How to unfuse_lora only the first one after I have added multiple lora?

[yanchaoguo]

base.load_lora_weights("models/safetensors/SDXL/国风插画SDXL.safetensors")
base.fuse_lora(lora_scale=.7)
base.load_lora_weights("models/safetensors/SDXL/sd_xl_offset_example-lora_1.0.safetensors")
base.fuse_lora(lora_scale=.8)
Now, When I execute unfuse_lora() only the most recent one has been unfuse .

so,how to unfuse '国风插画SDXL.safetensors' or unfuse all lora weights

[patrickvonplaten]

Good question! This is currently not possible sadly 😅

As soon as you've fused two LoRAs you can sadly not go back anymore and will have to reload the full original model weights.

It's a TODO!

cc @apolinario @sayakpaul

[alexblattner]

@patrickvonplaten I highly recommend getting inspiration from this: https://gist.github.com/takuma104/e38d683d72b1e448b8d9b3835f7cfa44#gistcomment-4666052

it's better than what is currently implemented in diffusers except it has no lycoris support

[yanchaoguo]

@patrickvonplaten I highly recommend getting inspiration from this: https://gist.github.com/takuma104/e38d683d72b1e448b8d9b3835f7cfa44#gistcomment-4666052

it's better than what is currently implemented in diffusers except it has no lycoris support

It isn’t suppored SDXL

[alexblattner]

@yanchaoguo to be fair, it was made before SDXL came out. My point is that this code is better overall if you ignore the fact that it doesn't support SDXL or Lycoris

[Damarcreative]

I tried using stable diffusion 1.5 I got the output:

if Use_LoRA_add_detail:
  pipe.load_lora_weights(".", weight_name="/content/model/lora/add_detail.safetensors")
  pipe.fuse_lora()

if Use_LoRA_Kimono:
  pipe.load_lora_weights(".", weight_name="/content/model/lora/kimono.safetensors")
  pipe.fuse_lora()

if Use_LoRA_Graffiti_Poster_Comic_Style:
  pipe.load_lora_weights(".", weight_name="/content/model/lora/mjty-000016.safetensors")
  pipe.fuse_lora()

Download Results:
gid   |stat|avg speed  |path/URI
======+====+===========+=======================================================
bfa0f2|OK  |    12MiB/s|/content/model/lora/add_detail.safetensors

Status Legend:
(OK):download completed.

Download Results:
gid   |stat|avg speed  |path/URI
======+====+===========+=======================================================
e1f247|OK  |    18MiB/s|/content/model/lora/kimono.safetensors

Status Legend:
(OK):download completed.

Download Results:
gid   |stat|avg speed  |path/URI
======+====+===========+=======================================================
66da65|OK  |    28MiB/s|/content/model/lora/mjty-000016.safetensors

Status Legend:
(OK):download completed.
The current API is supported for operating with a single LoRA file. You are trying to load and fuse more than one LoRA which is not well-supported.
The current API is supported for operating with a single LoRA file. You are trying to load and fuse more than one LoRA which is not well-supported.

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.