Merge branch 'main' into modular-diffusers-wan

a-r-r-o-w · web-flow · commit 348b07a75a1c · 2025-07-23T18:48:24.000+05:30
diff --git a/src/diffusers/pipelines/chroma/pipeline_chroma.py b/src/diffusers/pipelines/chroma/pipeline_chroma.py
@@ -663,11 +663,11 @@ def __call__(
                 their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
                 will be used.
             guidance_scale (`float`, *optional*, defaults to 3.5):
-                Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
-                `guidance_scale` is defined as `w` of equation 2. of [Imagen
-                Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale >
-                1`. Higher guidance scale encourages to generate images that are closely linked to the text `prompt`,
-                usually at the expense of lower image quality.
+                Embedded guiddance scale is enabled by setting `guidance_scale` > 1. Higher `guidance_scale` encourages
+                a model to generate images more aligned with `prompt` at the expense of lower image quality.
+
+                Guidance-distilled models approximates true classifer-free guidance for `guidance_scale` > 1. Refer to
+                the [paper](https://huggingface.co/papers/2210.03142) to learn more.
             num_images_per_prompt (`int`, *optional*, defaults to 1):
                 The number of images to generate per prompt.
             generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
diff --git a/src/diffusers/pipelines/chroma/pipeline_chroma_img2img.py b/src/diffusers/pipelines/chroma/pipeline_chroma_img2img.py
@@ -725,11 +725,11 @@ def __call__(
                 their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
                 will be used.
             guidance_scale (`float`, *optional*, defaults to 5.0):
-                Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
-                `guidance_scale` is defined as `w` of equation 2. of [Imagen
-                Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale >
-                1`. Higher guidance scale encourages to generate images that are closely linked to the text `prompt`,
-                usually at the expense of lower image quality.
+                Embedded guiddance scale is enabled by setting `guidance_scale` > 1. Higher `guidance_scale` encourages
+                a model to generate images more aligned with `prompt` at the expense of lower image quality.
+
+                Guidance-distilled models approximates true classifer-free guidance for `guidance_scale` > 1. Refer to
+                the [paper](https://huggingface.co/papers/2210.03142) to learn more.
             strength (`float, *optional*, defaults to 0.9):
                 Conceptually, indicates how much to transform the reference image. Must be between 0 and 1. image will
                 be used as a starting point, adding more noise to it the larger the strength. The number of denoising
diff --git a/src/diffusers/pipelines/flux/pipeline_flux.py b/src/diffusers/pipelines/flux/pipeline_flux.py
@@ -674,7 +674,8 @@ def __call__(
                 The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
                 `text_encoder_2`. If not defined, `negative_prompt` is used in all the text-encoders.
             true_cfg_scale (`float`, *optional*, defaults to 1.0):
-                When > 1.0 and a provided `negative_prompt`, enables true classifier-free guidance.
+                True classifier-free guidance (guidance scale) is enabled when `true_cfg_scale` > 1 and
+                `negative_prompt` is provided.
             height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
                 The height in pixels of the generated image. This is set to 1024 by default for the best results.
             width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
@@ -687,11 +688,11 @@ def __call__(
                 their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
                 will be used.
             guidance_scale (`float`, *optional*, defaults to 3.5):
-                Guidance scale as defined in [Classifier-Free Diffusion
-                Guidance](https://huggingface.co/papers/2207.12598). `guidance_scale` is defined as `w` of equation 2.
-                of [Imagen Paper](https://huggingface.co/papers/2205.11487). Guidance scale is enabled by setting
-                `guidance_scale > 1`. Higher guidance scale encourages to generate images that are closely linked to
-                the text `prompt`, usually at the expense of lower image quality.
+                Embedded guiddance scale is enabled by setting `guidance_scale` > 1. Higher `guidance_scale` encourages
+                a model to generate images more aligned with `prompt` at the expense of lower image quality.
+
+                Guidance-distilled models approximates true classifer-free guidance for `guidance_scale` > 1. Refer to
+                the [paper](https://huggingface.co/papers/2210.03142) to learn more.
             num_images_per_prompt (`int`, *optional*, defaults to 1):
                 The number of images to generate per prompt.
             generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
diff --git a/src/diffusers/pipelines/flux/pipeline_flux_control.py b/src/diffusers/pipelines/flux/pipeline_flux_control.py
@@ -661,11 +661,11 @@ def __call__(
                 their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
                 will be used.
             guidance_scale (`float`, *optional*, defaults to 3.5):
-                Guidance scale as defined in [Classifier-Free Diffusion
-                Guidance](https://huggingface.co/papers/2207.12598). `guidance_scale` is defined as `w` of equation 2.
-                of [Imagen Paper](https://huggingface.co/papers/2205.11487). Guidance scale is enabled by setting
-                `guidance_scale > 1`. Higher guidance scale encourages to generate images that are closely linked to
-                the text `prompt`, usually at the expense of lower image quality.
+                Embedded guidance scale is enabled by setting `guidance_scale` > 1. Higher `guidance_scale` encourages
+                a model to generate images more aligned with prompt at the expense of lower image quality.
+
+                Guidance-distilled models approximates true classifier-free guidance for `guidance_scale` > 1. Refer to
+                the [paper](https://huggingface.co/papers/2210.03142) to learn more.
             num_images_per_prompt (`int`, *optional*, defaults to 1):
                 The number of images to generate per prompt.
             generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
diff --git a/src/diffusers/pipelines/flux/pipeline_flux_kontext.py b/src/diffusers/pipelines/flux/pipeline_flux_kontext.py
@@ -795,11 +795,11 @@ def __call__(
                 their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
                 will be used.
             guidance_scale (`float`, *optional*, defaults to 3.5):
-                Guidance scale as defined in [Classifier-Free Diffusion
-                Guidance](https://huggingface.co/papers/2207.12598). `guidance_scale` is defined as `w` of equation 2.
-                of [Imagen Paper](https://huggingface.co/papers/2205.11487). Guidance scale is enabled by setting
-                `guidance_scale > 1`. Higher guidance scale encourages to generate images that are closely linked to
-                the text `prompt`, usually at the expense of lower image quality.
+                Embedded guidance scale is enabled by setting `guidance_scale` > 1. Higher `guidance_scale` encourages
+                a model to generate images more aligned with prompt at the expense of lower image quality.
+
+                Guidance-distilled models approximates true classifier-free guidance for `guidance_scale` > 1. Refer to
+                the [paper](https://huggingface.co/papers/2210.03142) to learn more.
             num_images_per_prompt (`int`, *optional*, defaults to 1):
                 The number of images to generate per prompt.
             generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
diff --git a/src/diffusers/pipelines/flux/pipeline_flux_kontext_inpaint.py b/src/diffusers/pipelines/flux/pipeline_flux_kontext_inpaint.py
@@ -989,7 +989,8 @@ def __call__(
                 The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
                 `text_encoder_2`. If not defined, `negative_prompt` is used in all the text-encoders.
             true_cfg_scale (`float`, *optional*, defaults to 1.0):
-                When > 1.0 and a provided `negative_prompt`, enables true classifier-free guidance.
+                True classifier-free guidance (guidance scale) is enabled when `true_cfg_scale` > 1 and
+                `negative_prompt` is provided.
             height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
                 The height in pixels of the generated image. This is set to 1024 by default for the best results.
             width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
@@ -1015,11 +1016,11 @@ def __call__(
                 their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
                 will be used.
             guidance_scale (`float`, *optional*, defaults to 3.5):
-                Guidance scale as defined in [Classifier-Free Diffusion
-                Guidance](https://huggingface.co/papers/2207.12598). `guidance_scale` is defined as `w` of equation 2.
-                of [Imagen Paper](https://huggingface.co/papers/2205.11487). Guidance scale is enabled by setting
-                `guidance_scale > 1`. Higher guidance scale encourages to generate images that are closely linked to
-                the text `prompt`, usually at the expense of lower image quality.
+                Embedded guidance scale is enabled by setting `guidance_scale` > 1. Higher `guidance_scale` encourages
+                a model to generate images more aligned with `prompt` at the expense of lower image quality.
+
+                Guidance-distilled models approximates true classifier-free guidance for `guidance_scale` > 1. Refer to
+                the [paper](https://huggingface.co/papers/2210.03142) to learn more.
             num_images_per_prompt (`int`, *optional*, defaults to 1):
                 The number of images to generate per prompt.
             generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
diff --git a/src/diffusers/pipelines/hidream_image/pipeline_hidream_image.py b/src/diffusers/pipelines/hidream_image/pipeline_hidream_image.py
@@ -763,11 +763,11 @@ def __call__(
                 their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
                 will be used.
             guidance_scale (`float`, *optional*, defaults to 3.5):
-                Guidance scale as defined in [Classifier-Free Diffusion
-                Guidance](https://huggingface.co/papers/2207.12598). `guidance_scale` is defined as `w` of equation 2.
-                of [Imagen Paper](https://huggingface.co/papers/2205.11487). Guidance scale is enabled by setting
-                `guidance_scale > 1`. Higher guidance scale encourages to generate images that are closely linked to
-                the text `prompt`, usually at the expense of lower image quality.
+                Embedded guiddance scale is enabled by setting `guidance_scale` > 1. Higher `guidance_scale` encourages
+                a model to generate images more aligned with `prompt` at the expense of lower image quality.
+
+                Guidance-distilled models approximates true classifer-free guidance for `guidance_scale` > 1. Refer to
+                the [paper](https://huggingface.co/papers/2210.03142) to learn more.
             negative_prompt (`str` or `List[str]`, *optional*):
                 The prompt or prompts not to guide the image generation. If not defined, one has to pass
                 `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `true_cfg_scale` is
diff --git a/src/diffusers/pipelines/hunyuan_video/pipeline_hunyuan_video.py b/src/diffusers/pipelines/hunyuan_video/pipeline_hunyuan_video.py
@@ -529,15 +529,14 @@ def __call__(
                 their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
                 will be used.
             true_cfg_scale (`float`, *optional*, defaults to 1.0):
-                When > 1.0 and a provided `negative_prompt`, enables true classifier-free guidance.
+                True classifier-free guidance (guidance scale) is enabled when `true_cfg_scale` > 1 and
+                `negative_prompt` is provided.
             guidance_scale (`float`, defaults to `6.0`):
-                Guidance scale as defined in [Classifier-Free Diffusion
-                Guidance](https://huggingface.co/papers/2207.12598). `guidance_scale` is defined as `w` of equation 2.
-                of [Imagen Paper](https://huggingface.co/papers/2205.11487). Guidance scale is enabled by setting
-                `guidance_scale > 1`. Higher guidance scale encourages to generate images that are closely linked to
-                the text `prompt`, usually at the expense of lower image quality. Note that the only available
-                HunyuanVideo model is CFG-distilled, which means that traditional guidance between unconditional and
-                conditional latent is not applied.
+                Embedded guiddance scale is enabled by setting `guidance_scale` > 1. Higher `guidance_scale` encourages
+                a model to generate images more aligned with `prompt` at the expense of lower image quality.
+
+                Guidance-distilled models approximates true classifer-free guidance for `guidance_scale` > 1. Refer to
+                the [paper](https://huggingface.co/papers/2210.03142) to learn more.
             num_videos_per_prompt (`int`, *optional*, defaults to 1):
                 The number of images to generate per prompt.
             generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
diff --git a/src/diffusers/pipelines/sana/pipeline_sana_sprint.py b/src/diffusers/pipelines/sana/pipeline_sana_sprint.py
@@ -643,11 +643,11 @@ def __call__(
                 in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is
                 passed will be used. Must be in descending order.
             guidance_scale (`float`, *optional*, defaults to 4.5):
-                Guidance scale as defined in [Classifier-Free Diffusion
-                Guidance](https://huggingface.co/papers/2207.12598). `guidance_scale` is defined as `w` of equation 2.
-                of [Imagen Paper](https://huggingface.co/papers/2205.11487). Guidance scale is enabled by setting
-                `guidance_scale > 1`. Higher guidance scale encourages to generate images that are closely linked to
-                the text `prompt`, usually at the expense of lower image quality.
+                Embedded guiddance scale is enabled by setting `guidance_scale` > 1. Higher `guidance_scale` encourages
+                a model to generate images more aligned with `prompt` at the expense of lower image quality.
+
+                Guidance-distilled models approximates true classifer-free guidance for `guidance_scale` > 1. Refer to
+                the [paper](https://huggingface.co/papers/2210.03142) to learn more.
             num_images_per_prompt (`int`, *optional*, defaults to 1):
                 The number of images to generate per prompt.
             height (`int`, *optional*, defaults to self.unet.config.sample_size):
diff --git a/tests/pipelines/wan/test_wan.py b/tests/pipelines/wan/test_wan.py
@@ -15,7 +15,6 @@
 import gc
 import unittest
 
-import numpy as np
 import torch
 from transformers import AutoTokenizer, T5EncoderModel
 
@@ -29,9 +28,7 @@
 )
 
 from ..pipeline_params import TEXT_TO_IMAGE_BATCH_PARAMS, TEXT_TO_IMAGE_IMAGE_PARAMS, TEXT_TO_IMAGE_PARAMS
-from ..test_pipelines_common import (
-    PipelineTesterMixin,
-)
+from ..test_pipelines_common import PipelineTesterMixin
 
 
 enable_full_determinism()
@@ -127,11 +124,15 @@ def test_inference(self):
         inputs = self.get_dummy_inputs(device)
         video = pipe(**inputs).frames
         generated_video = video[0]
-
         self.assertEqual(generated_video.shape, (9, 3, 16, 16))
-        expected_video = torch.randn(9, 3, 16, 16)
-        max_diff = np.abs(generated_video - expected_video).max()
-        self.assertLessEqual(max_diff, 1e10)
+
+        # fmt: off
+        expected_slice = torch.tensor([0.4525, 0.452, 0.4485, 0.4534, 0.4524, 0.4529, 0.454, 0.453, 0.5127, 0.5326, 0.5204, 0.5253, 0.5439, 0.5424, 0.5133, 0.5078])
+        # fmt: on
+
+        generated_slice = generated_video.flatten()
+        generated_slice = torch.cat([generated_slice[:8], generated_slice[-8:]])
+        self.assertTrue(torch.allclose(generated_slice, expected_slice, atol=1e-3))
 
     @unittest.skip("Test not supported")
     def test_attention_slicing_forward_pass(self):
diff --git a/tests/pipelines/wan/test_wan_image_to_video.py b/tests/pipelines/wan/test_wan_image_to_video.py
@@ -14,7 +14,6 @@
 
 import unittest
 
-import numpy as np
 import torch
 from PIL import Image
 from transformers import (
@@ -147,11 +146,15 @@ def test_inference(self):
         inputs = self.get_dummy_inputs(device)
         video = pipe(**inputs).frames
         generated_video = video[0]
-
         self.assertEqual(generated_video.shape, (9, 3, 16, 16))
-        expected_video = torch.randn(9, 3, 16, 16)
-        max_diff = np.abs(generated_video - expected_video).max()
-        self.assertLessEqual(max_diff, 1e10)
+
+        # fmt: off
+        expected_slice = torch.tensor([0.4525, 0.4525, 0.4497, 0.4536, 0.452, 0.4529, 0.454, 0.4535, 0.5072, 0.5527, 0.5165, 0.5244, 0.5481, 0.5282, 0.5208, 0.5214])
+        # fmt: on
+
+        generated_slice = generated_video.flatten()
+        generated_slice = torch.cat([generated_slice[:8], generated_slice[-8:]])
+        self.assertTrue(torch.allclose(generated_slice, expected_slice, atol=1e-3))
 
     @unittest.skip("Test not supported")
     def test_attention_slicing_forward_pass(self):
@@ -162,7 +165,25 @@ def test_inference_batch_single_identical(self):
         pass
 
 
-class WanFLFToVideoPipelineFastTests(WanImageToVideoPipelineFastTests):
+class WanFLFToVideoPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
+    pipeline_class = WanImageToVideoPipeline
+    params = TEXT_TO_IMAGE_PARAMS - {"cross_attention_kwargs", "height", "width"}
+    batch_params = TEXT_TO_IMAGE_BATCH_PARAMS
+    image_params = TEXT_TO_IMAGE_IMAGE_PARAMS
+    image_latents_params = TEXT_TO_IMAGE_IMAGE_PARAMS
+    required_optional_params = frozenset(
+        [
+            "num_inference_steps",
+            "generator",
+            "latents",
+            "return_dict",
+            "callback_on_step_end",
+            "callback_on_step_end_tensor_inputs",
+        ]
+    )
+    test_xformers_attention = False
+    supports_dduf = False
+
     def get_dummy_components(self):
         torch.manual_seed(0)
         vae = AutoencoderKLWan(
@@ -247,3 +268,32 @@ def get_dummy_inputs(self, device, seed=0):
             "output_type": "pt",
         }
         return inputs
+
+    def test_inference(self):
+        device = "cpu"
+
+        components = self.get_dummy_components()
+        pipe = self.pipeline_class(**components)
+        pipe.to(device)
+        pipe.set_progress_bar_config(disable=None)
+
+        inputs = self.get_dummy_inputs(device)
+        video = pipe(**inputs).frames
+        generated_video = video[0]
+        self.assertEqual(generated_video.shape, (9, 3, 16, 16))
+
+        # fmt: off
+        expected_slice = torch.tensor([0.4531, 0.4527, 0.4498, 0.4542, 0.4526, 0.4527, 0.4534, 0.4534, 0.5061, 0.5185, 0.5283, 0.5181, 0.5309, 0.5365, 0.5113, 0.5244])
+        # fmt: on
+
+        generated_slice = generated_video.flatten()
+        generated_slice = torch.cat([generated_slice[:8], generated_slice[-8:]])
+        self.assertTrue(torch.allclose(generated_slice, expected_slice, atol=1e-3))
+
+    @unittest.skip("Test not supported")
+    def test_attention_slicing_forward_pass(self):
+        pass
+
+    @unittest.skip("TODO: revisit failing as it requires a very high threshold to pass")
+    def test_inference_batch_single_identical(self):
+        pass
diff --git a/tests/pipelines/wan/test_wan_video_to_video.py b/tests/pipelines/wan/test_wan_video_to_video.py
