Add QwenImage Inpainting and Img2Img pipeline (#12117)

* feat/qwenimage-img2img-inpaint

* Update qwenimage.md to reflect new pipelines and add # Copied from convention

* tiny fix for passing ruff check

* reformat code

* fix copied from statement

* fix copied from statement

* copy and style fix

* fix dummies

---------
Co-authored-by: TuanNT-ZenAI <tuannt.zenai@gmail.com>
Co-authored-by: DN6 <dhruv.nair@gmail.com>

docs/source/en/api/pipelines/qwenimage.md

@@ -90,3 +90,15 @@ image.save("qwen_fewsteps.png")
 ## QwenImagePipelineOutput
 
 [[autodoc]] pipelines.qwenimage.pipeline_output.QwenImagePipelineOutput
+
+## QwenImageImg2ImgPipeline
+
+[[autodoc]] QwenImageImg2ImgPipeline
+  - all
+  - __call__
+
+## QwenImageInpaintPipeline
+
+[[autodoc]] QwenImageInpaintPipeline
+  - all
+  - __call__

src/diffusers/__init__.py

@@ -489,6 +489,8 @@ else:
             "PixArtAlphaPipeline",
             "PixArtSigmaPAGPipeline",
             "PixArtSigmaPipeline",
+            "QwenImageImg2ImgPipeline",
+            "QwenImageInpaintPipeline",
             "QwenImagePipeline",
             "ReduxImageEncoder",
             "SanaControlNetPipeline",
@@ -1121,6 +1123,8 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
             PixArtAlphaPipeline,
             PixArtSigmaPAGPipeline,
             PixArtSigmaPipeline,
+            QwenImageImg2ImgPipeline,
+            QwenImageInpaintPipeline,
             QwenImagePipeline,
             ReduxImageEncoder,
             SanaControlNetPipeline,

src/diffusers/pipelines/__init__.py

@@ -387,7 +387,11 @@ else:
         "SkyReelsV2ImageToVideoPipeline",
         "SkyReelsV2Pipeline",
     ]
-    _import_structure["qwenimage"] = ["QwenImagePipeline"]
+    _import_structure["qwenimage"] = [
+        "QwenImagePipeline",
+        "QwenImageImg2ImgPipeline",
+        "QwenImageInpaintPipeline",
+    ]
 try:
     if not is_onnx_available():
         raise OptionalDependencyNotAvailable()
@@ -704,7 +708,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
         from .paint_by_example import PaintByExamplePipeline
         from .pia import PIAPipeline
         from .pixart_alpha import PixArtAlphaPipeline, PixArtSigmaPipeline
-        from .qwenimage import QwenImagePipeline
+        from .qwenimage import QwenImageImg2ImgPipeline, QwenImageInpaintPipeline, QwenImagePipeline
         from .sana import SanaControlNetPipeline, SanaPipeline, SanaSprintImg2ImgPipeline, SanaSprintPipeline
         from .semantic_stable_diffusion import SemanticStableDiffusionPipeline
         from .shap_e import ShapEImg2ImgPipeline, ShapEPipeline

src/diffusers/pipelines/qwenimage/__init__.py

@@ -24,6 +24,8 @@ except OptionalDependencyNotAvailable:
 else:
     _import_structure["modeling_qwenimage"] = ["ReduxImageEncoder"]
     _import_structure["pipeline_qwenimage"] = ["QwenImagePipeline"]
+    _import_structure["pipeline_qwenimage_img2img"] = ["QwenImageImg2ImgPipeline"]
+    _import_structure["pipeline_qwenimage_inpaint"] = ["QwenImageInpaintPipeline"]
 
 if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
     try:
@@ -33,6 +35,8 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
         from ...utils.dummy_torch_and_transformers_objects import *  # noqa F403
     else:
         from .pipeline_qwenimage import QwenImagePipeline
+        from .pipeline_qwenimage_img2img import QwenImageImg2ImgPipeline
+        from .pipeline_qwenimage_inpaint import QwenImageInpaintPipeline
 else:
     import sys
 

src/diffusers/utils/dummy_torch_and_transformers_objects.py

@@ -1742,6 +1742,36 @@ class PixArtSigmaPipeline(metaclass=DummyObject):
         requires_backends(cls, ["torch", "transformers"])
 
 
+class QwenImageImg2ImgPipeline(metaclass=DummyObject):
+    _backends = ["torch", "transformers"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch", "transformers"])
+
+    @classmethod
+    def from_config(cls, *args, **kwargs):
+        requires_backends(cls, ["torch", "transformers"])
+
+    @classmethod
+    def from_pretrained(cls, *args, **kwargs):
+        requires_backends(cls, ["torch", "transformers"])
+
+
+class QwenImageInpaintPipeline(metaclass=DummyObject):
+    _backends = ["torch", "transformers"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch", "transformers"])
+
+    @classmethod
+    def from_config(cls, *args, **kwargs):
+        requires_backends(cls, ["torch", "transformers"])
+
+    @classmethod
+    def from_pretrained(cls, *args, **kwargs):
+        requires_backends(cls, ["torch", "transformers"])
+
+
 class QwenImagePipeline(metaclass=DummyObject):
     _backends = ["torch", "transformers"]
 

tests/pipelines/qwenimage/test_qwenimage_img2img.py

+import random
+import unittest
+
+import numpy as np
+import torch
+from transformers import Qwen2_5_VLConfig, Qwen2_5_VLForConditionalGeneration, Qwen2Tokenizer
+
+from diffusers import (
+    AutoencoderKLQwenImage,
+    FlowMatchEulerDiscreteScheduler,
+    QwenImageImg2ImgPipeline,
+    QwenImageTransformer2DModel,
+)
+from diffusers.utils.testing_utils import (
+    enable_full_determinism,
+    floats_tensor,
+    torch_device,
+)
+
+from ..test_pipelines_common import PipelineTesterMixin, to_np
+
+
+enable_full_determinism()
+
+
+class QwenImageImg2ImgPipelineFastTests(unittest.TestCase, PipelineTesterMixin):
+    pipeline_class = QwenImageImg2ImgPipeline
+    params = frozenset(["prompt", "image", "height", "width", "guidance_scale", "true_cfg_scale", "strength"])
+    batch_params = frozenset(["prompt", "image"])
+    image_params = frozenset(["image"])
+    image_latents_params = frozenset(["latents"])
+    required_optional_params = frozenset(
+        [
+            "num_inference_steps",
+            "generator",
+            "latents",
+            "return_dict",
+            "callback_on_step_end",
+            "callback_on_step_end_tensor_inputs",
+        ]
+    )
+    supports_dduf = False
+    test_xformers_attention = False
+    test_attention_slicing = True
+    test_layerwise_casting = True
+    test_group_offloading = True
+
+    def get_dummy_components(self):
+        torch.manual_seed(0)
+        transformer = QwenImageTransformer2DModel(
+            patch_size=2,
+            in_channels=16,
+            out_channels=4,
+            num_layers=2,
+            attention_head_dim=16,
+            num_attention_heads=3,
+            joint_attention_dim=16,
+            guidance_embeds=False,
+            axes_dims_rope=(8, 4, 4),
+        )
+
+        torch.manual_seed(0)
+        z_dim = 4
+        vae = AutoencoderKLQwenImage(
+            base_dim=z_dim * 6,
+            z_dim=z_dim,
+            dim_mult=[1, 2, 4],
+            num_res_blocks=1,
+            temperal_downsample=[False, True],
+            latents_mean=[0.0] * 4,
+            latents_std=[1.0] * 4,
+        )
+
+        torch.manual_seed(0)
+        scheduler = FlowMatchEulerDiscreteScheduler()
+
+        torch.manual_seed(0)
+        config = Qwen2_5_VLConfig(
+            text_config={
+                "hidden_size": 16,
+                "intermediate_size": 16,
+                "num_hidden_layers": 2,
+                "num_attention_heads": 2,
+                "num_key_value_heads": 2,
+                "rope_scaling": {
+                    "mrope_section": [1, 1, 2],
+                    "rope_type": "default",
+                    "type": "default",
+                },
+                "rope_theta": 1000000.0,
+            },
+            vision_config={
+                "depth": 2,
+                "hidden_size": 16,
+                "intermediate_size": 16,
+                "num_heads": 2,
+                "out_hidden_size": 16,
+            },
+            hidden_size=16,
+            vocab_size=152064,
+            vision_end_token_id=151653,
+            vision_start_token_id=151652,
+            vision_token_id=151654,
+        )
+        text_encoder = Qwen2_5_VLForConditionalGeneration(config)
+        tokenizer = Qwen2Tokenizer.from_pretrained("hf-internal-testing/tiny-random-Qwen2VLForConditionalGeneration")
+
+        return {
+            "transformer": transformer,
+            "vae": vae,
+            "scheduler": scheduler,
+            "text_encoder": text_encoder,
+            "tokenizer": tokenizer,
+        }
+
+    def get_dummy_inputs(self, device, seed=0):
+        image = floats_tensor((1, 3, 32, 32), rng=random.Random(seed)).to(device)
+        if str(device).startswith("mps"):
+            generator = torch.manual_seed(seed)
+        else:
+            generator = torch.Generator(device="cpu").manual_seed(seed)
+
+        inputs = {
+            "image": image,
+            "prompt": "dance monkey",
+            "negative_prompt": "bad quality",
+            "generator": generator,
+            "num_inference_steps": 2,
+            "guidance_scale": 3.0,
+            "true_cfg_scale": 1.0,
+            "height": 32,
+            "width": 32,
+            "max_sequence_length": 16,
+            "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)
+        image = pipe(**inputs).images
+        generated_image = image[0]
+        self.assertEqual(generated_image.shape, (3, 32, 32))
+
+    def test_inference_batch_single_identical(self):
+        self._test_inference_batch_single_identical(batch_size=3, expected_max_diff=1e-1)
+
+    def test_attention_slicing_forward_pass(
+        self, test_max_difference=True, test_mean_pixel_difference=True, expected_max_diff=1e-3
+    ):
+        if not self.test_attention_slicing:
+            return
+
+        components = self.get_dummy_components()
+        pipe = self.pipeline_class(**components)
+        for component in pipe.components.values():
+            if hasattr(component, "set_default_attn_processor"):
+                component.set_default_attn_processor()
+        pipe.to(torch_device)
+        pipe.set_progress_bar_config(disable=None)
+
+        generator_device = "cpu"
+        inputs = self.get_dummy_inputs(generator_device)
+        output_without_slicing = pipe(**inputs).images[0]
+
+        pipe.enable_attention_slicing(slice_size=1)
+        inputs = self.get_dummy_inputs(generator_device)
+        output_with_slicing1 = pipe(**inputs).images[0]
+
+        pipe.enable_attention_slicing(slice_size=2)
+        inputs = self.get_dummy_inputs(generator_device)
+        output_with_slicing2 = pipe(**inputs).images[0]
+
+        if test_max_difference:
+            max_diff1 = np.abs(to_np(output_with_slicing1) - to_np(output_without_slicing)).max()
+            max_diff2 = np.abs(to_np(output_with_slicing2) - to_np(output_without_slicing)).max()
+            self.assertLess(
+                max(max_diff1, max_diff2),
+                expected_max_diff,
+                "Attention slicing should not affect the inference results",
+            )
+
+    def test_vae_tiling(self, expected_diff_max: float = 0.2):
+        generator_device = "cpu"
+        components = self.get_dummy_components()
+
+        pipe = self.pipeline_class(**components)
+        pipe.to("cpu")
+        pipe.set_progress_bar_config(disable=None)
+
+        # Without tiling
+        inputs = self.get_dummy_inputs(generator_device)
+        inputs["height"] = inputs["width"] = 128
+        output_without_tiling = pipe(**inputs)[0]
+
+        # With tiling
+        pipe.vae.enable_tiling(
+            tile_sample_min_height=96,
+            tile_sample_min_width=96,
+            tile_sample_stride_height=64,
+            tile_sample_stride_width=64,
+        )
+        inputs = self.get_dummy_inputs(generator_device)
+        inputs["height"] = inputs["width"] = 128
+        output_with_tiling = pipe(**inputs)[0]
+
+        self.assertLess(
+            (to_np(output_without_tiling) - to_np(output_with_tiling)).max(),
+            expected_diff_max,
+            "VAE tiling should not affect the inference results",
+        )

tests/pipelines/qwenimage/test_qwenimage_inpaint.py

+# Copyright 2025 The HuggingFace Team.
+#
+# 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.
+
+import random
+import unittest
+
+import numpy as np
+import torch
+from transformers import Qwen2_5_VLConfig, Qwen2_5_VLForConditionalGeneration, Qwen2Tokenizer
+
+from diffusers import (
+    AutoencoderKLQwenImage,
+    FlowMatchEulerDiscreteScheduler,
+    QwenImageInpaintPipeline,
+    QwenImageTransformer2DModel,
+)
+from diffusers.utils.testing_utils import enable_full_determinism, floats_tensor, torch_device
+
+from ..pipeline_params import TEXT_TO_IMAGE_BATCH_PARAMS, TEXT_TO_IMAGE_IMAGE_PARAMS, TEXT_TO_IMAGE_PARAMS
+from ..test_pipelines_common import PipelineTesterMixin, to_np
+
+
+enable_full_determinism()
+
+
+class QwenImageInpaintPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
+    pipeline_class = QwenImageInpaintPipeline
+    params = TEXT_TO_IMAGE_PARAMS - {"cross_attention_kwargs"}
+    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",
+        ]
+    )
+    supports_dduf = False
+    test_xformers_attention = False
+    test_layerwise_casting = True
+    test_group_offloading = True
+
+    def get_dummy_components(self):
+        torch.manual_seed(0)
+        transformer = QwenImageTransformer2DModel(
+            patch_size=2,
+            in_channels=16,
+            out_channels=4,
+            num_layers=2,
+            attention_head_dim=16,
+            num_attention_heads=3,
+            joint_attention_dim=16,
+            guidance_embeds=False,
+            axes_dims_rope=(8, 4, 4),
+        )
+
+        torch.manual_seed(0)
+        z_dim = 4
+        vae = AutoencoderKLQwenImage(
+            base_dim=z_dim * 6,
+            z_dim=z_dim,
+            dim_mult=[1, 2, 4],
+            num_res_blocks=1,
+            temperal_downsample=[False, True],
+            # fmt: off
+            latents_mean=[0.0] * 4,
+            latents_std=[1.0] * 4,
+            # fmt: on
+        )
+
+        torch.manual_seed(0)
+        scheduler = FlowMatchEulerDiscreteScheduler()
+
+        torch.manual_seed(0)
+        config = Qwen2_5_VLConfig(
+            text_config={
+                "hidden_size": 16,
+                "intermediate_size": 16,
+                "num_hidden_layers": 2,
+                "num_attention_heads": 2,
+                "num_key_value_heads": 2,
+                "rope_scaling": {
+                    "mrope_section": [1, 1, 2],
+                    "rope_type": "default",
+                    "type": "default",
+                },
+                "rope_theta": 1000000.0,
+            },
+            vision_config={
+                "depth": 2,
+                "hidden_size": 16,
+                "intermediate_size": 16,
+                "num_heads": 2,
+                "out_hidden_size": 16,
+            },
+            hidden_size=16,
+            vocab_size=152064,
+            vision_end_token_id=151653,
+            vision_start_token_id=151652,
+            vision_token_id=151654,
+        )
+        text_encoder = Qwen2_5_VLForConditionalGeneration(config)
+        tokenizer = Qwen2Tokenizer.from_pretrained("hf-internal-testing/tiny-random-Qwen2VLForConditionalGeneration")
+
+        components = {
+            "transformer": transformer,
+            "vae": vae,
+            "scheduler": scheduler,
+            "text_encoder": text_encoder,
+            "tokenizer": tokenizer,
+        }
+        return components
+
+    def get_dummy_inputs(self, device, seed=0):
+        image = floats_tensor((1, 3, 32, 32), rng=random.Random(seed)).to(device)
+        mask_image = torch.ones((1, 1, 32, 32)).to(device)
+        if str(device).startswith("mps"):
+            generator = torch.manual_seed(seed)
+        else:
+            generator = torch.Generator(device=device).manual_seed(seed)
+
+        inputs = {
+            "prompt": "dance monkey",
+            "negative_prompt": "bad quality",
+            "image": image,
+            "mask_image": mask_image,
+            "generator": generator,
+            "num_inference_steps": 2,
+            "guidance_scale": 3.0,
+            "true_cfg_scale": 1.0,
+            "height": 32,
+            "width": 32,
+            "max_sequence_length": 16,
+            "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)
+        image = pipe(**inputs).images
+        generated_image = image[0]
+        self.assertEqual(generated_image.shape, (3, 32, 32))
+
+    def test_inference_batch_single_identical(self):
+        self._test_inference_batch_single_identical(batch_size=3, expected_max_diff=1e-1)
+
+    def test_attention_slicing_forward_pass(
+        self, test_max_difference=True, test_mean_pixel_difference=True, expected_max_diff=1e-3
+    ):
+        if not self.test_attention_slicing:
+            return
+
+        components = self.get_dummy_components()
+        pipe = self.pipeline_class(**components)
+        for component in pipe.components.values():
+            if hasattr(component, "set_default_attn_processor"):
+                component.set_default_attn_processor()
+        pipe.to(torch_device)
+        pipe.set_progress_bar_config(disable=None)
+
+        generator_device = "cpu"
+        inputs = self.get_dummy_inputs(generator_device)
+        output_without_slicing = pipe(**inputs)[0]
+
+        pipe.enable_attention_slicing(slice_size=1)
+        inputs = self.get_dummy_inputs(generator_device)
+        output_with_slicing1 = pipe(**inputs)[0]
+
+        pipe.enable_attention_slicing(slice_size=2)
+        inputs = self.get_dummy_inputs(generator_device)
+        output_with_slicing2 = pipe(**inputs)[0]
+
+        if test_max_difference:
+            max_diff1 = np.abs(to_np(output_with_slicing1) - to_np(output_without_slicing)).max()
+            max_diff2 = np.abs(to_np(output_with_slicing2) - to_np(output_without_slicing)).max()
+            self.assertLess(
+                max(max_diff1, max_diff2),
+                expected_max_diff,
+                "Attention slicing should not affect the inference results",
+            )
+
+    def test_vae_tiling(self, expected_diff_max: float = 0.2):
+        generator_device = "cpu"
+        components = self.get_dummy_components()
+
+        pipe = self.pipeline_class(**components)
+        pipe.to("cpu")
+        pipe.set_progress_bar_config(disable=None)
+
+        # Without tiling
+        inputs = self.get_dummy_inputs(generator_device)
+        inputs["height"] = inputs["width"] = 128
+        output_without_tiling = pipe(**inputs)[0]
+
+        # With tiling
+        pipe.vae.enable_tiling(
+            tile_sample_min_height=96,
+            tile_sample_min_width=96,
+            tile_sample_stride_height=64,
+            tile_sample_stride_width=64,
+        )
+        inputs = self.get_dummy_inputs(generator_device)
+        inputs["height"] = inputs["width"] = 128
+        output_with_tiling = pipe(**inputs)[0]
+
+        self.assertLess(
+            (to_np(output_without_tiling) - to_np(output_with_tiling)).max(),
+            expected_diff_max,
+            "VAE tiling should not affect the inference results",
+        )