@@ -35,7 +35,7 @@ Follow the style introduced in [#14113](https://github.com/huggingface/diffusers
   - `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.
+- **`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.
 - **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.
 
@@ -54,7 +54,7 @@ jobs:
 
       - name: Pipeline Tests Artifacts
         if: ${{ always() }}
-        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f  # v6
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f  # v6.0.0
         with:
           name: test-pipelines.json
           path: reports
@@ -108,7 +108,7 @@ jobs:
           cat reports/tests_pipeline_${{ matrix.module }}_cuda_failures_short.txt
       - name: Test suite reports artifacts
         if: ${{ always() }}
-        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f  # v6
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f  # v6.0.0
         with:
           name: pipeline_${{ matrix.module }}_test_reports
           path: reports
@@ -180,7 +180,7 @@ jobs:
 
     - name: Test suite reports artifacts
       if: ${{ always() }}
-      uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f  # v6
+      uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f  # v6.0.0
       with:
         name: torch_${{ matrix.module }}_cuda_test_reports
         path: reports
@@ -225,7 +225,7 @@ jobs:
 
     - name: Test suite reports artifacts
       if: ${{ always() }}
-      uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f  # v6
+      uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f  # v6.0.0
       with:
         name: torch_compile_test_reports
         path: reports
@@ -279,7 +279,7 @@ jobs:
           cat reports/tests_big_gpu_torch_cuda_failures_short.txt
       - name: Test suite reports artifacts
         if: ${{ always() }}
-        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f  # v6
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f  # v6.0.0
         with:
           name: torch_cuda_big_gpu_test_reports
           path: reports
@@ -323,7 +323,6 @@ jobs:
           pytest -n 1 --max-worker-restart=0 --dist=loadfile \
             --make-reports=tests_torch_minimum_version_cuda \
             tests/models/test_modeling_common.py \
-            tests/pipelines/test_pipelines_common.py \
             tests/pipelines/test_pipeline_utils.py \
             tests/pipelines/test_pipelines.py \
             tests/pipelines/test_pipelines_auto.py \
@@ -338,7 +337,7 @@ jobs:
 
       - name: Test suite reports artifacts
         if: ${{ always() }}
-        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f  # v6
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f  # v6.0.0
         with:
           name: torch_minimum_version_cuda_test_reports
           path: reports
@@ -410,7 +409,7 @@ jobs:
           cat reports/tests_${{ matrix.config.backend }}_torch_cuda_failures_short.txt
       - name: Test suite reports artifacts
         if: ${{ always() }}
-        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f  # v6
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f  # v6.0.0
         with:
           name: torch_cuda_${{ matrix.config.backend }}_reports
           path: reports
@@ -463,7 +462,7 @@ jobs:
           cat reports/tests_pipeline_level_quant_torch_cuda_failures_short.txt
       - name: Test suite reports artifacts
         if: ${{ always() }}
-        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f  # v6
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f  # v6.0.0
         with:
           name: torch_cuda_pipeline_level_quant_reports
           path: reports
@@ -496,7 +495,7 @@ jobs:
         run: mkdir -p combined_reports
 
       - name: Download all test reports
-        uses: actions/download-artifact@37930b1c2abaa49bbe596cd826c3c89aef350131  # v7
+        uses: actions/download-artifact@37930b1c2abaa49bbe596cd826c3c89aef350131  # v7.0.0
         with:
           path: artifacts
 
@@ -522,7 +521,7 @@ jobs:
           cat $CONSOLIDATED_REPORT_PATH >> $GITHUB_STEP_SUMMARY
 
       - name: Upload consolidated report
-        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f  # v6
+        uses: actions/upload-artifact@b7c566a772e6b6bfb58ed0dc250532a479d7789f  # v6.0.0
         with:
           name: consolidated_test_report
           path: ${{ env.CONSOLIDATED_REPORT_PATH }}
 
@@ -12,8 +12,9 @@ on:
       - "src/diffusers/loaders/lora_base.py"
       - "src/diffusers/loaders/lora_pipeline.py"
       - "src/diffusers/loaders/peft.py"
-      - "tests/pipelines/test_pipelines_common.py"
+      - "tests/pipelines/testing_utils/**.py"
       - "tests/models/test_modeling_common.py"
+      - "tests/models/testing_utils/**.py"
       - "examples/**/*.py"
   workflow_dispatch:
 
@@ -94,7 +95,7 @@ jobs:
     runs-on:
       group: aws-general-8-plus
     container:
-      image: diffusers/diffusers-pytorch-cpu
+      image: diffusers/diffusers-pytorch-cpu:latest
     outputs:
       pipeline_test_matrix: ${{ steps.fetch_pipeline_matrix.outputs.pipeline_test_matrix }}
     steps:
@@ -135,7 +136,7 @@ jobs:
     runs-on:
       group: aws-g4dn-2xlarge
     container:
-      image: diffusers/diffusers-pytorch-cuda
+      image: diffusers/diffusers-pytorch-cuda:latest
       options: --shm-size "16gb" --ipc host --gpus all
     steps:
       - name: Checkout diffusers
@@ -201,7 +202,7 @@ jobs:
     runs-on:
       group: aws-g4dn-2xlarge
     container:
-      image: diffusers/diffusers-pytorch-cuda
+      image: diffusers/diffusers-pytorch-cuda:latest
       options: --shm-size "16gb" --ipc host --gpus all
     defaults:
       run:
@@ -273,7 +274,7 @@ jobs:
       group: aws-g4dn-2xlarge
 
     container:
-      image: diffusers/diffusers-pytorch-cuda
+      image: diffusers/diffusers-pytorch-cuda:latest
       options: --gpus all --shm-size "16gb" --ipc host
     steps:
     - name: Checkout diffusers
 
@@ -213,7 +213,6 @@ jobs:
           pytest -n 1 --max-worker-restart=0 --dist=loadfile \
             --make-reports=tests_torch_minimum_cuda \
             tests/models/test_modeling_common.py \
-            tests/pipelines/test_pipelines_common.py \
             tests/pipelines/test_pipeline_utils.py \
             tests/pipelines/test_pipelines.py \
             tests/pipelines/test_pipelines_auto.py \
 
@@ -575,6 +575,8 @@
         title: DreamLite
       - local: api/pipelines/easyanimate
         title: EasyAnimate
+      - local: api/pipelines/echo
+        title: Echo
       - local: api/pipelines/ernie_image
         title: ERNIE-Image
       - local: api/pipelines/flux
 
@@ -0,0 +1,98 @@
+<!-- Copyright 2026 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License. -->
+
+# Echo
+
+[Echo](https://github.com/jd-opensource/JoyAI-Echo) is a long-video generation model. It supports an optional clean
+first frame, ordered image/audio memory slots, and a stochastic few-step Distribution Matching Distillation (DMD)
+sampler.
+The pipeline generates synchronized video and audio.
+
+Echo is implemented as a [`ModularPipeline`] so its text encoding, memory conditioning, stochastic DMD denoising, and
+decoding blocks can be run as a complete workflow or composed independently.
+
+## Inference
+
+Load the official [jdopensource/JoyAI-Echo](https://huggingface.co/jdopensource/JoyAI-Echo) checkpoint directly.
+It uses the text encoder and tokenizer from [google/gemma-3-12b-it](https://huggingface.co/google/gemma-3-12b-it).
+The Gemma repository is gated, so accept its license and authenticate with Hugging Face before loading the pipeline.
+
+The released model uses 241 frames in its long-video example. The video RoPE coordinates remain at the training rate
+of 24 fps, independently of the output container rate.
+
+```py
+import torch
+import torchaudio
+from PIL import Image
+
+from diffusers import ComponentsManager, ModularPipeline
+from diffusers.utils import encode_video
+
+manager = ComponentsManager()
+pipe = ModularPipeline.from_pretrained("jdopensource/JoyAI-Echo", components_manager=manager)
+pipe.load_components(dtype={"default": torch.bfloat16, "audio_vae": torch.float32})
+manager.enable_auto_cpu_offload(device="cuda")
+pipe.vae.enable_tiling()
+
+first_frame = Image.open("first_frame.png").convert("RGB")
+memory_images = [Image.open(path).convert("RGB") for path in ["memory_0.png", "memory_1.png"]]
+memory_audio_with_rates = [torchaudio.load(path) for path in ["memory_0.wav", "memory_1.wav"]]
+memory_audio = [waveform for waveform, _ in memory_audio_with_rates]
+memory_audio_rates = [sample_rate for _, sample_rate in memory_audio_with_rates]
+
+output = pipe(
+    prompt="A cinematic dialogue scene in a quiet cafe.",
+    image=first_frame,
+    memory_images=memory_images,
+    memory_audio_waveforms=memory_audio,
+    memory_audio_sample_rates=memory_audio_rates,
+    width=1280,
+    height=736,
+    num_frames=241,
+    frame_rate=25.0,
+    model_frame_rate=24.0,
+    generator=torch.Generator(device="cuda").manual_seed(42),
+    output_type="np",
+    output=["videos", "audio"],
+)
+
+encode_video(
+    output["videos"][0],
+    fps=25,
+    audio=output["audio"][0].float().cpu(),
+    audio_sample_rate=pipe.vocoder.config.output_sampling_rate,
+    output_path="echo.mp4",
+)
+```
+
+The default DMD sigma schedule is the released eight-step schedule. It predicts `x0` at every step and re-noises with
+fresh Gaussian noise at the next sigma, so a seeded `torch.Generator` controls both the initial noise and all
+intermediate re-noising.
+
+Raw audio-memory encoding requires `torchaudio`. For reference parity, keep `audio_vae` in FP32 as shown above.
+Modular workflows can cache the VAE encoder's normalized, unpacked tensors. Video latents have shape
+`(batch, channels, frames, height, width)` and audio latents have shape `(batch, channels, time, mel_bins)`.
+The core `denoise` block packs those tensors, expands conditioning for `num_videos_per_prompt`, and unpacks
+denoised outputs back to the same VAE form. Pass initial `latents` and `audio_latents` in that unpacked form too.
+Decoders take normalized VAE tensors and denormalize immediately before
+decoding. `output=["latents", "audio_latents"]` returns normalized VAE tensors. With `output_type="latent"` and
+`output=["videos", "audio"]`, you get the denormalized VAE tensors without decoding.
+
+## EchoModularPipeline
+
+[[autodoc]] EchoModularPipeline
+
+## EchoBlocks
+
+[[autodoc]] EchoBlocks
@@ -77,6 +77,26 @@ pipe = QwenImage21Pipeline.from_pretrained("Qwen/Qwen-Image-2.1", transformer=tr
 )
 ```
 
+## Sampling sigmas
+
+Model authors can configure a default sampling grid with `sample_sigmas` in the pipeline config. When you load a
+released checkpoint, its default grid and scheduler settings are restored automatically.
+
+To experiment with a different grid at runtime, pass `sigmas` to the pipeline call:
+
+```python
+# Use the checkpoint's default sampling grid.
+image = pipe(prompt).images[0]
+
+# Override the default grid for this call.
+image = pipe(prompt, sigmas=[1.0, 0.8, 0.5, 0.2]).images[0]
+```
+
+The custom grid above illustrates the API; generation quality depends on the checkpoint and grid. Sigma lists exclude
+the terminal sigma, which the scheduler appends. Explicit `sigmas` override the configured `sample_sigmas`, and either
+list determines the number of steps instead of `num_inference_steps`. If neither is provided, the pipeline uses
+`num_inference_steps` to generate the schedule. The scheduler applies its configured processing to either grid.
+
 ## QwenImage21Pipeline
 
 [[autodoc]] QwenImage21Pipeline