huggingface · sayakpaul · Sep 1, 2026 · Aug 29, 2026 · Aug 30, 2026 · Aug 31, 2026
diff --git a/.ai/references/testing.md b/.ai/references/testing.md
@@ -34,6 +34,10 @@ Follow the style introduced in [#14113](https://github.com/huggingface/diffusers
   - `torch.nn.MultiheadAttention` is the common instance: it passes `self.out_proj.weight` straight to `torch.nn.functional.multi_head_attention_forward` instead of calling `self.out_proj`, so the hook on `out_proj` never fires. `SiglipVisionModel`'s attention pooling head wraps one — see `tests/pipelines/hunyuan_video/test_hunyuan_video_framepack.py`, whose `image_encoder` is excluded for this reason.
   - `HunyuanDiTAttentionPool` (`src/diffusers/models/embeddings.py`) shows the same failure without an MHA module: a plain `nn.Module` that hands its `q_proj` / `k_proj` / `v_proj` / `c_proj` weights to `torch.nn.functional.multi_head_attention_forward`, so all four projections stay offloaded rather than just one. `HunyuanDiT2DModel` opts out of group offloading entirely with `_supports_group_offloading = False`.
   - Before adding a skip or an exclusion, confirm the failure still reproduces — several existing skips are stale, having outlived the upstream cause.
+- **A migration that surfaces a `src/` gap marks the test `xfail`, it does not patch the pipeline.** Give the marker a module-level name and a `reason` naming the exact gap (`PNDM_*` in `tests/pipelines/pndm/test_pndm.py` is the worked example), and prefer `strict=True` so the marker reports XPASS — and gets deleted — the day the pipeline is fixed. Use `strict=False` only when one mark covers a group whose members do not all fail. Marking a whole test class keeps the mixin's own marks (`@is_memory`, `@require_accelerator`) intact; overriding individual inherited tests drops the decorators they were declared with, so re-declare those too.
+- **`from_pipe` tests** (a pipeline that is a variant of an existing one — PAG, AnimateDiff, ...) compose the shared `FromPipeTesterMixin` (`tests/pipelines/testing_utils/from_pipe.py`, exported from `..testing_utils`) in their own test class. It derives the original pipeline from `pipeline_class.__name__`; set `original_pipeline_repo` on the test class to pull it from a repo other than the default for that class. The unittest-era `PipelineFromPipeTesterMixin` in `tests/pipelines/test_pipelines_common.py` is what it replaces.
+- **A hardware gap is a conditional skip, not an xfail.** When a test fails only because the runner's cuDNN build has no kernel for an op — `RuntimeError: GET was unable to find an engine to execute this computation`, as Sana's depthwise `Conv2d` hits in bfloat16 — wrap the call in `skip_if_no_cudnn_engine()` (`tests/testing_utils.py`). It skips on that error and re-raises every other `RuntimeError`, so the test still runs wherever the kernel exists.
+- **PAG pipelines** also compose `PAGPipelineTesterMixin` (`tests/pipelines/pag/testing_utils.py`) in place of `PipelineTesterMixin`: it adds `test_pag_disable_enable` and `test_pag_inference` on top, driven by `base_pipeline_class` and the `pag_*` knobs on the test class. Keep `test_pag_applied_layers` per pipeline — which layers PAG resolves to is model-specific.
 - **`encode_prompt` reading a component that isn't a text encoder or tokenizer?** `test_encode_prompt_works_in_isolation` rebuilds the pipeline with only the components whose names contain `text` or `tokenizer`. When `encode_prompt` also needs another one — a `processor` used for chat templating, say — list it in `text_stack_component_names` on the config class rather than re-implementing the test.
 - **IP-Adapter tests** live in their own class decorated with `@is_ip_adapter`, subclassing only the config (not `PipelineTesterMixin`). UNet pipelines that load adapters through the standard `IPAdapterMixin` API compose the shared `IPAdapterTesterMixin` (`tests/pipelines/testing_utils/ip_adapter.py`, exported from `..testing_utils`); pipelines whose IP-Adapter API differs (Flux, for example) keep a bespoke mixin next to their own tests.
 

diff --git a/src/diffusers/pipelines/omnigen/pipeline_omnigen.py b/src/diffusers/pipelines/omnigen/pipeline_omnigen.py
@@ -116,9 +116,7 @@ def retrieve_timesteps(
     return timesteps, num_inference_steps
 
 
-class OmniGenPipeline(
-    DiffusionPipeline,
-):
+class OmniGenPipeline(DiffusionPipeline):
     r"""
     The OmniGen pipeline for multimodal-to-image generation.
 

diff --git a/tests/pipelines/animatediff/test_animatediff.py b/tests/pipelines/animatediff/test_animatediff.py
@@ -10,7 +10,6 @@
     AutoencoderKL,
     DDIMScheduler,
     MotionAdapter,
-    StableDiffusionPipeline,
     UNet2DConditionModel,
 )
 
@@ -22,16 +21,15 @@
     torch_device,
 )
 from ..pipeline_params import TEXT_TO_IMAGE_BATCH_PARAMS, TEXT_TO_IMAGE_PARAMS
-from ..test_pipelines_common import PipelineFromPipeTesterMixin
 from ..testing_utils import (
+    FromPipeTesterMixin,
     IPAdapterTesterMixin,
     LoraMemoryTesterMixin,
     LoraTesterMixin,
     MemoryTesterMixin,
     UNetLoraTesterMixin,
 )
 from .testing_utils import (
-    FROM_PIPE_SKIP_REASON,
     FreeInitTesterMixin,
     FreeNoiseSplitInferenceTesterMixin,
     MotionPipelineTesterConfig,
@@ -128,32 +126,6 @@ class TestAnimateDiffPipeline(
     FreeInitTesterMixin,
     FreeNoiseSplitInferenceTesterMixin,
 ):
-    def test_from_pipe_consistent_config(self):
-        original_repo = "hf-internal-testing/tinier-stable-diffusion-pipe"
-
-        # create StableDiffusionPipeline
-        pipe_original = StableDiffusionPipeline.from_pretrained(original_repo, requires_safety_checker=False)
-
-        # StableDiffusionPipeline -> AnimateDiffPipeline
-        pipe_components = self.get_dummy_components()
-        pipe_additional_components = {
-            name: component for name, component in pipe_components.items() if name not in pipe_original.components
-        }
-        pipe = self.pipeline_class.from_pipe(pipe_original, **pipe_additional_components)
-
-        # AnimateDiffPipeline -> StableDiffusionPipeline
-        original_pipe_additional_components = {}
-        for name, component in pipe_original.components.items():
-            if name not in pipe.components or not isinstance(component, pipe.components[name].__class__):
-                original_pipe_additional_components[name] = component
-
-        pipe_original_2 = StableDiffusionPipeline.from_pipe(pipe, **original_pipe_additional_components)
-
-        # compare the config
-        original_config = {k: v for k, v in pipe_original.config.items() if not k.startswith("_")}
-        original_config_2 = {k: v for k, v in pipe_original_2.config.items() if not k.startswith("_")}
-        assert original_config_2 == original_config
-
     def test_dict_tuple_outputs_equivalent(self, expected_slice=None, expected_max_difference=1e-4):
         if torch_device == "cpu" and expected_slice is None:
             # fmt: off
@@ -244,10 +216,7 @@ def test_animatediff(self):
         assert numpy_cosine_similarity_distance(image_slice.flatten(), expected_slice.flatten()) < 1e-3
 
 
-@pytest.mark.skip(FROM_PIPE_SKIP_REASON)
-class TestAnimateDiffPipelineFromPipe(AnimateDiffPipelineTesterConfig, PipelineFromPipeTesterMixin):
-    """`from_pipe` forward-pass parity and offload round trip for the AnimateDiff pipeline.
+class TestAnimateDiffPipelineFromPipe(AnimateDiffPipelineTesterConfig, FromPipeTesterMixin):
+    """`from_pipe` round-trip tests against `StableDiffusionPipeline` for the AnimateDiff pipeline."""
 
-    Parked, not deleted: `test_from_pipe_consistent_config` runs for real as a method on the main test class above,
-    but the forward-pass checks in `PipelineFromPipeTesterMixin` have no pytest-style equivalent yet.
-    """
+    original_pipeline_repo = "hf-internal-testing/tinier-stable-diffusion-pipe"
diff --git a/tests/pipelines/animatediff/test_animatediff_controlnet.py b/tests/pipelines/animatediff/test_animatediff_controlnet.py
@@ -1,4 +1,3 @@
-import pytest
 import torch
 from PIL import Image
 from transformers import CLIPTextConfig, CLIPTextModel, CLIPTokenizer
@@ -9,22 +8,20 @@
     ControlNetModel,
     DDIMScheduler,
     MotionAdapter,
-    StableDiffusionPipeline,
     UNet2DConditionModel,
 )
 
 from ...testing_utils import torch_device
 from ..pipeline_params import TEXT_TO_IMAGE_BATCH_PARAMS, TEXT_TO_IMAGE_PARAMS
-from ..test_pipelines_common import PipelineFromPipeTesterMixin
 from ..testing_utils import (
+    FromPipeTesterMixin,
     IPAdapterTesterMixin,
     LoraMemoryTesterMixin,
     LoraTesterMixin,
     MemoryTesterMixin,
     UNetLoraTesterMixin,
 )
 from .testing_utils import (
-    FROM_PIPE_SKIP_REASON,
     FreeInitTesterMixin,
     FreeNoiseTesterMixin,
     MotionPipelineTesterConfig,
@@ -142,32 +139,6 @@ def get_free_noise_inputs(self):
         # `get_dummy_inputs` rather than by overriding `num_frames` on the returned dict.
         return self.get_dummy_inputs(num_frames=16)
 
-    def test_from_pipe_consistent_config(self):
-        original_repo = "hf-internal-testing/tinier-stable-diffusion-pipe"
-
-        # create StableDiffusionPipeline
-        pipe_original = StableDiffusionPipeline.from_pretrained(original_repo, requires_safety_checker=False)
-
-        # StableDiffusionPipeline -> AnimateDiffControlNetPipeline
-        pipe_components = self.get_dummy_components()
-        pipe_additional_components = {
-            name: component for name, component in pipe_components.items() if name not in pipe_original.components
-        }
-        pipe = self.pipeline_class.from_pipe(pipe_original, **pipe_additional_components)
-
-        # AnimateDiffControlNetPipeline -> StableDiffusionPipeline
-        original_pipe_additional_components = {}
-        for name, component in pipe_original.components.items():
-            if name not in pipe.components or not isinstance(component, pipe.components[name].__class__):
-                original_pipe_additional_components[name] = component
-
-        pipe_original_2 = StableDiffusionPipeline.from_pipe(pipe, **original_pipe_additional_components)
-
-        # compare the config
-        original_config = {k: v for k, v in pipe_original.config.items() if not k.startswith("_")}
-        original_config_2 = {k: v for k, v in pipe_original_2.config.items() if not k.startswith("_")}
-        assert original_config_2 == original_config
-
     def test_dict_tuple_outputs_equivalent(self, expected_slice=None, expected_max_difference=1e-4):
         if torch_device == "cpu" and expected_slice is None:
             # fmt: off
@@ -198,12 +169,7 @@ class TestAnimateDiffControlNetPipelineLoRAMemory(AnimateDiffControlNetPipelineT
     """LoRA x memory-optimization tests (group offload, CPU offload) for the pipeline."""
 
 
-@pytest.mark.skip(FROM_PIPE_SKIP_REASON)
-class TestAnimateDiffControlNetPipelineFromPipe(
-    AnimateDiffControlNetPipelineTesterConfig, PipelineFromPipeTesterMixin
-):
-    """`from_pipe` forward-pass parity and offload round trip for the AnimateDiff ControlNet pipeline.
+class TestAnimateDiffControlNetPipelineFromPipe(AnimateDiffControlNetPipelineTesterConfig, FromPipeTesterMixin):
+    """`from_pipe` round-trip tests against `StableDiffusionPipeline` for the AnimateDiff ControlNet pipeline."""
 
-    Parked, not deleted: `test_from_pipe_consistent_config` runs for real as a method on the main test class above,
-    but the forward-pass checks in `PipelineFromPipeTesterMixin` have no pytest-style equivalent yet.
-    """
+    original_pipeline_repo = "hf-internal-testing/tinier-stable-diffusion-pipe"
diff --git a/tests/pipelines/animatediff/test_animatediff_sparsectrl.py b/tests/pipelines/animatediff/test_animatediff_sparsectrl.py
@@ -1,4 +1,3 @@
-import pytest
 import torch
 from PIL import Image
 from transformers import CLIPTextConfig, CLIPTextModel, CLIPTokenizer
@@ -9,22 +8,20 @@
     DDIMScheduler,
     MotionAdapter,
     SparseControlNetModel,
-    StableDiffusionPipeline,
     UNet2DConditionModel,
 )
 
 from ...testing_utils import assert_tensors_close, torch_device
 from ..pipeline_params import TEXT_TO_IMAGE_BATCH_PARAMS, TEXT_TO_IMAGE_PARAMS
-from ..test_pipelines_common import PipelineFromPipeTesterMixin
 from ..testing_utils import (
+    FromPipeTesterMixin,
     IPAdapterTesterMixin,
     LoraMemoryTesterMixin,
     LoraTesterMixin,
     MemoryTesterMixin,
     UNetLoraTesterMixin,
 )
 from .testing_utils import (
-    FROM_PIPE_SKIP_REASON,
     FreeInitTesterMixin,
     MotionPipelineTesterConfig,
     MotionPipelineTesterMixin,
@@ -138,32 +135,6 @@ class TestAnimateDiffSparseControlNetPipeline(
     MotionPipelineTesterMixin,
     FreeInitTesterMixin,
 ):
-    def test_from_pipe_consistent_config(self):
-        original_repo = "hf-internal-testing/tinier-stable-diffusion-pipe"
-
-        # create StableDiffusionPipeline
-        pipe_original = StableDiffusionPipeline.from_pretrained(original_repo, requires_safety_checker=False)
-
-        # StableDiffusionPipeline -> AnimateDiffSparseControlNetPipeline
-        pipe_components = self.get_dummy_components()
-        pipe_additional_components = {
-            name: component for name, component in pipe_components.items() if name not in pipe_original.components
-        }
-        pipe = self.pipeline_class.from_pipe(pipe_original, **pipe_additional_components)
-
-        # AnimateDiffSparseControlNetPipeline -> StableDiffusionPipeline
-        original_pipe_additional_components = {}
-        for name, component in pipe_original.components.items():
-            if name not in pipe.components or not isinstance(component, pipe.components[name].__class__):
-                original_pipe_additional_components[name] = component
-
-        pipe_original_2 = StableDiffusionPipeline.from_pipe(pipe, **original_pipe_additional_components)
-
-        # compare the config
-        original_config = {k: v for k, v in pipe_original.config.items() if not k.startswith("_")}
-        original_config_2 = {k: v for k, v in pipe_original_2.config.items() if not k.startswith("_")}
-        assert original_config_2 == original_config
-
     def test_dict_tuple_outputs_equivalent(self, expected_slice=None, expected_max_difference=1e-4):
         if torch_device == "cpu" and expected_slice is None:
             # fmt: off
@@ -229,12 +200,9 @@ class TestAnimateDiffSparseControlNetPipelineLoRAMemory(
     """LoRA x memory-optimization tests (group offload, CPU offload) for the pipeline."""
 
 
-@pytest.mark.skip(FROM_PIPE_SKIP_REASON)
 class TestAnimateDiffSparseControlNetPipelineFromPipe(
-    AnimateDiffSparseControlNetPipelineTesterConfig, PipelineFromPipeTesterMixin
+    AnimateDiffSparseControlNetPipelineTesterConfig, FromPipeTesterMixin
 ):
-    """`from_pipe` forward-pass parity and offload round trip for the AnimateDiff SparseControlNet pipeline.
+    """`from_pipe` round-trip tests against `StableDiffusionPipeline` for the AnimateDiff SparseControlNet pipeline."""
 
-    Parked, not deleted: `test_from_pipe_consistent_config` runs for real as a method on the main test class above,
-    but the forward-pass checks in `PipelineFromPipeTesterMixin` have no pytest-style equivalent yet.
-    """
+    original_pipeline_repo = "hf-internal-testing/tinier-stable-diffusion-pipe"
diff --git a/tests/pipelines/animatediff/test_animatediff_video2video.py b/tests/pipelines/animatediff/test_animatediff_video2video.py
@@ -1,4 +1,3 @@
-import pytest
 import torch
 from PIL import Image
 from transformers import CLIPTextConfig, CLIPTextModel, CLIPTokenizer
@@ -8,22 +7,20 @@
     AutoencoderKL,
     DDIMScheduler,
     MotionAdapter,
-    StableDiffusionPipeline,
     UNet2DConditionModel,
 )
 
 from ...testing_utils import torch_device
 from ..pipeline_params import TEXT_TO_IMAGE_PARAMS, VIDEO_TO_VIDEO_BATCH_PARAMS
-from ..test_pipelines_common import PipelineFromPipeTesterMixin
 from ..testing_utils import (
+    FromPipeTesterMixin,
     IPAdapterTesterMixin,
     LoraMemoryTesterMixin,
     LoraTesterMixin,
     MemoryTesterMixin,
     UNetLoraTesterMixin,
 )
 from .testing_utils import (
-    FROM_PIPE_SKIP_REASON,
     FreeInitTesterMixin,
     FreeNoiseSplitInferenceTesterMixin,
     MotionPipelineTesterConfig,
@@ -133,32 +130,6 @@ def get_free_noise_inputs(self):
         inputs["strength"] = 0.5
         return inputs
 
-    def test_from_pipe_consistent_config(self):
-        original_repo = "hf-internal-testing/tinier-stable-diffusion-pipe"
-
-        # create StableDiffusionPipeline
-        pipe_original = StableDiffusionPipeline.from_pretrained(original_repo, requires_safety_checker=False)
-
-        # StableDiffusionPipeline -> AnimateDiffVideoToVideoPipeline
-        pipe_components = self.get_dummy_components()
-        pipe_additional_components = {
-            name: component for name, component in pipe_components.items() if name not in pipe_original.components
-        }
-        pipe = self.pipeline_class.from_pipe(pipe_original, **pipe_additional_components)
-
-        # AnimateDiffVideoToVideoPipeline -> StableDiffusionPipeline
-        original_pipe_additional_components = {}
-        for name, component in pipe_original.components.items():
-            if name not in pipe.components or not isinstance(component, pipe.components[name].__class__):
-                original_pipe_additional_components[name] = component
-
-        pipe_original_2 = StableDiffusionPipeline.from_pipe(pipe, **original_pipe_additional_components)
-
-        # compare the config
-        original_config = {k: v for k, v in pipe_original.config.items() if not k.startswith("_")}
-        original_config_2 = {k: v for k, v in pipe_original_2.config.items() if not k.startswith("_")}
-        assert original_config_2 == original_config
-
     def test_latent_inputs(self):
         pipe = self.get_pipeline().to(torch_device)
 
@@ -191,12 +162,7 @@ class TestAnimateDiffVideoToVideoPipelineLoRAMemory(
     """LoRA x memory-optimization tests (group offload, CPU offload) for the pipeline."""
 
 
-@pytest.mark.skip(FROM_PIPE_SKIP_REASON)
-class TestAnimateDiffVideoToVideoPipelineFromPipe(
-    AnimateDiffVideoToVideoPipelineTesterConfig, PipelineFromPipeTesterMixin
-):
-    """`from_pipe` forward-pass parity and offload round trip for the AnimateDiff video-to-video pipeline.
+class TestAnimateDiffVideoToVideoPipelineFromPipe(AnimateDiffVideoToVideoPipelineTesterConfig, FromPipeTesterMixin):
+    """`from_pipe` round-trip tests against `StableDiffusionPipeline` for the AnimateDiff video-to-video pipeline."""
 
-    Parked, not deleted: `test_from_pipe_consistent_config` runs for real as a method on the main test class above,
-    but the forward-pass checks in `PipelineFromPipeTesterMixin` have no pytest-style equivalent yet.
-    """
+    original_pipeline_repo = "hf-internal-testing/tinier-stable-diffusion-pipe"