diff --git a/source/isaaclab_visualizers/test/conftest.py b/source/isaaclab_visualizers/test/conftest.py
new file mode 100644
index 000000000000..398b711a5cc5
--- /dev/null
+++ b/source/isaaclab_visualizers/test/conftest.py
@@ -0,0 +1,107 @@
+# Copyright (c) 2022-2026, The Isaac Lab Project Developers (https://github.com/isaac-sim/IsaacLab/blob/main/CONTRIBUTORS.md).
+# All rights reserved.
+#
+# SPDX-License-Identifier: BSD-3-Clause
+
+"""H4 hypothesis: warp 1.13 changed implicit CUDA-OpenGL sync semantics.
+
+Monkey-patches ``newton._src.viewer.viewer_gl.ViewerGL.get_frame`` to insert an
+explicit ``glFinish()`` and ``warp.synchronize_device()`` between
+``glReadPixels`` (PBO async readback) and ``self._wp_pbo.map()`` (CUDA-side
+buffer access). If pre-rc2/1.13 the implicit sync was tighter, this restores it.
+
+If the failing test passes with this in place, the regression is a missed
+sync between GL framebuffer-to-PBO and CUDA-mapped read.
+
+Activated by setting VIZ_H4_EXPLICIT_SYNC=1 in the environment.
+"""
+
+from __future__ import annotations
+
+import ctypes
+
+
+def _patch_viewergl_with_explicit_sync():
+    try:
+        import warp as wp
+
+        from newton._src.utils.render import copy_rgb_frame_uint8
+        from newton._src.viewer.gl.opengl import RendererGL
+        from newton._src.viewer.viewer_gl import ViewerGL
+    except ImportError:
+        return
+
+    if getattr(ViewerGL, "_h4_patched", False):
+        return
+
+    def patched_get_frame(self, target_image=None, render_ui=False):
+        gl = RendererGL.gl
+        w, h = self.renderer._screen_width, self.renderer._screen_height
+
+        if self._pbo is None:
+            pbo_id = (gl.GLuint * 1)()
+            gl.glGenBuffers(1, pbo_id)
+            self._pbo = pbo_id[0]
+            gl.glBindBuffer(gl.GL_PIXEL_PACK_BUFFER, self._pbo)
+            gl.glBufferData(gl.GL_PIXEL_PACK_BUFFER, gl.GLsizeiptr(w * h * 3), None, gl.GL_STREAM_READ)
+            gl.glBindBuffer(gl.GL_PIXEL_PACK_BUFFER, 0)
+            self._wp_pbo = wp.RegisteredGLBuffer(
+                gl_buffer_id=int(self._pbo),
+                device=self.device,
+                flags=wp.RegisteredGLBuffer.READ_ONLY,
+            )
+            gl.glPixelStorei(gl.GL_PACK_ALIGNMENT, 1)
+
+        gl.glBindFramebuffer(gl.GL_FRAMEBUFFER, self.renderer._frame_fbo)
+        gl.glBindBuffer(gl.GL_PIXEL_PACK_BUFFER, self._pbo)
+
+        if render_ui and self.ui:
+            self.ui.begin_frame()
+            self._render_ui()
+            self.ui.end_frame()
+            self.ui.render()
+
+        gl.glReadPixels(0, 0, w, h, gl.GL_RGB, gl.GL_UNSIGNED_BYTE, ctypes.c_void_p(0))
+        gl.glBindBuffer(gl.GL_PIXEL_PACK_BUFFER, 0)
+        gl.glBindFramebuffer(gl.GL_FRAMEBUFFER, 0)
+
+        # H4 patch: force an explicit GL fence + CUDA sync before mapping the
+        # PBO into CUDA address space. warp 1.13 may have weakened the
+        # implicit sync inside ``RegisteredGLBuffer.map``.
+        gl.glFinish()
+        wp.synchronize_device(self.device)
+
+        buf = self._wp_pbo.map(dtype=wp.uint8, shape=(w * h * 3,))
+
+        if target_image is None:
+            target_image = wp.empty(shape=(h, w, 3), dtype=wp.uint8, device=self.device)
+
+        if target_image.shape != (h, w, 3):
+            raise ValueError(f"Shape of `target_image` must be ({h}, {w}, 3), got {target_image.shape}")
+
+        wp.launch(
+            copy_rgb_frame_uint8,
+            dim=(w, h),
+            inputs=[buf, w, h],
+            outputs=[target_image],
+            device=self.device,
+        )
+        self._wp_pbo.unmap()
+
+        # Diagnostic: post-call sample.
+        try:
+            arr = target_image.numpy()
+            print(f"[H4] returned shape={arr.shape}  nonzero={int((arr != 0).sum())}  first24={arr.reshape(-1)[:24].tolist()}")
+        except Exception:
+            pass
+
+        return target_image
+
+    ViewerGL.get_frame = patched_get_frame
+    ViewerGL._h4_patched = True
+    print("[H4] ViewerGL.get_frame patched with explicit glFinish + wp.synchronize_device")
+
+
+def pytest_collection_modifyitems(config, items):  # noqa: ARG001
+    """Hook fires after test files are collected (i.e., after AppLauncher in test files runs)."""
+    _patch_viewergl_with_explicit_sync()