeline_controlnet_img2img.py
+++ b/src/diffusers/pipelines/controlnet/pipeline_controlnet_img2img.py
@@ -238,27 +238,6 @@ def disable_vae_tiling(self):
         """
         self.vae.disable_tiling()
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae, controlnet, and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        Note that offloading happens on a submodule basis. Memory savings are higher than with
-        `enable_model_cpu_offload`, but performance is lower.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae, self.controlnet]:
-            cpu_offload(cpu_offloaded_model, device)
-
-        if self.safety_checker is not None:
-            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
-
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
         Offloads all models to CPU using accelerate, reducing memory usage with a low impact on performance. Compared
@@ -287,25 +266,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._encode_prompt
     def _encode_prompt(
         self,
diff --git a/src/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint.py b/src/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint.py
index 7de3f1dd9d88..2d850f204319 100644
--- a/src/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint.py
+++ b/src/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint.py
@@ -355,27 +355,6 @@ def disable_vae_tiling(self):
         """
         self.vae.disable_tiling()
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae, controlnet, and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        Note that offloading happens on a submodule basis. Memory savings are higher than with
-        `enable_model_cpu_offload`, but performance is lower.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae, self.controlnet]:
-            cpu_offload(cpu_offloaded_model, device)
-
-        if self.safety_checker is not None:
-            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
-
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
         Offloads all models to CPU using accelerate, reducing memory usage with a low impact on performance. Compared
@@ -404,25 +383,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._encode_prompt
     def _encode_prompt(
         self,
diff --git a/src/diffusers/pipelines/dance_diffusion/pipeline_dance_diffusion.py b/src/diffusers/pipelines/dance_diffusion/pipeline_dance_diffusion.py
index 8dd63fcc7f04..c3eb32273b6d 100644
--- a/src/diffusers/pipelines/dance_diffusion/pipeline_dance_diffusion.py
+++ b/src/diffusers/pipelines/dance_diffusion/pipeline_dance_diffusion.py
@@ -102,7 +102,7 @@ def __call__(
                 f" size of {batch_size}. Make sure the batch size matches the length of the generators."
             )
 
-        audio = randn_tensor(shape, generator=generator, device=self.device, dtype=dtype)
+        audio = randn_tensor(shape, generator=generator, device=self._execution_device, dtype=dtype)
 
         # set step values
         self.scheduler.set_timesteps(num_inference_steps, device=audio.device)
diff --git a/src/diffusers/pipelines/ddim/pipeline_ddim.py b/src/diffusers/pipelines/ddim/pipeline_ddim.py
index aaf53589b969..c24aa6c79793 100644
--- a/src/diffusers/pipelines/ddim/pipeline_ddim.py
+++ b/src/diffusers/pipelines/ddim/pipeline_ddim.py
@@ -95,7 +95,7 @@ def __call__(
                 f" size of {batch_size}. Make sure the batch size matches the length of the generators."
             )
 
-        image = randn_tensor(image_shape, generator=generator, device=self.device, dtype=self.unet.dtype)
+        image = randn_tensor(image_shape, generator=generator, device=self._execution_device, dtype=self.unet.dtype)
 
         # set step values
         self.scheduler.set_timesteps(num_inference_steps)
diff --git a/src/diffusers/pipelines/deepfloyd_if/pipeline_if.py b/src/diffusers/pipelines/deepfloyd_if/pipeline_if.py
index cd1015dc03bb..ffcb1ab32d35 100644
--- a/src/diffusers/pipelines/deepfloyd_if/pipeline_if.py
+++ b/src/diffusers/pipelines/deepfloyd_if/pipeline_if.py
@@ -144,30 +144,6 @@ def __init__(
         )
         self.register_to_config(requires_safety_checker=requires_safety_checker)
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, the pipeline's
-        models have their state dicts saved to CPU and then are moved to a `torch.device('meta') and loaded to GPU only
-        when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        models = [
-            self.text_encoder,
-            self.unet,
-        ]
-        for cpu_offloaded_model in models:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
-        if self.safety_checker is not None:
-            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
-
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
         Offloads all models to CPU using accelerate, reducing memory usage with a low impact on performance. Compared
@@ -223,25 +199,6 @@ def remove_all_hooks(self):
         self.text_encoder_offload_hook = None
         self.final_offload_hook = None
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     @torch.no_grad()
     def encode_prompt(
         self,
diff --git a/src/diffusers/pipelines/deepfloyd_if/pipeline_if_img2img.py b/src/diffusers/pipelines/deepfloyd_if/pipeline_if_img2img.py
index 6bae2071173b..9c4c5229e1c4 100644
--- a/src/diffusers/pipelines/deepfloyd_if/pipeline_if_img2img.py
+++ b/src/diffusers/pipelines/deepfloyd_if/pipeline_if_img2img.py
@@ -168,31 +168,6 @@ def __init__(
         )
         self.register_to_config(requires_safety_checker=requires_safety_checker)
 
-    # Copied from diffusers.pipelines.deepfloyd_if.pipeline_if.IFPipeline.enable_sequential_cpu_offload
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, the pipeline's
-        models have their state dicts saved to CPU and then are moved to a `torch.device('meta') and loaded to GPU only
-        when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        models = [
-            self.text_encoder,
-            self.unet,
-        ]
-        for cpu_offloaded_model in models:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
-        if self.safety_checker is not None:
-            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
-
     # Copied from diffusers.pipelines.deepfloyd_if.pipeline_if.IFPipeline.enable_model_cpu_offload
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
@@ -250,25 +225,6 @@ def remove_all_hooks(self):
         self.text_encoder_offload_hook = None
         self.final_offload_hook = None
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     @torch.no_grad()
     # Copied from diffusers.pipelines.deepfloyd_if.pipeline_if.IFPipeline.encode_prompt
     def encode_prompt(
diff --git a/src/diffusers/pipelines/deepfloyd_if/pipeline_if_img2img_superresolution.py b/src/diffusers/pipelines/deepfloyd_if/pipeline_if_img2img_superresolution.py
index 0ee9c6ba331d..1ea0bafa51a0 100644
--- a/src/diffusers/pipelines/deepfloyd_if/pipeline_if_img2img_superresolution.py
+++ b/src/diffusers/pipelines/deepfloyd_if/pipeline_if_img2img_superresolution.py
@@ -179,31 +179,6 @@ def __init__(
         )
         self.register_to_config(requires_safety_checker=requires_safety_checker)
 
-    # Copied from diffusers.pipelines.deepfloyd_if.pipeline_if.IFPipeline.enable_sequential_cpu_offload
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, the pipeline's
-        models have their state dicts saved to CPU and then are moved to a `torch.device('meta') and loaded to GPU only
-        when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        models = [
-            self.text_encoder,
-            self.unet,
-        ]
-        for cpu_offloaded_model in models:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
-        if self.safety_checker is not None:
-            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
-
     # Copied from diffusers.pipelines.deepfloyd_if.pipeline_if.IFPipeline.enable_model_cpu_offload
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
@@ -401,25 +376,6 @@ def _clean_caption(self, caption):
 
         return caption.strip()
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     @torch.no_grad()
     # Copied from diffusers.pipelines.deepfloyd_if.pipeline_if.IFPipeline.encode_prompt
     def encode_prompt(
diff --git a/src/diffusers/pipelines/deepfloyd_if/pipeline_if_inpainting.py b/src/diffusers/pipelines/deepfloyd_if/pipeline_if_inpainting.py
index 9c1f71126ac5..b11bf780de47 100644
--- a/src/diffusers/pipelines/deepfloyd_if/pipeline_if_inpainting.py
+++ b/src/diffusers/pipelines/deepfloyd_if/pipeline_if_inpainting.py
@@ -171,31 +171,6 @@ def __init__(
         )
         self.register_to_config(requires_safety_checker=requires_safety_checker)
 
-    # Copied from diffusers.pipelines.deepfloyd_if.pipeline_if.IFPipeline.enable_sequential_cpu_offload
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, the pipeline's
-        models have their state dicts saved to CPU and then are moved to a `torch.device('meta') and loaded to GPU only
-        when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        models = [
-            self.text_encoder,
-            self.unet,
-        ]
-        for cpu_offloaded_model in models:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
-        if self.safety_checker is not None:
-            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
-
     # Copied from diffusers.pipelines.deepfloyd_if.pipeline_if.IFPipeline.enable_model_cpu_offload
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
@@ -253,25 +228,6 @@ def remove_all_hooks(self):
         self.text_encoder_offload_hook = None
         self.final_offload_hook = None
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     @torch.no_grad()
     # Copied from diffusers.pipelines.deepfloyd_if.pipeline_if.IFPipeline.encode_prompt
     def encode_prompt(
diff --git a/src/diffusers/pipelines/deepfloyd_if/pipeline_if_inpainting_superresolution.py b/src/diffusers/pipelines/deepfloyd_if/pipeline_if_inpainting_superresolution.py
index 6a90f2b765d4..2971570aada2 100644
--- a/src/diffusers/pipelines/deepfloyd_if/pipeline_if_inpainting_superresolution.py
+++ b/src/diffusers/pipelines/deepfloyd_if/pipeline_if_inpainting_superresolution.py
@@ -181,31 +181,6 @@ def __init__(
         )
         self.register_to_config(requires_safety_checker=requires_safety_checker)
 
-    # Copied from diffusers.pipelines.deepfloyd_if.pipeline_if.IFPipeline.enable_sequential_cpu_offload
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, the pipeline's
-        models have their state dicts saved to CPU and then are moved to a `torch.device('meta') and loaded to GPU only
-        when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        models = [
-            self.text_encoder,
-            self.unet,
-        ]
-        for cpu_offloaded_model in models:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
-        if self.safety_checker is not None:
-            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
-
     # Copied from diffusers.pipelines.deepfloyd_if.pipeline_if.IFPipeline.enable_model_cpu_offload
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
@@ -403,25 +378,6 @@ def _clean_caption(self, caption):
 
         return caption.strip()
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     @torch.no_grad()
     # Copied from diffusers.pipelines.deepfloyd_if.pipeline_if.IFPipeline.encode_prompt
     def encode_prompt(
diff --git a/src/diffusers/pipelines/deepfloyd_if/pipeline_if_superresolution.py b/src/diffusers/pipelines/deepfloyd_if/pipeline_if_superresolution.py
index 86d9574b97e1..8bdd01fe748e 100644
--- a/src/diffusers/pipelines/deepfloyd_if/pipeline_if_superresolution.py
+++ b/src/diffusers/pipelines/deepfloyd_if/pipeline_if_superresolution.py
@@ -137,31 +137,6 @@ def __init__(
         )
         self.register_to_config(requires_safety_checker=requires_safety_checker)
 
-    # Copied from diffusers.pipelines.deepfloyd_if.pipeline_if.IFPipeline.enable_sequential_cpu_offload
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, the pipeline's
-        models have their state dicts saved to CPU and then are moved to a `torch.device('meta') and loaded to GPU only
-        when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        models = [
-            self.text_encoder,
-            self.unet,
-        ]
-        for cpu_offloaded_model in models:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
-        if self.safety_checker is not None:
-            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
-
     # Copied from diffusers.pipelines.deepfloyd_if.pipeline_if.IFPipeline.enable_model_cpu_offload
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
@@ -359,25 +334,6 @@ def _clean_caption(self, caption):
 
         return caption.strip()
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     @torch.no_grad()
     # Copied from diffusers.pipelines.deepfloyd_if.pipeline_if.IFPipeline.encode_prompt
     def encode_prompt(
diff --git a/src/diffusers/pipelines/dit/pipeline_dit.py b/src/diffusers/pipelines/dit/pipeline_dit.py
index f0d30697af43..07fd2835ccf0 100644
--- a/src/diffusers/pipelines/dit/pipeline_dit.py
+++ b/src/diffusers/pipelines/dit/pipeline_dit.py
@@ -121,13 +121,13 @@ def __call__(
         latents = randn_tensor(
             shape=(batch_size, latent_channels, latent_size, latent_size),
             generator=generator,
-            device=self.device,
+            device=self._execution_device,
             dtype=self.transformer.dtype,
         )
         latent_model_input = torch.cat([latents] * 2) if guidance_scale > 1 else latents
 
-        class_labels = torch.tensor(class_labels, device=self.device).reshape(-1)
-        class_null = torch.tensor([1000] * batch_size, device=self.device)
+        class_labels = torch.tensor(class_labels, device=self._execution_device).reshape(-1)
+        class_null = torch.tensor([1000] * batch_size, device=self._execution_device)
         class_labels_input = torch.cat([class_labels, class_null], 0) if guidance_scale > 1 else class_labels
 
         # set step values
diff --git a/src/diffusers/pipelines/kandinsky/pipeline_kandinsky.py b/src/diffusers/pipelines/kandinsky/pipeline_kandinsky.py
index 2b4c6b963598..31edb915e8f7 100644
--- a/src/diffusers/pipelines/kandinsky/pipeline_kandinsky.py
+++ b/src/diffusers/pipelines/kandinsky/pipeline_kandinsky.py
@@ -229,28 +229,6 @@ def _encode_prompt(
 
         return prompt_embeds, text_encoder_hidden_states, text_mask
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, the pipeline's
-        models have their state dicts saved to CPU and then are moved to a `torch.device('meta') and loaded to GPU only
-        when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        models = [
-            self.unet,
-            self.text_encoder,
-            self.movq,
-        ]
-        for cpu_offloaded_model in models:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
         Offloads all models to CPU using accelerate, reducing memory usage with a low impact on performance. Compared
@@ -276,25 +254,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     @torch.no_grad()
     @replace_example_docstring(EXAMPLE_DOC_STRING)
     def __call__(
diff --git a/src/diffusers/pipelines/kandinsky/pipeline_kandinsky_img2img.py b/src/diffusers/pipelines/kandinsky/pipeline_kandinsky_img2img.py
index 8d8d85ebb7b6..e3e4b04febac 100644
--- a/src/diffusers/pipelines/kandinsky/pipeline_kandinsky_img2img.py
+++ b/src/diffusers/pipelines/kandinsky/pipeline_kandinsky_img2img.py
@@ -264,28 +264,6 @@ def _encode_prompt(
 
         return prompt_embeds, text_encoder_hidden_states, text_mask
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, the pipeline's
-        models have their state dicts saved to CPU and then are moved to a `torch.device('meta') and loaded to GPU only
-        when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        models = [
-            self.unet,
-            self.text_encoder,
-            self.movq,
-        ]
-        for cpu_offloaded_model in models:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
         Offloads all models to CPU using accelerate, reducing memory usage with a low impact on performance. Compared
@@ -311,25 +289,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     #  add_noise method to overwrite the one in schedule because it use a different beta schedule for adding noise vs sampling
     def add_noise(
         self,
diff --git a/src/diffusers/pipelines/kandinsky/pipeline_kandinsky_inpaint.py b/src/diffusers/pipelines/kandinsky/pipeline_kandinsky_inpaint.py
index cc41e9eb4fc8..a4a7dc225947 100644
--- a/src/diffusers/pipelines/kandinsky/pipeline_kandinsky_inpaint.py
+++ b/src/diffusers/pipelines/kandinsky/pipeline_kandinsky_inpaint.py
@@ -389,28 +389,6 @@ def _encode_prompt(
 
         return prompt_embeds, text_encoder_hidden_states, text_mask
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, the pipeline's
-        models have their state dicts saved to CPU and then are moved to a `torch.device('meta') and loaded to GPU only
-        when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        models = [
-            self.unet,
-            self.text_encoder,
-            self.movq,
-        ]
-        for cpu_offloaded_model in models:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
         Offloads all models to CPU using accelerate, reducing memory usage with a low impact on performance. Compared
@@ -436,25 +414,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     @torch.no_grad()
     @replace_example_docstring(EXAMPLE_DOC_STRING)
     def __call__(
diff --git a/src/diffusers/pipelines/kandinsky/pipeline_kandinsky_prior.py b/src/diffusers/pipelines/kandinsky/pipeline_kandinsky_prior.py
index 9322ff6080d1..4d99180b3d11 100644
--- a/src/diffusers/pipelines/kandinsky/pipeline_kandinsky_prior.py
+++ b/src/diffusers/pipelines/kandinsky/pipeline_kandinsky_prior.py
@@ -25,7 +25,6 @@
 from ...schedulers import UnCLIPScheduler
 from ...utils import (
     BaseOutput,
-    is_accelerate_available,
     logging,
     randn_tensor,
     replace_example_docstring,
@@ -147,6 +146,8 @@ class KandinskyPriorPipeline(DiffusionPipeline):
             A scheduler to be used in combination with `prior` to generate image embedding.
     """
 
+    _exclude_from_cpu_offload = ["prior"]
+
     def __init__(
         self,
         prior: PriorTransformer,
@@ -295,45 +296,6 @@ def get_zero_embed(self, batch_size=1, device=None):
         zero_image_emb = zero_image_emb.repeat(batch_size, 1)
         return zero_image_emb
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, the pipeline's
-        models have their state dicts saved to CPU and then are moved to a `torch.device('meta') and loaded to GPU only
-        when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        models = [
-            self.image_encoder,
-            self.text_encoder,
-        ]
-        for cpu_offloaded_model in models:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
-    @property
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if self.device != torch.device("meta") or not hasattr(self.text_encoder, "_hf_hook"):
-            return self.device
-        for module in self.text_encoder.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     def _encode_prompt(
         self,
         prompt,
diff --git a/src/diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2.py b/src/diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2.py
index 4a116e1e600b..6db5260b04d8 100644
--- a/src/diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2.py
+++ b/src/diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2.py
@@ -110,27 +110,6 @@ def prepare_latents(self, shape, dtype, device, generator, latents, scheduler):
         latents = latents * scheduler.init_noise_sigma
         return latents
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, the pipeline's
-        models have their state dicts saved to CPU and then are moved to a `torch.device('meta') and loaded to GPU only
-        when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        models = [
-            self.unet,
-            self.movq,
-        ]
-        for cpu_offloaded_model in models:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
         Offloads all models to CPU using accelerate, reducing memory usage with a low impact on performance. Compared
@@ -156,25 +135,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     @torch.no_grad()
     @replace_example_docstring(EXAMPLE_DOC_STRING)
     def __call__(
diff --git a/src/diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_controlnet.py b/src/diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_controlnet.py
index 73fc20b5e0f2..d6fc26911747 100644
--- a/src/diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_controlnet.py
+++ b/src/diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_controlnet.py
@@ -150,28 +150,6 @@ def prepare_latents(self, shape, dtype, device, generator, latents, scheduler):
         latents = latents * scheduler.init_noise_sigma
         return latents
 
-    # Copied from diffusers.pipelines.kandinsky2_2.pipeline_kandinsky2_2.KandinskyV22Pipeline.enable_sequential_cpu_offload
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, the pipeline's
-        models have their state dicts saved to CPU and then are moved to a `torch.device('meta') and loaded to GPU only
-        when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        models = [
-            self.unet,
-            self.movq,
-        ]
-        for cpu_offloaded_model in models:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
     # Copied from diffusers.pipelines.kandinsky2_2.pipeline_kandinsky2_2.KandinskyV22Pipeline.enable_model_cpu_offload
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
@@ -198,25 +176,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     @torch.no_grad()
     @replace_example_docstring(EXAMPLE_DOC_STRING)
     def __call__(
diff --git a/src/diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_controlnet_img2img.py b/src/diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_controlnet_img2img.py
index 3e001e89e490..8a25624b7267 100644
--- a/src/diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_controlnet_img2img.py
+++ b/src/diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_controlnet_img2img.py
@@ -205,28 +205,6 @@ def prepare_latents(self, image, timestep, batch_size, num_images_per_prompt, dt
 
         return latents
 
-    # Copied from diffusers.pipelines.kandinsky2_2.pipeline_kandinsky2_2.KandinskyV22Pipeline.enable_sequential_cpu_offload
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, the pipeline's
-        models have their state dicts saved to CPU and then are moved to a `torch.device('meta') and loaded to GPU only
-        when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        models = [
-            self.unet,
-            self.movq,
-        ]
-        for cpu_offloaded_model in models:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
     # Copied from diffusers.pipelines.kandinsky2_2.pipeline_kandinsky2_2.KandinskyV22Pipeline.enable_model_cpu_offload
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
@@ -253,25 +231,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     @torch.no_grad()
     @replace_example_docstring(EXAMPLE_DOC_STRING)
     def __call__(
diff --git a/src/diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_img2img.py b/src/diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_img2img.py
index 0a5f77b0ff2d..26976ad0c925 100644
--- a/src/diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_img2img.py
+++ b/src/diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_img2img.py
@@ -178,28 +178,6 @@ def prepare_latents(self, image, timestep, batch_size, num_images_per_prompt, dt
 
         return latents
 
-    # Copied from diffusers.pipelines.kandinsky2_2.pipeline_kandinsky2_2.KandinskyV22Pipeline.enable_sequential_cpu_offload
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, the pipeline's
-        models have their state dicts saved to CPU and then are moved to a `torch.device('meta') and loaded to GPU only
-        when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        models = [
-            self.unet,
-            self.movq,
-        ]
-        for cpu_offloaded_model in models:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
     # Copied from diffusers.pipelines.kandinsky2_2.pipeline_kandinsky2_2.KandinskyV22Pipeline.enable_model_cpu_offload
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
@@ -226,25 +204,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     @torch.no_grad()
     @replace_example_docstring(EXAMPLE_DOC_STRING)
     def __call__(
diff --git a/src/diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_inpainting.py b/src/diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_inpainting.py
index 151312979f81..27bd01984dbf 100644
--- a/src/diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_inpainting.py
+++ b/src/diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_inpainting.py
@@ -277,28 +277,6 @@ def prepare_latents(self, shape, dtype, device, generator, latents, scheduler):
         latents = latents * scheduler.init_noise_sigma
         return latents
 
-    # Copied from diffusers.pipelines.kandinsky2_2.pipeline_kandinsky2_2.KandinskyV22Pipeline.enable_sequential_cpu_offload
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, the pipeline's
-        models have their state dicts saved to CPU and then are moved to a `torch.device('meta') and loaded to GPU only
-        when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        models = [
-            self.unet,
-            self.movq,
-        ]
-        for cpu_offloaded_model in models:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
     # Copied from diffusers.pipelines.kandinsky2_2.pipeline_kandinsky2_2.KandinskyV22Pipeline.enable_model_cpu_offload
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
@@ -325,25 +303,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     @torch.no_grad()
     @replace_example_docstring(EXAMPLE_DOC_STRING)
     def __call__(
diff --git a/src/diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_prior.py b/src/diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_prior.py
index 3b9974a5dd70..1bfc6523cdf9 100644
--- a/src/diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_prior.py
+++ b/src/diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_prior.py
@@ -8,7 +8,6 @@
 from ...pipelines import DiffusionPipeline
 from ...schedulers import UnCLIPScheduler
 from ...utils import (
-    is_accelerate_available,
     logging,
     randn_tensor,
     replace_example_docstring,
@@ -105,6 +104,8 @@ class KandinskyV22PriorPipeline(DiffusionPipeline):
             A image_processor to be used to preprocess image from clip.
     """
 
+    _exclude_from_cpu_offload = ["prior"]
+
     def __init__(
         self,
         prior: PriorTransformer,
@@ -254,47 +255,6 @@ def get_zero_embed(self, batch_size=1, device=None):
         zero_image_emb = zero_image_emb.repeat(batch_size, 1)
         return zero_image_emb
 
-    # Copied from diffusers.pipelines.kandinsky.pipeline_kandinsky_prior.KandinskyPriorPipeline.enable_sequential_cpu_offload
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, the pipeline's
-        models have their state dicts saved to CPU and then are moved to a `torch.device('meta') and loaded to GPU only
-        when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        models = [
-            self.image_encoder,
-            self.text_encoder,
-        ]
-        for cpu_offloaded_model in models:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
-    @property
-    # Copied from diffusers.pipelines.kandinsky.pipeline_kandinsky_prior.KandinskyPriorPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if self.device != torch.device("meta") or not hasattr(self.text_encoder, "_hf_hook"):
-            return self.device
-        for module in self.text_encoder.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     # Copied from diffusers.pipelines.kandinsky.pipeline_kandinsky_prior.KandinskyPriorPipeline._encode_prompt
     def _encode_prompt(
         self,
diff --git a/src/diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_prior_emb2emb.py b/src/diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_prior_emb2emb.py
index ae46af9c4551..bcbeb78bac10 100644
--- a/src/diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_prior_emb2emb.py
+++ b/src/diffusers/pipelines/kandinsky2_2/pipeline_kandinsky2_2_prior_emb2emb.py
@@ -8,7 +8,6 @@
 from ...pipelines import DiffusionPipeline
 from ...schedulers import UnCLIPScheduler
 from ...utils import (
-    is_accelerate_available,
     logging,
     randn_tensor,
     replace_example_docstring,
@@ -121,6 +120,8 @@ class KandinskyV22PriorEmb2EmbPipeline(DiffusionPipeline):
             A scheduler to be used in combination with `prior` to generate image embedding.
     """
 
+    _exclude_from_cpu_offload = ["prior"]
+
     def __init__(
         self,
         prior: PriorTransformer,
@@ -293,47 +294,6 @@ def get_zero_embed(self, batch_size=1, device=None):
         zero_image_emb = zero_image_emb.repeat(batch_size, 1)
         return zero_image_emb
 
-    # Copied from diffusers.pipelines.kandinsky.pipeline_kandinsky_prior.KandinskyPriorPipeline.enable_sequential_cpu_offload
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, the pipeline's
-        models have their state dicts saved to CPU and then are moved to a `torch.device('meta') and loaded to GPU only
-        when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        models = [
-            self.image_encoder,
-            self.text_encoder,
-        ]
-        for cpu_offloaded_model in models:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
-    @property
-    # Copied from diffusers.pipelines.kandinsky.pipeline_kandinsky_prior.KandinskyPriorPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if self.device != torch.device("meta") or not hasattr(self.text_encoder, "_hf_hook"):
-            return self.device
-        for module in self.text_encoder.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     # Copied from diffusers.pipelines.kandinsky.pipeline_kandinsky_prior.KandinskyPriorPipeline._encode_prompt
     def _encode_prompt(
         self,
diff --git a/src/diffusers/pipelines/latent_diffusion/pipeline_latent_diffusion.py b/src/diffusers/pipelines/latent_diffusion/pipeline_latent_diffusion.py
index 3e4f9425b0f6..ab7c28b96cc8 100644
--- a/src/diffusers/pipelines/latent_diffusion/pipeline_latent_diffusion.py
+++ b/src/diffusers/pipelines/latent_diffusion/pipeline_latent_diffusion.py
@@ -128,11 +128,11 @@ def __call__(
             uncond_input = self.tokenizer(
                 [""] * batch_size, padding="max_length", max_length=77, truncation=True, return_tensors="pt"
             )
-            negative_prompt_embeds = self.bert(uncond_input.input_ids.to(self.device))[0]
+            negative_prompt_embeds = self.bert(uncond_input.input_ids.to(self._execution_device))[0]
 
         # get prompt text embeddings
         text_input = self.tokenizer(prompt, padding="max_length", max_length=77, truncation=True, return_tensors="pt")
-        prompt_embeds = self.bert(text_input.input_ids.to(self.device))[0]
+        prompt_embeds = self.bert(text_input.input_ids.to(self._execution_device))[0]
 
         # get the initial random noise unless the user supplied it
         latents_shape = (batch_size, self.unet.config.in_channels, height // 8, width // 8)
@@ -143,11 +143,13 @@ def __call__(
             )
 
         if latents is None:
-            latents = randn_tensor(latents_shape, generator=generator, device=self.device, dtype=prompt_embeds.dtype)
+            latents = randn_tensor(
+                latents_shape, generator=generator, device=self._execution_device, dtype=prompt_embeds.dtype
+            )
         else:
             if latents.shape != latents_shape:
                 raise ValueError(f"Unexpected latents shape, got {latents.shape}, expected {latents_shape}")
-        latents = latents.to(self.device)
+        latents = latents.to(self._execution_device)
 
         self.scheduler.set_timesteps(num_inference_steps)
 
diff --git a/src/diffusers/pipelines/paint_by_example/pipeline_paint_by_example.py b/src/diffusers/pipelines/paint_by_example/pipeline_paint_by_example.py
index c8f3e8a9ee11..f844834b527d 100644
--- a/src/diffusers/pipelines/paint_by_example/pipeline_paint_by_example.py
+++ b/src/diffusers/pipelines/paint_by_example/pipeline_paint_by_example.py
@@ -21,8 +21,6 @@
 import torch
 from transformers import CLIPImageProcessor
 
-from diffusers.utils import is_accelerate_available
-
 from ...image_processor import VaeImageProcessor
 from ...models import AutoencoderKL, UNet2DConditionModel
 from ...schedulers import DDIMScheduler, LMSDiscreteScheduler, PNDMScheduler
@@ -189,44 +187,6 @@ def __init__(
         self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
         self.register_to_config(requires_safety_checker=requires_safety_checker)
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        for cpu_offloaded_model in [self.unet, self.vae, self.image_encoder]:
-            cpu_offload(cpu_offloaded_model, execution_device=device)
-
-        if self.safety_checker is not None:
-            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
-
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.run_safety_checker
     def run_safety_checker(self, image, device, dtype):
         if self.safety_checker is None:
diff --git a/src/diffusers/pipelines/pipeline_utils.py b/src/diffusers/pipelines/pipeline_utils.py
index c2b7399be080..ad52c6ac1c59 100644
--- a/src/diffusers/pipelines/pipeline_utils.py
+++ b/src/diffusers/pipelines/pipeline_utils.py
@@ -474,6 +474,7 @@ class DiffusionPipeline(ConfigMixin):
     """
     config_name = "model_index.json"
     _optional_components = []
+    _exclude_from_cpu_offload = []
 
     def register_modules(self, **kwargs):
         # import it here to avoid circular import
@@ -1086,6 +1087,60 @@ def load_module(name, value):
     def name_or_path(self) -> str:
         return getattr(self.config, "_name_or_path", None)
 
+    @property
+    def _execution_device(self):
+        r"""
+        Returns the device on which the pipeline's models will be executed. After calling
+        [`~DiffusionPipeline.enable_sequential_cpu_offload`] the execution device can only be inferred from
+        Accelerate's module hooks.
+        """
+        for name, model in self.components.items():
+            if not isinstance(model, torch.nn.Module) or name in self._exclude_from_cpu_offload:
+                continue
+
+            if not hasattr(model, "_hf_hook"):
+                return self.device
+            for module in model.modules():
+                if (
+                    hasattr(module, "_hf_hook")
+                    and hasattr(module._hf_hook, "execution_device")
+                    and module._hf_hook.execution_device is not None
+                ):
+                    return torch.device(module._hf_hook.execution_device)
+        return self.device
+
+    def enable_sequential_cpu_offload(self, gpu_id: int = 0, device: Union[torch.device, str] = "cuda"):
+        r"""
+        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
+        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
+        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
+        Note that offloading happens on a submodule basis. Memory savings are higher than with
+        `enable_model_cpu_offload`, but performance is lower.
+        """
+        if is_accelerate_available() and is_accelerate_version(">=", "0.14.0"):
+            from accelerate import cpu_offload
+        else:
+            raise ImportError("`enable_sequential_cpu_offload` requires `accelerate v0.14.0` or higher")
+
+        if device == "cuda":
+            device = torch.device(f"{device}:{gpu_id}")
+
+        if self.device.type != "cpu":
+            self.to("cpu", silence_dtype_warnings=True)
+            torch.cuda.empty_cache()  # otherwise we don't see the memory savings (but they probably exist)
+
+        for name, model in self.components.items():
+            if not isinstance(model, torch.nn.Module):
+                continue
+
+            if name in self._exclude_from_cpu_offload:
+                model.to(device)
+            else:
+                # make sure to offload buffers if not all high level weights
+                # are of type nn.Module
+                offload_buffers = len(model._parameters) > 0
+                cpu_offload(model, device, offload_buffers=offload_buffers)
+
     @classmethod
     def download(cls, pretrained_model_name, **kwargs) -> Union[str, os.PathLike]:
         r"""
diff --git a/src/diffusers/pipelines/repaint/pipeline_repaint.py b/src/diffusers/pipelines/repaint/pipeline_repaint.py
index d2aa1d4f1f77..6527a023a74f 100644
--- a/src/diffusers/pipelines/repaint/pipeline_repaint.py
+++ b/src/diffusers/pipelines/repaint/pipeline_repaint.py
@@ -132,9 +132,9 @@ def __call__(
         original_image = image
 
         original_image = _preprocess_image(original_image)
-        original_image = original_image.to(device=self.device, dtype=self.unet.dtype)
+        original_image = original_image.to(device=self._execution_device, dtype=self.unet.dtype)
         mask_image = _preprocess_mask(mask_image)
-        mask_image = mask_image.to(device=self.device, dtype=self.unet.dtype)
+        mask_image = mask_image.to(device=self._execution_device, dtype=self.unet.dtype)
 
         batch_size = original_image.shape[0]
 
@@ -146,10 +146,10 @@ def __call__(
             )
 
         image_shape = original_image.shape
-        image = randn_tensor(image_shape, generator=generator, device=self.device, dtype=self.unet.dtype)
+        image = randn_tensor(image_shape, generator=generator, device=self._execution_device, dtype=self.unet.dtype)
 
         # set step values
-        self.scheduler.set_timesteps(num_inference_steps, jump_length, jump_n_sample, self.device)
+        self.scheduler.set_timesteps(num_inference_steps, jump_length, jump_n_sample, self._execution_device)
         self.scheduler.eta = eta
 
         t_last = self.scheduler.timesteps[0] + 1
diff --git a/src/diffusers/pipelines/shap_e/pipeline_shap_e.py b/src/diffusers/pipelines/shap_e/pipeline_shap_e.py
index 5d96fc7bb9f4..fdcbe550860a 100644
--- a/src/diffusers/pipelines/shap_e/pipeline_shap_e.py
+++ b/src/diffusers/pipelines/shap_e/pipeline_shap_e.py
@@ -130,24 +130,6 @@ def prepare_latents(self, shape, dtype, device, generator, latents, scheduler):
         latents = latents * scheduler.init_noise_sigma
         return latents
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, the pipeline's
-        models have their state dicts saved to CPU and then are moved to a `torch.device('meta') and loaded to GPU only
-        when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        models = [self.text_encoder, self.prior]
-        for cpu_offloaded_model in models:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
         Offloads all models to CPU using accelerate, reducing memory usage with a low impact on performance. Compared
@@ -176,24 +158,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if self.device != torch.device("meta") or not hasattr(self.text_encoder, "_hf_hook"):
-            return self.device
-        for module in self.text_encoder.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     def _encode_prompt(
         self,
         prompt,
diff --git a/src/diffusers/pipelines/shap_e/pipeline_shap_e_img2img.py b/src/diffusers/pipelines/shap_e/pipeline_shap_e_img2img.py
index b99b808e5953..08c585c5ad73 100644
--- a/src/diffusers/pipelines/shap_e/pipeline_shap_e_img2img.py
+++ b/src/diffusers/pipelines/shap_e/pipeline_shap_e_img2img.py
@@ -25,7 +25,6 @@
 from ...schedulers import HeunDiscreteScheduler
 from ...utils import (
     BaseOutput,
-    is_accelerate_available,
     logging,
     randn_tensor,
     replace_example_docstring,
@@ -130,42 +129,6 @@ def prepare_latents(self, shape, dtype, device, generator, latents, scheduler):
         latents = latents * scheduler.init_noise_sigma
         return latents
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, the pipeline's
-        models have their state dicts saved to CPU and then are moved to a `torch.device('meta') and loaded to GPU only
-        when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        models = [self.image_encoder, self.prior]
-        for cpu_offloaded_model in models:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
-    @property
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if self.device != torch.device("meta") or not hasattr(self.image_encoder, "_hf_hook"):
-            return self.device
-        for module in self.image_encoder.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     def _encode_image(
         self,
         image,
diff --git a/src/diffusers/pipelines/stable_diffusion/pipeline_cycle_diffusion.py b/src/diffusers/pipelines/stable_diffusion/pipeline_cycle_diffusion.py
index b8360f512405..9a68c4d059c6 100644
--- a/src/diffusers/pipelines/stable_diffusion/pipeline_cycle_diffusion.py
+++ b/src/diffusers/pipelines/stable_diffusion/pipeline_cycle_diffusion.py
@@ -231,32 +231,6 @@ def __init__(
         self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
         self.register_to_config(requires_safety_checker=requires_safety_checker)
 
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_sequential_cpu_offload
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        Note that offloading happens on a submodule basis. Memory savings are higher than with
-        `enable_model_cpu_offload`, but performance is lower.
-        """
-        if is_accelerate_available() and is_accelerate_version(">=", "0.14.0"):
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("`enable_sequential_cpu_offload` requires `accelerate v0.14.0` or higher")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        if self.device.type != "cpu":
-            self.to("cpu", silence_dtype_warnings=True)
-            torch.cuda.empty_cache()  # otherwise we don't see the memory savings (but they probably exist)
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae]:
-            cpu_offload(cpu_offloaded_model, device)
-
-        if self.safety_checker is not None:
-            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_model_cpu_offload
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
@@ -286,25 +260,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._encode_prompt
     def _encode_prompt(
         self,
diff --git a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py
index 9ad4d404fdbe..54927049571c 100644
--- a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py
+++ b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py
@@ -228,31 +228,6 @@ def disable_vae_tiling(self):
         """
         self.vae.disable_tiling()
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        Note that offloading happens on a submodule basis. Memory savings are higher than with
-        `enable_model_cpu_offload`, but performance is lower.
-        """
-        if is_accelerate_available() and is_accelerate_version(">=", "0.14.0"):
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("`enable_sequential_cpu_offload` requires `accelerate v0.14.0` or higher")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        if self.device.type != "cpu":
-            self.to("cpu", silence_dtype_warnings=True)
-            torch.cuda.empty_cache()  # otherwise we don't see the memory savings (but they probably exist)
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae]:
-            cpu_offload(cpu_offloaded_model, device)
-
-        if self.safety_checker is not None:
-            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
-
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
         Offloads all models to CPU using accelerate, reducing memory usage with a low impact on performance. Compared
@@ -281,24 +256,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     def _encode_prompt(
         self,
         prompt,
diff --git a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_attend_and_excite.py b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_attend_and_excite.py
index f76268463707..15a5d7eb1362 100644
--- a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_attend_and_excite.py
+++ b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_attend_and_excite.py
@@ -27,7 +27,7 @@
 from ...models import AutoencoderKL, UNet2DConditionModel
 from ...models.attention_processor import Attention
 from ...schedulers import KarrasDiffusionSchedulers
-from ...utils import is_accelerate_available, is_accelerate_version, logging, randn_tensor, replace_example_docstring
+from ...utils import logging, randn_tensor, replace_example_docstring
 from ..pipeline_utils import DiffusionPipeline
 from . import StableDiffusionPipelineOutput
 from .safety_checker import StableDiffusionSafetyChecker
@@ -251,51 +251,6 @@ def disable_vae_slicing(self):
         """
         self.vae.disable_slicing()
 
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_sequential_cpu_offload
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        Note that offloading happens on a submodule basis. Memory savings are higher than with
-        `enable_model_cpu_offload`, but performance is lower.
-        """
-        if is_accelerate_available() and is_accelerate_version(">=", "0.14.0"):
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("`enable_sequential_cpu_offload` requires `accelerate v0.14.0` or higher")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        if self.device.type != "cpu":
-            self.to("cpu", silence_dtype_warnings=True)
-            torch.cuda.empty_cache()  # otherwise we don't see the memory savings (but they probably exist)
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae]:
-            cpu_offload(cpu_offloaded_model, device)
-
-        if self.safety_checker is not None:
-            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
-
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._encode_prompt
     def _encode_prompt(
         self,
diff --git a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_depth2img.py b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_depth2img.py
index 002014681040..cae0f3a347de 100644
--- a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_depth2img.py
+++ b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_depth2img.py
@@ -28,7 +28,7 @@
 from ...loaders import LoraLoaderMixin, TextualInversionLoaderMixin
 from ...models import AutoencoderKL, UNet2DConditionModel
 from ...schedulers import KarrasDiffusionSchedulers
-from ...utils import PIL_INTERPOLATION, deprecate, is_accelerate_available, logging, randn_tensor
+from ...utils import PIL_INTERPOLATION, deprecate, logging, randn_tensor
 from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput
 
 
@@ -137,42 +137,6 @@ def __init__(
         self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)
         self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae, self.depth_estimator]:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._encode_prompt
     def _encode_prompt(
         self,
diff --git a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_diffedit.py b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_diffedit.py
index 837811baae64..f9bb8d0af418 100644
--- a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_diffedit.py
+++ b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_diffedit.py
@@ -403,32 +403,6 @@ def disable_vae_tiling(self):
         """
         self.vae.disable_tiling()
 
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_sequential_cpu_offload
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        Note that offloading happens on a submodule basis. Memory savings are higher than with
-        `enable_model_cpu_offload`, but performance is lower.
-        """
-        if is_accelerate_available() and is_accelerate_version(">=", "0.14.0"):
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("`enable_sequential_cpu_offload` requires `accelerate v0.14.0` or higher")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        if self.device.type != "cpu":
-            self.to("cpu", silence_dtype_warnings=True)
-            torch.cuda.empty_cache()  # otherwise we don't see the memory savings (but they probably exist)
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae]:
-            cpu_offload(cpu_offloaded_model, device)
-
-        if self.safety_checker is not None:
-            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_model_cpu_offload
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
@@ -458,25 +432,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._encode_prompt
     def _encode_prompt(
         self,
diff --git a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_image_variation.py b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_image_variation.py
index 640fd7f2d94b..ebcb55a7cff9 100644
--- a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_image_variation.py
+++ b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_image_variation.py
@@ -25,7 +25,7 @@
 from ...image_processor import VaeImageProcessor
 from ...models import AutoencoderKL, UNet2DConditionModel
 from ...schedulers import KarrasDiffusionSchedulers
-from ...utils import deprecate, is_accelerate_available, logging, randn_tensor
+from ...utils import deprecate, logging, randn_tensor
 from ..pipeline_utils import DiffusionPipeline
 from . import StableDiffusionPipelineOutput
 from .safety_checker import StableDiffusionSafetyChecker
@@ -123,42 +123,6 @@ def __init__(
         self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
         self.register_to_config(requires_safety_checker=requires_safety_checker)
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        for cpu_offloaded_model in [self.unet, self.image_encoder, self.vae, self.safety_checker]:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     def _encode_image(self, image, device, num_images_per_prompt, do_classifier_free_guidance):
         dtype = next(self.image_encoder.parameters()).dtype
 
diff --git a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_img2img.py b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_img2img.py
index f8874ba2cfae..ab4420f28386 100644
--- a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_img2img.py
+++ b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_img2img.py
@@ -227,32 +227,6 @@ def __init__(
         self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
         self.register_to_config(requires_safety_checker=requires_safety_checker)
 
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_sequential_cpu_offload
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        Note that offloading happens on a submodule basis. Memory savings are higher than with
-        `enable_model_cpu_offload`, but performance is lower.
-        """
-        if is_accelerate_available() and is_accelerate_version(">=", "0.14.0"):
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("`enable_sequential_cpu_offload` requires `accelerate v0.14.0` or higher")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        if self.device.type != "cpu":
-            self.to("cpu", silence_dtype_warnings=True)
-            torch.cuda.empty_cache()  # otherwise we don't see the memory savings (but they probably exist)
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae]:
-            cpu_offload(cpu_offloaded_model, device)
-
-        if self.safety_checker is not None:
-            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_model_cpu_offload
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
@@ -282,25 +256,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._encode_prompt
     def _encode_prompt(
         self,
diff --git a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint.py b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint.py
index b012a79ba855..b11ebfb6cfc5 100644
--- a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint.py
+++ b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint.py
@@ -296,32 +296,6 @@ def __init__(
         self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
         self.register_to_config(requires_safety_checker=requires_safety_checker)
 
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_sequential_cpu_offload
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        Note that offloading happens on a submodule basis. Memory savings are higher than with
-        `enable_model_cpu_offload`, but performance is lower.
-        """
-        if is_accelerate_available() and is_accelerate_version(">=", "0.14.0"):
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("`enable_sequential_cpu_offload` requires `accelerate v0.14.0` or higher")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        if self.device.type != "cpu":
-            self.to("cpu", silence_dtype_warnings=True)
-            torch.cuda.empty_cache()  # otherwise we don't see the memory savings (but they probably exist)
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae]:
-            cpu_offload(cpu_offloaded_model, device)
-
-        if self.safety_checker is not None:
-            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_model_cpu_offload
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
@@ -351,25 +325,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._encode_prompt
     def _encode_prompt(
         self,
diff --git a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint_legacy.py b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint_legacy.py
index 483f27ae3966..049e3d18f3de 100644
--- a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint_legacy.py
+++ b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint_legacy.py
@@ -220,32 +220,6 @@ def __init__(
         self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
         self.register_to_config(requires_safety_checker=requires_safety_checker)
 
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_sequential_cpu_offload
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        Note that offloading happens on a submodule basis. Memory savings are higher than with
-        `enable_model_cpu_offload`, but performance is lower.
-        """
-        if is_accelerate_available() and is_accelerate_version(">=", "0.14.0"):
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("`enable_sequential_cpu_offload` requires `accelerate v0.14.0` or higher")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        if self.device.type != "cpu":
-            self.to("cpu", silence_dtype_warnings=True)
-            torch.cuda.empty_cache()  # otherwise we don't see the memory savings (but they probably exist)
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae]:
-            cpu_offload(cpu_offloaded_model, device)
-
-        if self.safety_checker is not None:
-            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_model_cpu_offload
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
@@ -275,25 +249,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._encode_prompt
     def _encode_prompt(
         self,
diff --git a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_instruct_pix2pix.py b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_instruct_pix2pix.py
index 367e401d57f8..341ff8daad42 100644
--- a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_instruct_pix2pix.py
+++ b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_instruct_pix2pix.py
@@ -428,32 +428,6 @@ def __call__(
 
         return StableDiffusionPipelineOutput(images=image, nsfw_content_detected=has_nsfw_concept)
 
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_sequential_cpu_offload
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        Note that offloading happens on a submodule basis. Memory savings are higher than with
-        `enable_model_cpu_offload`, but performance is lower.
-        """
-        if is_accelerate_available() and is_accelerate_version(">=", "0.14.0"):
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("`enable_sequential_cpu_offload` requires `accelerate v0.14.0` or higher")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        if self.device.type != "cpu":
-            self.to("cpu", silence_dtype_warnings=True)
-            torch.cuda.empty_cache()  # otherwise we don't see the memory savings (but they probably exist)
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae]:
-            cpu_offload(cpu_offloaded_model, device)
-
-        if self.safety_checker is not None:
-            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_model_cpu_offload
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
@@ -483,25 +457,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     def _encode_prompt(
         self,
         prompt,
diff --git a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_k_diffusion.py b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_k_diffusion.py
index 283ddb6f19b3..29a57470a341 100755
--- a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_k_diffusion.py
+++ b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_k_diffusion.py
@@ -127,32 +127,6 @@ def set_scheduler(self, scheduler_type: str):
         sampling = getattr(library, "sampling")
         self.sampler = getattr(sampling, scheduler_type)
 
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_sequential_cpu_offload
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        Note that offloading happens on a submodule basis. Memory savings are higher than with
-        `enable_model_cpu_offload`, but performance is lower.
-        """
-        if is_accelerate_available() and is_accelerate_version(">=", "0.14.0"):
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("`enable_sequential_cpu_offload` requires `accelerate v0.14.0` or higher")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        if self.device.type != "cpu":
-            self.to("cpu", silence_dtype_warnings=True)
-            torch.cuda.empty_cache()  # otherwise we don't see the memory savings (but they probably exist)
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae]:
-            cpu_offload(cpu_offloaded_model, device)
-
-        if self.safety_checker is not None:
-            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_model_cpu_offload
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
@@ -182,25 +156,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._encode_prompt
     def _encode_prompt(
         self,
diff --git a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_latent_upscale.py b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_latent_upscale.py
index d67a7f894886..9d5022c854a8 100644
--- a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_latent_upscale.py
+++ b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_latent_upscale.py
@@ -24,7 +24,7 @@
 from ...image_processor import VaeImageProcessor
 from ...models import AutoencoderKL, UNet2DConditionModel
 from ...schedulers import EulerDiscreteScheduler
-from ...utils import is_accelerate_available, logging, randn_tensor
+from ...utils import logging, randn_tensor
 from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput
 
 
@@ -101,42 +101,6 @@ def __init__(
         self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)
         self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor, resample="bicubic")
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae]:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     def _encode_prompt(self, prompt, device, do_classifier_free_guidance, negative_prompt):
         r"""
         Encodes the prompt into text encoder hidden states.
diff --git a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_ldm3d.py b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_ldm3d.py
index 85f628ca8229..95dd207f9d12 100644
--- a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_ldm3d.py
+++ b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_ldm3d.py
@@ -193,31 +193,6 @@ def disable_vae_tiling(self):
         """
         self.vae.disable_tiling()
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        Note that offloading happens on a submodule basis. Memory savings are higher than with
-        `enable_model_cpu_offload`, but performance is lower.
-        """
-        if is_accelerate_available() and is_accelerate_version(">=", "0.14.0"):
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("`enable_sequential_cpu_offload` requires `accelerate v0.14.0` or higher")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        if self.device.type != "cpu":
-            self.to("cpu", silence_dtype_warnings=True)
-            torch.cuda.empty_cache()  # otherwise we don't see the memory savings (but they probably exist)
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae]:
-            cpu_offload(cpu_offloaded_model, device)
-
-        if self.safety_checker is not None:
-            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_model_cpu_offload
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
@@ -247,25 +222,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._encode_prompt
     def _encode_prompt(
         self,
diff --git a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_model_editing.py b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_model_editing.py
index 1d30b9ee0347..2ecb3f9dbaf7 100644
--- a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_model_editing.py
+++ b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_model_editing.py
@@ -24,7 +24,7 @@
 from ...models import AutoencoderKL, UNet2DConditionModel
 from ...schedulers import PNDMScheduler
 from ...schedulers.scheduling_utils import SchedulerMixin
-from ...utils import is_accelerate_available, is_accelerate_version, logging, randn_tensor
+from ...utils import logging, randn_tensor
 from ..pipeline_utils import DiffusionPipeline
 from . import StableDiffusionPipelineOutput
 from .safety_checker import StableDiffusionSafetyChecker
@@ -182,51 +182,6 @@ def disable_vae_slicing(self):
         """
         self.vae.disable_slicing()
 
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_sequential_cpu_offload
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        Note that offloading happens on a submodule basis. Memory savings are higher than with
-        `enable_model_cpu_offload`, but performance is lower.
-        """
-        if is_accelerate_available() and is_accelerate_version(">=", "0.14.0"):
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("`enable_sequential_cpu_offload` requires `accelerate v0.14.0` or higher")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        if self.device.type != "cpu":
-            self.to("cpu", silence_dtype_warnings=True)
-            torch.cuda.empty_cache()  # otherwise we don't see the memory savings (but they probably exist)
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae]:
-            cpu_offload(cpu_offloaded_model, device)
-
-        if self.safety_checker is not None:
-            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
-
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._encode_prompt
     def _encode_prompt(
         self,
diff --git a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_panorama.py b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_panorama.py
index 7ebeaa17c0bb..37e705d1bc5a 100644
--- a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_panorama.py
+++ b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_panorama.py
@@ -23,7 +23,7 @@
 from ...loaders import LoraLoaderMixin, TextualInversionLoaderMixin
 from ...models import AutoencoderKL, UNet2DConditionModel
 from ...schedulers import DDIMScheduler
-from ...utils import is_accelerate_available, is_accelerate_version, logging, randn_tensor, replace_example_docstring
+from ...utils import logging, randn_tensor, replace_example_docstring
 from ..pipeline_utils import DiffusionPipeline
 from . import StableDiffusionPipelineOutput
 from .safety_checker import StableDiffusionSafetyChecker
@@ -144,51 +144,6 @@ def disable_vae_slicing(self):
         """
         self.vae.disable_slicing()
 
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_sequential_cpu_offload
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        Note that offloading happens on a submodule basis. Memory savings are higher than with
-        `enable_model_cpu_offload`, but performance is lower.
-        """
-        if is_accelerate_available() and is_accelerate_version(">=", "0.14.0"):
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("`enable_sequential_cpu_offload` requires `accelerate v0.14.0` or higher")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        if self.device.type != "cpu":
-            self.to("cpu", silence_dtype_warnings=True)
-            torch.cuda.empty_cache()  # otherwise we don't see the memory savings (but they probably exist)
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae]:
-            cpu_offload(cpu_offloaded_model, device)
-
-        if self.safety_checker is not None:
-            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
-
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._encode_prompt
     def _encode_prompt(
         self,
diff --git a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_paradigms.py b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_paradigms.py
index 2239e3853a8e..073f02e8ee98 100644
--- a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_paradigms.py
+++ b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_paradigms.py
@@ -182,32 +182,6 @@ def disable_vae_tiling(self):
         """
         self.vae.disable_tiling()
 
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_sequential_cpu_offload
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        Note that offloading happens on a submodule basis. Memory savings are higher than with
-        `enable_model_cpu_offload`, but performance is lower.
-        """
-        if is_accelerate_available() and is_accelerate_version(">=", "0.14.0"):
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("`enable_sequential_cpu_offload` requires `accelerate v0.14.0` or higher")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        if self.device.type != "cpu":
-            self.to("cpu", silence_dtype_warnings=True)
-            torch.cuda.empty_cache()  # otherwise we don't see the memory savings (but they probably exist)
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae]:
-            cpu_offload(cpu_offloaded_model, device)
-
-        if self.safety_checker is not None:
-            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_model_cpu_offload
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
@@ -237,25 +211,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._encode_prompt
     def _encode_prompt(
         self,
diff --git a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_pix2pix_zero.py b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_pix2pix_zero.py
index 75ac4f777756..902bcdd0496b 100644
--- a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_pix2pix_zero.py
+++ b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_pix2pix_zero.py
@@ -368,32 +368,6 @@ def __init__(
         self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
         self.register_to_config(requires_safety_checker=requires_safety_checker)
 
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_sequential_cpu_offload
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        Note that offloading happens on a submodule basis. Memory savings are higher than with
-        `enable_model_cpu_offload`, but performance is lower.
-        """
-        if is_accelerate_available() and is_accelerate_version(">=", "0.14.0"):
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("`enable_sequential_cpu_offload` requires `accelerate v0.14.0` or higher")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        if self.device.type != "cpu":
-            self.to("cpu", silence_dtype_warnings=True)
-            torch.cuda.empty_cache()  # otherwise we don't see the memory savings (but they probably exist)
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae]:
-            cpu_offload(cpu_offloaded_model, device)
-
-        if self.safety_checker is not None:
-            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
-
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
         Offloads all models to CPU using accelerate, reducing memory usage with a low impact on performance. Compared
@@ -418,25 +392,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._encode_prompt
     def _encode_prompt(
         self,
diff --git a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_sag.py b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_sag.py
index ba1c0d2b9d49..9c583de9ca9c 100644
--- a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_sag.py
+++ b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_sag.py
@@ -24,7 +24,7 @@
 from ...loaders import LoraLoaderMixin, TextualInversionLoaderMixin
 from ...models import AutoencoderKL, UNet2DConditionModel
 from ...schedulers import KarrasDiffusionSchedulers
-from ...utils import is_accelerate_available, is_accelerate_version, logging, randn_tensor, replace_example_docstring
+from ...utils import logging, randn_tensor, replace_example_docstring
 from ..pipeline_utils import DiffusionPipeline
 from . import StableDiffusionPipelineOutput
 from .safety_checker import StableDiffusionSafetyChecker
@@ -163,51 +163,6 @@ def disable_vae_slicing(self):
         """
         self.vae.disable_slicing()
 
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_sequential_cpu_offload
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        Note that offloading happens on a submodule basis. Memory savings are higher than with
-        `enable_model_cpu_offload`, but performance is lower.
-        """
-        if is_accelerate_available() and is_accelerate_version(">=", "0.14.0"):
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("`enable_sequential_cpu_offload` requires `accelerate v0.14.0` or higher")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        if self.device.type != "cpu":
-            self.to("cpu", silence_dtype_warnings=True)
-            torch.cuda.empty_cache()  # otherwise we don't see the memory savings (but they probably exist)
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae]:
-            cpu_offload(cpu_offloaded_model, device)
-
-        if self.safety_checker is not None:
-            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
-
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._encode_prompt
     def _encode_prompt(
         self,
diff --git a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_upscale.py b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_upscale.py
index 59aa87f46c19..61b1419d5ced 100644
--- a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_upscale.py
+++ b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_upscale.py
@@ -140,23 +140,6 @@ def __init__(
         self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor, resample="bicubic")
         self.register_to_config(max_noise_level=max_noise_level)
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae]:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
         Offloads all models to CPU using accelerate, reducing memory usage with a low impact on performance. Compared
@@ -183,25 +166,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     # Copied from diffusers.pipelines.deepfloyd_if.pipeline_if.IFPipeline.run_safety_checker
     def run_safety_checker(self, image, device, dtype):
         if self.safety_checker is not None:
diff --git a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_unclip.py b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_unclip.py
index e36ebfbb70f1..7c89bfedbd59 100644
--- a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_unclip.py
+++ b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_unclip.py
@@ -85,6 +85,8 @@ class StableUnCLIPPipeline(DiffusionPipeline, TextualInversionLoaderMixin, LoraL
             Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
     """
 
+    _exclude_from_cpu_offload = ["prior", "image_normalizer"]
+
     # prior components
     prior_tokenizer: CLIPTokenizer
     prior_text_encoder: CLIPTextModelWithProjection
@@ -158,30 +160,6 @@ def disable_vae_slicing(self):
         """
         self.vae.disable_slicing()
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, the pipeline's
-        models have their state dicts saved to CPU and then are moved to a `torch.device('meta') and loaded to GPU only
-        when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        # TODO: self.prior.post_process_latents and self.image_noiser.{scale,unscale} are not covered by the offload hooks, so they fails if added to the list
-        models = [
-            self.prior_text_encoder,
-            self.text_encoder,
-            self.unet,
-            self.vae,
-        ]
-        for cpu_offloaded_model in models:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
         Offloads all models to CPU using accelerate, reducing memory usage with a low impact on performance. Compared
@@ -207,25 +185,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     # Copied from diffusers.pipelines.unclip.pipeline_unclip.UnCLIPPipeline._encode_prompt with _encode_prompt->_encode_prior_prompt, tokenizer->prior_tokenizer, text_encoder->prior_text_encoder
     def _encode_prior_prompt(
         self,
diff --git a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_unclip_img2img.py b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_unclip_img2img.py
index 0187c86b4239..003c82ff4f8a 100644
--- a/src/diffusers/pipelines/stable_diffusion/pipeline_stable_unclip_img2img.py
+++ b/src/diffusers/pipelines/stable_diffusion/pipeline_stable_unclip_img2img.py
@@ -93,6 +93,8 @@ class StableUnCLIPImg2ImgPipeline(DiffusionPipeline, TextualInversionLoaderMixin
             Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
     """
 
+    _exclude_from_cpu_offload = ["image_normalizer"]
+
     # image encoding components
     feature_extractor: CLIPImageProcessor
     image_encoder: CLIPVisionModelWithProjection
@@ -160,30 +162,6 @@ def disable_vae_slicing(self):
         """
         self.vae.disable_slicing()
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, the pipeline's
-        models have their state dicts saved to CPU and then are moved to a `torch.device('meta') and loaded to GPU only
-        when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        # TODO: self.image_normalizer.{scale,unscale} are not covered by the offload hooks, so they fails if added to the list
-        models = [
-            self.image_encoder,
-            self.text_encoder,
-            self.unet,
-            self.vae,
-        ]
-        for cpu_offloaded_model in models:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
         Offloads all models to CPU using accelerate, reducing memory usage with a low impact on performance. Compared
@@ -209,25 +187,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._encode_prompt
     def _encode_prompt(
         self,
diff --git a/src/diffusers/pipelines/stable_diffusion_safe/pipeline_stable_diffusion_safe.py b/src/diffusers/pipelines/stable_diffusion_safe/pipeline_stable_diffusion_safe.py
index d770ee290517..f172575bc6c7 100644
--- a/src/diffusers/pipelines/stable_diffusion_safe/pipeline_stable_diffusion_safe.py
+++ b/src/diffusers/pipelines/stable_diffusion_safe/pipeline_stable_diffusion_safe.py
@@ -10,7 +10,7 @@
 from ...configuration_utils import FrozenDict
 from ...models import AutoencoderKL, UNet2DConditionModel
 from ...schedulers import KarrasDiffusionSchedulers
-from ...utils import deprecate, is_accelerate_available, logging, randn_tensor
+from ...utils import deprecate, logging, randn_tensor
 from ..pipeline_utils import DiffusionPipeline
 from . import StableDiffusionSafePipelineOutput
 from .safety_checker import SafeStableDiffusionSafetyChecker
@@ -167,42 +167,6 @@ def safety_concept(self, concept):
         """
         self._safety_text_concept = concept
 
-    def enable_sequential_cpu_offload(self):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device("cuda")
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae, self.safety_checker]:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     def _encode_prompt(
         self,
         prompt,
diff --git a/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py b/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py
index bd33068862c1..9863c663910f 100644
--- a/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py
+++ b/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py
@@ -179,28 +179,6 @@ def disable_vae_tiling(self):
         """
         self.vae.disable_tiling()
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        Note that offloading happens on a submodule basis. Memory savings are higher than with
-        `enable_model_cpu_offload`, but performance is lower.
-        """
-        if is_accelerate_available() and is_accelerate_version(">=", "0.14.0"):
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("`enable_sequential_cpu_offload` requires `accelerate v0.14.0` or higher")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        if self.device.type != "cpu":
-            self.to("cpu", silence_dtype_warnings=True)
-            torch.cuda.empty_cache()  # otherwise we don't see the memory savings (but they probably exist)
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder, self.text_encoder_2, self.vae]:
-            cpu_offload(cpu_offloaded_model, device)
-
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
         Offloads all models to CPU using accelerate, reducing memory usage with a low impact on performance. Compared
@@ -231,25 +209,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     def encode_prompt(
         self,
         prompt,
diff --git a/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_img2img.py b/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_img2img.py
index 03f6ec9a5006..a106ee7ac5f6 100644
--- a/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_img2img.py
+++ b/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_img2img.py
@@ -186,31 +186,6 @@ def disable_vae_tiling(self):
         """
         self.vae.disable_tiling()
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        Note that offloading happens on a submodule basis. Memory savings are higher than with
-        `enable_model_cpu_offload`, but performance is lower.
-        """
-        if is_accelerate_available() and is_accelerate_version(">=", "0.14.0"):
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("`enable_sequential_cpu_offload` requires `accelerate v0.14.0` or higher")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        if self.device.type != "cpu":
-            self.to("cpu", silence_dtype_warnings=True)
-            torch.cuda.empty_cache()  # otherwise we don't see the memory savings (but they probably exist)
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder_2, self.vae]:
-            cpu_offload(cpu_offloaded_model, device)
-
-        if self.text_encoder is not None:
-            cpu_offload(self.text_encoder, device)
-
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
         Offloads all models to CPU using accelerate, reducing memory usage with a low impact on performance. Compared
@@ -241,25 +216,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.StableDiffusionXLPipeline.encode_prompt
     def encode_prompt(
         self,
diff --git a/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_inpaint.py b/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_inpaint.py
index 571de38ab111..55f20660afc7 100644
--- a/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_inpaint.py
+++ b/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_inpaint.py
@@ -291,32 +291,6 @@ def disable_vae_tiling(self):
         """
         self.vae.disable_tiling()
 
-    # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl_img2img.StableDiffusionXLImg2ImgPipeline.enable_sequential_cpu_offload
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        Note that offloading happens on a submodule basis. Memory savings are higher than with
-        `enable_model_cpu_offload`, but performance is lower.
-        """
-        if is_accelerate_available() and is_accelerate_version(">=", "0.14.0"):
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("`enable_sequential_cpu_offload` requires `accelerate v0.14.0` or higher")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        if self.device.type != "cpu":
-            self.to("cpu", silence_dtype_warnings=True)
-            torch.cuda.empty_cache()  # otherwise we don't see the memory savings (but they probably exist)
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder_2, self.vae]:
-            cpu_offload(cpu_offloaded_model, device)
-
-        if self.text_encoder is not None:
-            cpu_offload(self.text_encoder, device)
-
     # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl_img2img.StableDiffusionXLImg2ImgPipeline.enable_model_cpu_offload
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
@@ -535,25 +509,6 @@ def encode_prompt(
 
         return prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs
     def prepare_extra_step_kwargs(self, generator, eta):
         # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
diff --git a/src/diffusers/pipelines/t2i_adapter/pipeline_stable_diffusion_adapter.py b/src/diffusers/pipelines/t2i_adapter/pipeline_stable_diffusion_adapter.py
index c84c99e6d19a..97c6fb99157f 100644
--- a/src/diffusers/pipelines/t2i_adapter/pipeline_stable_diffusion_adapter.py
+++ b/src/diffusers/pipelines/t2i_adapter/pipeline_stable_diffusion_adapter.py
@@ -218,31 +218,6 @@ def disable_vae_slicing(self):
         """
         self.vae.disable_slicing()
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        Note that offloading happens on a submodule basis. Memory savings are higher than with
-        `enable_model_cpu_offload`, but performance is lower.
-        """
-        if is_accelerate_available() and is_accelerate_version(">=", "0.14.0"):
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("`enable_sequential_cpu_offload` requires `accelerate v0.14.0` or higher")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        if self.device.type != "cpu":
-            self.to("cpu", silence_dtype_warnings=True)
-            torch.cuda.empty_cache()  # otherwise we don't see the memory savings (but they probably exist)
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae, self.adapter]:
-            cpu_offload(cpu_offloaded_model, device)
-
-        if self.safety_checker is not None:
-            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
-
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
         Offloads all models to CPU using accelerate, reducing memory usage with a low impact on performance. Compared
@@ -271,25 +246,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._encode_prompt
     def _encode_prompt(
         self,
diff --git a/src/diffusers/pipelines/text_to_video_synthesis/pipeline_text_to_video_synth.py b/src/diffusers/pipelines/text_to_video_synthesis/pipeline_text_to_video_synth.py
index ecc330b5f504..dad7d5639892 100644
--- a/src/diffusers/pipelines/text_to_video_synthesis/pipeline_text_to_video_synth.py
+++ b/src/diffusers/pipelines/text_to_video_synthesis/pipeline_text_to_video_synth.py
@@ -149,27 +149,6 @@ def disable_vae_tiling(self):
         """
         self.vae.disable_tiling()
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae have their state dicts saved to CPU and then are moved to a `torch.device('meta') and loaded
-        to GPU only when their specific submodule has its `forward` method called. Note that offloading happens on a
-        submodule basis. Memory savings are higher than with `enable_model_cpu_offload`, but performance is lower.
-        """
-        if is_accelerate_available() and is_accelerate_version(">=", "0.14.0"):
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("`enable_sequential_cpu_offload` requires `accelerate v0.14.0` or higher")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        if self.device.type != "cpu":
-            self.to("cpu", silence_dtype_warnings=True)
-            torch.cuda.empty_cache()  # otherwise we don't see the memory savings (but they probably exist)
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae]:
-            cpu_offload(cpu_offloaded_model, device)
-
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
         Offloads all models to CPU using accelerate, reducing memory usage with a low impact on performance. Compared
@@ -195,25 +174,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._encode_prompt
     def _encode_prompt(
         self,
diff --git a/src/diffusers/pipelines/text_to_video_synthesis/pipeline_text_to_video_synth_img2img.py b/src/diffusers/pipelines/text_to_video_synthesis/pipeline_text_to_video_synth_img2img.py
index 7a4b73cd3c35..72a5b762d504 100644
--- a/src/diffusers/pipelines/text_to_video_synthesis/pipeline_text_to_video_synth_img2img.py
+++ b/src/diffusers/pipelines/text_to_video_synthesis/pipeline_text_to_video_synth_img2img.py
@@ -211,27 +211,6 @@ def disable_vae_tiling(self):
         """
         self.vae.disable_tiling()
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae have their state dicts saved to CPU and then are moved to a `torch.device('meta') and loaded
-        to GPU only when their specific submodule has its `forward` method called. Note that offloading happens on a
-        submodule basis. Memory savings are higher than with `enable_model_cpu_offload`, but performance is lower.
-        """
-        if is_accelerate_available() and is_accelerate_version(">=", "0.14.0"):
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("`enable_sequential_cpu_offload` requires `accelerate v0.14.0` or higher")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        if self.device.type != "cpu":
-            self.to("cpu", silence_dtype_warnings=True)
-            torch.cuda.empty_cache()  # otherwise we don't see the memory savings (but they probably exist)
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae]:
-            cpu_offload(cpu_offloaded_model, device)
-
     def enable_model_cpu_offload(self, gpu_id=0):
         r"""
         Offloads all models to CPU using accelerate, reducing memory usage with a low impact on performance. Compared
@@ -257,25 +236,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._encode_prompt
     def _encode_prompt(
         self,
diff --git a/src/diffusers/pipelines/unclip/pipeline_unclip.py b/src/diffusers/pipelines/unclip/pipeline_unclip.py
index abbb48ce8f46..f3b67bebfc8d 100644
--- a/src/diffusers/pipelines/unclip/pipeline_unclip.py
+++ b/src/diffusers/pipelines/unclip/pipeline_unclip.py
@@ -24,7 +24,7 @@
 from ...pipelines import DiffusionPipeline
 from ...pipelines.pipeline_utils import ImagePipelineOutput
 from ...schedulers import UnCLIPScheduler
-from ...utils import is_accelerate_available, logging, randn_tensor
+from ...utils import logging, randn_tensor
 from .text_proj import UnCLIPTextProjModel
 
 
@@ -63,6 +63,8 @@ class UnCLIPPipeline(DiffusionPipeline):
 
     """
 
+    _exclude_from_cpu_offload = ["prior"]
+
     prior: PriorTransformer
     decoder: UNet2DConditionModel
     text_proj: UnCLIPTextProjModel
@@ -205,49 +207,6 @@ def _encode_prompt(
 
         return prompt_embeds, text_encoder_hidden_states, text_mask
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, the pipeline's
-        models have their state dicts saved to CPU and then are moved to a `torch.device('meta') and loaded to GPU only
-        when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        # TODO: self.prior.post_process_latents is not covered by the offload hooks, so it fails if added to the list
-        models = [
-            self.decoder,
-            self.text_proj,
-            self.text_encoder,
-            self.super_res_first,
-            self.super_res_last,
-        ]
-        for cpu_offloaded_model in models:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
-    @property
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if self.device != torch.device("meta") or not hasattr(self.decoder, "_hf_hook"):
-            return self.device
-        for module in self.decoder.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     @torch.no_grad()
     def __call__(
         self,
diff --git a/src/diffusers/pipelines/unclip/pipeline_unclip_image_variation.py b/src/diffusers/pipelines/unclip/pipeline_unclip_image_variation.py
index 30d74cd36bb0..580417f517be 100644
--- a/src/diffusers/pipelines/unclip/pipeline_unclip_image_variation.py
+++ b/src/diffusers/pipelines/unclip/pipeline_unclip_image_variation.py
@@ -28,7 +28,7 @@
 from ...models import UNet2DConditionModel, UNet2DModel
 from ...pipelines import DiffusionPipeline, ImagePipelineOutput
 from ...schedulers import UnCLIPScheduler
-from ...utils import is_accelerate_available, logging, randn_tensor
+from ...utils import logging, randn_tensor
 from .text_proj import UnCLIPTextProjModel
 
 
@@ -198,49 +198,6 @@ def _encode_image(self, image, device, num_images_per_prompt, image_embeddings:
 
         return image_embeddings
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, the pipeline's
-        models have their state dicts saved to CPU and then are moved to a `torch.device('meta') and loaded to GPU only
-        when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        models = [
-            self.decoder,
-            self.text_proj,
-            self.text_encoder,
-            self.super_res_first,
-            self.super_res_last,
-        ]
-        for cpu_offloaded_model in models:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
-    @property
-    # Copied from diffusers.pipelines.unclip.pipeline_unclip.UnCLIPPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if self.device != torch.device("meta") or not hasattr(self.decoder, "_hf_hook"):
-            return self.device
-        for module in self.decoder.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     @torch.no_grad()
     def __call__(
         self,
diff --git a/src/diffusers/pipelines/unidiffuser/pipeline_unidiffuser.py b/src/diffusers/pipelines/unidiffuser/pipeline_unidiffuser.py
index e25da12414e7..3632d74d1c12 100644
--- a/src/diffusers/pipelines/unidiffuser/pipeline_unidiffuser.py
+++ b/src/diffusers/pipelines/unidiffuser/pipeline_unidiffuser.py
@@ -172,33 +172,6 @@ def __init__(
         # TODO: handle safety checking?
         self.safety_checker = None
 
-    # Modified from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_sequential_cpu_offload
-    # Add self.image_encoder, self.text_decoder to cpu_offloaded_models list
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta')` and loaded to GPU only when their specific submodule has its `forward` method called.
-        Note that offloading happens on a submodule basis. Memory savings are higher than with
-        `enable_model_cpu_offload`, but performance is lower.
-        """
-        if is_accelerate_available() and is_accelerate_version(">=", "0.14.0"):
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("`enable_sequential_cpu_offload` requires `accelerate v0.14.0` or higher")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        if self.device.type != "cpu":
-            self.to("cpu", silence_dtype_warnings=True)
-            torch.cuda.empty_cache()  # otherwise we don't see the memory savings (but they probably exist)
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae, self.image_encoder, self.text_decoder]:
-            cpu_offload(cpu_offloaded_model, device)
-
-        if self.safety_checker is not None:
-            cpu_offload(self.safety_checker, execution_device=device, offload_buffers=True)
-
     # Modified from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_model_cpu_offload
     # Add self.image_encoder, self.text_decoder to cpu_offloaded_models list
     def enable_model_cpu_offload(self, gpu_id=0):
@@ -229,25 +202,6 @@ def enable_model_cpu_offload(self, gpu_id=0):
         # We'll offload the last model manually.
         self.final_offload_hook = hook
 
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs
     def prepare_extra_step_kwargs(self, generator, eta):
         # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
diff --git a/src/diffusers/pipelines/versatile_diffusion/pipeline_versatile_diffusion_dual_guided.py b/src/diffusers/pipelines/versatile_diffusion/pipeline_versatile_diffusion_dual_guided.py
index 1d2e61d86b90..5986d66a61e7 100644
--- a/src/diffusers/pipelines/versatile_diffusion/pipeline_versatile_diffusion_dual_guided.py
+++ b/src/diffusers/pipelines/versatile_diffusion/pipeline_versatile_diffusion_dual_guided.py
@@ -30,7 +30,7 @@
 from ...image_processor import VaeImageProcessor
 from ...models import AutoencoderKL, DualTransformer2DModel, Transformer2DModel, UNet2DConditionModel
 from ...schedulers import KarrasDiffusionSchedulers
-from ...utils import is_accelerate_available, logging, randn_tensor
+from ...utils import logging, randn_tensor
 from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput
 from .modeling_text_unet import UNetFlatConditionModel
 
@@ -148,42 +148,6 @@ def _revert_dual_attention(self):
 
         self.image_unet.register_to_config(dual_cross_attention=False)
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        for cpu_offloaded_model in [self.image_unet, self.text_unet, self.text_encoder, self.vae]:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device with unet->image_unet
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.image_unet, "_hf_hook"):
-            return self.device
-        for module in self.image_unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     def _encode_text_prompt(self, prompt, device, num_images_per_prompt, do_classifier_free_guidance):
         r"""
         Encodes the prompt into text encoder hidden states.
diff --git a/src/diffusers/pipelines/versatile_diffusion/pipeline_versatile_diffusion_image_variation.py b/src/diffusers/pipelines/versatile_diffusion/pipeline_versatile_diffusion_image_variation.py
index 4450846300fc..154548df7542 100644
--- a/src/diffusers/pipelines/versatile_diffusion/pipeline_versatile_diffusion_image_variation.py
+++ b/src/diffusers/pipelines/versatile_diffusion/pipeline_versatile_diffusion_image_variation.py
@@ -25,7 +25,7 @@
 from ...image_processor import VaeImageProcessor
 from ...models import AutoencoderKL, UNet2DConditionModel
 from ...schedulers import KarrasDiffusionSchedulers
-from ...utils import is_accelerate_available, logging, randn_tensor
+from ...utils import logging, randn_tensor
 from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput
 
 
@@ -75,42 +75,6 @@ def __init__(
         self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)
         self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        for cpu_offloaded_model in [self.image_unet, self.text_unet, self.text_encoder, self.vae]:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device with unet->image_unet
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.image_unet, "_hf_hook"):
-            return self.device
-        for module in self.image_unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     def _encode_prompt(self, prompt, device, num_images_per_prompt, do_classifier_free_guidance, negative_prompt):
         r"""
         Encodes the prompt into text encoder hidden states.
diff --git a/src/diffusers/pipelines/versatile_diffusion/pipeline_versatile_diffusion_text_to_image.py b/src/diffusers/pipelines/versatile_diffusion/pipeline_versatile_diffusion_text_to_image.py
index 1fdb21f2b745..1ba5d8451f2e 100644
--- a/src/diffusers/pipelines/versatile_diffusion/pipeline_versatile_diffusion_text_to_image.py
+++ b/src/diffusers/pipelines/versatile_diffusion/pipeline_versatile_diffusion_text_to_image.py
@@ -23,7 +23,7 @@
 from ...image_processor import VaeImageProcessor
 from ...models import AutoencoderKL, Transformer2DModel, UNet2DConditionModel
 from ...schedulers import KarrasDiffusionSchedulers
-from ...utils import is_accelerate_available, logging, randn_tensor
+from ...utils import logging, randn_tensor
 from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput
 from .modeling_text_unet import UNetFlatConditionModel
 
@@ -99,42 +99,6 @@ def _swap_unet_attention_blocks(self):
     def remove_unused_weights(self):
         self.register_modules(text_unet=None)
 
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        r"""
-        Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
-        text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
-        `torch.device('meta') and loaded to GPU only when their specific submodule has its `forward` method called.
-        """
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        for cpu_offloaded_model in [self.image_unet, self.text_unet, self.text_encoder, self.vae]:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
-    @property
-    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._execution_device with unet->image_unet
-    def _execution_device(self):
-        r"""
-        Returns the device on which the pipeline's models will be executed. After calling
-        `pipeline.enable_sequential_cpu_offload()` the execution device can only be inferred from Accelerate's module
-        hooks.
-        """
-        if not hasattr(self.image_unet, "_hf_hook"):
-            return self.device
-        for module in self.image_unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
     def _encode_prompt(self, prompt, device, num_images_per_prompt, do_classifier_free_guidance, negative_prompt):
         r"""
         Encodes the prompt into text encoder hidden states.
diff --git a/tests/pipelines/dance_diffusion/test_dance_diffusion.py b/tests/pipelines/dance_diffusion/test_dance_diffusion.py
index 0ba86daa61fc..fadb3ead7a78 100644
--- a/tests/pipelines/dance_diffusion/test_dance_diffusion.py
+++ b/tests/pipelines/dance_diffusion/test_dance_diffusion.py
@@ -42,7 +42,6 @@ class DanceDiffusionPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
     }
     batch_params = UNCONDITIONAL_AUDIO_GENERATION_BATCH_PARAMS
     test_attention_slicing = False
-    test_cpu_offload = False
 
     def get_dummy_components(self):
         torch.manual_seed(0)
diff --git a/tests/pipelines/ddim/test_ddim.py b/tests/pipelines/ddim/test_ddim.py
index 0861d7daab29..de513fe234fd 100644
--- a/tests/pipelines/ddim/test_ddim.py
+++ b/tests/pipelines/ddim/test_ddim.py
@@ -38,7 +38,6 @@ class DDIMPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
         "callback_steps",
     }
     batch_params = UNCONDITIONAL_IMAGE_GENERATION_BATCH_PARAMS
-    test_cpu_offload = False
 
     def get_dummy_components(self):
         torch.manual_seed(0)
diff --git a/tests/pipelines/dit/test_dit.py b/tests/pipelines/dit/test_dit.py
index 4937915696b4..9a493ab4eeaa 100644
--- a/tests/pipelines/dit/test_dit.py
+++ b/tests/pipelines/dit/test_dit.py
@@ -43,7 +43,6 @@ class DiTPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
         "callback_steps",
     }
     batch_params = CLASS_CONDITIONED_IMAGE_GENERATION_BATCH_PARAMS
-    test_cpu_offload = False
 
     def get_dummy_components(self):
         torch.manual_seed(0)
diff --git a/tests/pipelines/latent_diffusion/test_latent_diffusion.py b/tests/pipelines/latent_diffusion/test_latent_diffusion.py
index 88dc8ef9b17b..47aec73edc1e 100644
--- a/tests/pipelines/latent_diffusion/test_latent_diffusion.py
+++ b/tests/pipelines/latent_diffusion/test_latent_diffusion.py
@@ -51,7 +51,6 @@ class LDMTextToImagePipelineFastTests(PipelineTesterMixin, unittest.TestCase):
         "callback_steps",
     }
     batch_params = TEXT_TO_IMAGE_BATCH_PARAMS
-    test_cpu_offload = False
 
     def get_dummy_components(self):
         torch.manual_seed(0)
diff --git a/tests/pipelines/repaint/test_repaint.py b/tests/pipelines/repaint/test_repaint.py
index e372cf979ebb..607827854bf7 100644
--- a/tests/pipelines/repaint/test_repaint.py
+++ b/tests/pipelines/repaint/test_repaint.py
@@ -47,7 +47,6 @@ class RepaintPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
         "callback_steps",
     }
     batch_params = IMAGE_INPAINTING_BATCH_PARAMS
-    test_cpu_offload = False
 
     def get_dummy_components(self):
         torch.manual_seed(0)
diff --git a/tests/pipelines/spectrogram_diffusion/test_spectrogram_diffusion.py b/tests/pipelines/spectrogram_diffusion/test_spectrogram_diffusion.py
index cc8690eb87ca..7f5fe80bf10b 100644
--- a/tests/pipelines/spectrogram_diffusion/test_spectrogram_diffusion.py
+++ b/tests/pipelines/spectrogram_diffusion/test_spectrogram_diffusion.py
@@ -48,7 +48,7 @@ class SpectrogramDiffusionPipelineFastTests(PipelineTesterMixin, unittest.TestCa
         "num_images_per_prompt",
     }
     test_attention_slicing = False
-    test_cpu_offload = False
+
     batch_params = TOKENS_TO_AUDIO_GENERATION_PARAMS
     params = TOKENS_TO_AUDIO_GENERATION_BATCH_PARAMS
 
diff --git a/tests/pipelines/stable_diffusion/test_stable_diffusion_sag.py b/tests/pipelines/stable_diffusion/test_stable_diffusion_sag.py
index 91719ce7676f..79d76666c392 100644
--- a/tests/pipelines/stable_diffusion/test_stable_diffusion_sag.py
+++ b/tests/pipelines/stable_diffusion/test_stable_diffusion_sag.py
@@ -42,7 +42,6 @@ class StableDiffusionSAGPipelineFastTests(PipelineLatentTesterMixin, PipelineTes
     batch_params = TEXT_TO_IMAGE_BATCH_PARAMS
     image_params = TEXT_TO_IMAGE_IMAGE_PARAMS
     image_latents_params = TEXT_TO_IMAGE_IMAGE_PARAMS
-    test_cpu_offload = False
 
     def get_dummy_components(self):
         torch.manual_seed(0)
diff --git a/tests/pipelines/stable_diffusion_2/test_stable_diffusion_latent_upscale.py b/tests/pipelines/stable_diffusion_2/test_stable_diffusion_latent_upscale.py
index 5e1d610efcaf..ce55bddc4fe0 100644
--- a/tests/pipelines/stable_diffusion_2/test_stable_diffusion_latent_upscale.py
+++ b/tests/pipelines/stable_diffusion_2/test_stable_diffusion_latent_upscale.py
@@ -63,8 +63,6 @@ class StableDiffusionLatentUpscalePipelineFastTests(
     )  # TO-DO: update image_params once pipeline is refactored with VaeImageProcessor.preprocess
     image_latents_params = frozenset([])
 
-    test_cpu_offload = True
-
     @property
     def dummy_image(self):
         batch_size = 1
diff --git a/tests/pipelines/test_pipelines_common.py b/tests/pipelines/test_pipelines_common.py
index 8f67501d4b23..c522c4b89122 100644
--- a/tests/pipelines/test_pipelines_common.py
+++ b/tests/pipelines/test_pipelines_common.py
@@ -227,7 +227,7 @@ class PipelineTesterMixin:
 
     # set these parameters to False in the child class if the pipeline does not support the corresponding functionality
     test_attention_slicing = True
-    test_cpu_offload = True
+
     test_xformers_attention = True
 
     def get_generator(self, seed):
@@ -671,9 +671,6 @@ def _test_attention_slicing_forward_pass(
         reason="CPU offload is only available with CUDA and `accelerate v0.14.0` or higher",
     )
     def test_cpu_offload_forward_pass(self, expected_max_diff=1e-4):
-        if not self.test_cpu_offload:
-            return
-
         components = self.get_dummy_components()
         pipe = self.pipeline_class(**components)
         pipe.to(torch_device)