[SDXL] Make watermarker optional under certain circumstances to improve usability of SDXL 1.0 (#4346)

* improve sdxl

* more fixes

* improve sdxl

* improve sdxl

* improve sdxl

* finish

docs/source/en/api/pipelines/stable_diffusion/stable_diffusion_xl.md

@@ -38,9 +38,25 @@ You can install the libraries as follows:
 pip install transformers
 pip install accelerate
 pip install safetensors
+```
+
+### Watermarker
+
+We recommend to add an invisible watermark to images generating by Stable Diffusion XL, this can help with identifying if an image is machine-synthesised for downstream applications. To do so, please install
+the [invisible-watermark library](https://pypi.org/project/invisible-watermark/) via:
+
+```
 pip install invisible-watermark>=0.2.0
 ```
 
+If the `invisible-watermark` library is installed the watermarker will be used **by default**.
+
+If you have other provisions for generating or deploying images safely, you can disable the watermarker as follows:
+
+```py
+pipe = StableDiffusionXLPipeline.from_pretrained(..., add_watermarker=False)
+```
+
 ### Text-to-Image
 
 You can use SDXL as follows for *text-to-image*:

examples/controlnet/requirements_sdxl.txt

@@ -4,6 +4,5 @@ transformers>=4.25.1
 ftfy
 tensorboard
 Jinja2
-invisible-watermark>=0.2.0
 datasets
 wandb

examples/dreambooth/requirements_sdxl.txt

@@ -4,4 +4,3 @@ transformers>=4.25.1
 ftfy
 tensorboard
 Jinja2
-invisible-watermark>=0.2.0

src/diffusers/__init__.py

@@ -185,6 +185,11 @@ else:
         StableDiffusionPix2PixZeroPipeline,
         StableDiffusionSAGPipeline,
         StableDiffusionUpscalePipeline,
+        StableDiffusionXLControlNetPipeline,
+        StableDiffusionXLImg2ImgPipeline,
+        StableDiffusionXLInpaintPipeline,
+        StableDiffusionXLInstructPix2PixPipeline,
+        StableDiffusionXLPipeline,
         StableUnCLIPImg2ImgPipeline,
         StableUnCLIPPipeline,
         TextToVideoSDPipeline,
@@ -202,20 +207,6 @@ else:
         VQDiffusionPipeline,
     )
 
-try:
-    if not (is_torch_available() and is_transformers_available() and is_invisible_watermark_available()):
-        raise OptionalDependencyNotAvailable()
-except OptionalDependencyNotAvailable:
-    from .utils.dummy_torch_and_transformers_and_invisible_watermark_objects import *  # noqa F403
-else:
-    from .pipelines import (
-        StableDiffusionXLControlNetPipeline,
-        StableDiffusionXLImg2ImgPipeline,
-        StableDiffusionXLInpaintPipeline,
-        StableDiffusionXLInstructPix2PixPipeline,
-        StableDiffusionXLPipeline,
-    )
-
 try:
     if not (is_torch_available() and is_transformers_available() and is_k_diffusion_available()):
         raise OptionalDependencyNotAvailable()

src/diffusers/pipelines/__init__.py

@@ -1,7 +1,6 @@
 from ..utils import (
     OptionalDependencyNotAvailable,
     is_flax_available,
-    is_invisible_watermark_available,
     is_k_diffusion_available,
     is_librosa_available,
     is_note_seq_available,
@@ -51,6 +50,7 @@ else:
         StableDiffusionControlNetImg2ImgPipeline,
         StableDiffusionControlNetInpaintPipeline,
         StableDiffusionControlNetPipeline,
+        StableDiffusionXLControlNetPipeline,
     )
     from .deepfloyd_if import (
         IFImg2ImgPipeline,
@@ -108,6 +108,12 @@ else:
         StableUnCLIPPipeline,
     )
     from .stable_diffusion_safe import StableDiffusionPipelineSafe
+    from .stable_diffusion_xl import (
+        StableDiffusionXLImg2ImgPipeline,
+        StableDiffusionXLInpaintPipeline,
+        StableDiffusionXLInstructPix2PixPipeline,
+        StableDiffusionXLPipeline,
+    )
     from .t2i_adapter import StableDiffusionAdapterPipeline
     from .text_to_video_synthesis import TextToVideoSDPipeline, TextToVideoZeroPipeline, VideoToVideoSDPipeline
     from .unclip import UnCLIPImageVariationPipeline, UnCLIPPipeline
@@ -121,20 +127,6 @@ else:
     from .vq_diffusion import VQDiffusionPipeline
 
 
-try:
-    if not (is_torch_available() and is_transformers_available() and is_invisible_watermark_available()):
-        raise OptionalDependencyNotAvailable()
-except OptionalDependencyNotAvailable:
-    from ..utils.dummy_torch_and_transformers_and_invisible_watermark_objects import *  # noqa F403
-else:
-    from .controlnet import StableDiffusionXLControlNetPipeline
-    from .stable_diffusion_xl import (
-        StableDiffusionXLImg2ImgPipeline,
-        StableDiffusionXLInpaintPipeline,
-        StableDiffusionXLInstructPix2PixPipeline,
-        StableDiffusionXLPipeline,
-    )
-
 try:
     if not is_onnx_available():
         raise OptionalDependencyNotAvailable()

src/diffusers/pipelines/controlnet/__init__.py

@@ -1,21 +1,11 @@
 from ...utils import (
     OptionalDependencyNotAvailable,
     is_flax_available,
-    is_invisible_watermark_available,
     is_torch_available,
     is_transformers_available,
 )
 
 
-try:
-    if not (is_transformers_available() and is_torch_available() and is_invisible_watermark_available()):
-        raise OptionalDependencyNotAvailable()
-except OptionalDependencyNotAvailable:
-    from ...utils.dummy_torch_and_transformers_and_invisible_watermark_objects import *  # noqa F403
-else:
-    from .pipeline_controlnet_sd_xl import StableDiffusionXLControlNetPipeline
-
-
 try:
     if not (is_transformers_available() and is_torch_available()):
         raise OptionalDependencyNotAvailable()
@@ -26,6 +16,7 @@ else:
     from .pipeline_controlnet import StableDiffusionControlNetPipeline
     from .pipeline_controlnet_img2img import StableDiffusionControlNetImg2ImgPipeline
     from .pipeline_controlnet_inpaint import StableDiffusionControlNetInpaintPipeline
+    from .pipeline_controlnet_sd_xl import StableDiffusionXLControlNetPipeline
 
 
 if is_transformers_available() and is_flax_available():

src/diffusers/pipelines/controlnet/pipeline_controlnet_sd_xl.py

@@ -22,6 +22,8 @@ import torch
 import torch.nn.functional as F
 from transformers import CLIPTextModel, CLIPTextModelWithProjection, CLIPTokenizer
 
+from diffusers.utils.import_utils import is_invisible_watermark_available
+
 from ...image_processor import VaeImageProcessor
 from ...loaders import LoraLoaderMixin, TextualInversionLoaderMixin
 from ...models import AutoencoderKL, ControlNetModel, UNet2DConditionModel
@@ -42,7 +44,11 @@ from ...utils import (
 )
 from ..pipeline_utils import DiffusionPipeline
 from ..stable_diffusion_xl import StableDiffusionXLPipelineOutput
-from ..stable_diffusion_xl.watermark import StableDiffusionXLWatermarker
+
+
+if is_invisible_watermark_available():
+    from ..stable_diffusion_xl.watermark import StableDiffusionXLWatermarker
+
 from .multicontrolnet import MultiControlNetModel
 
 
@@ -109,6 +115,7 @@ class StableDiffusionXLControlNetPipeline(DiffusionPipeline, TextualInversionLoa
         controlnet: ControlNetModel,
         scheduler: KarrasDiffusionSchedulers,
         force_zeros_for_empty_prompt: bool = True,
+        add_watermarker: Optional[bool] = None,
     ):
         super().__init__()
 
@@ -130,7 +137,13 @@ class StableDiffusionXLControlNetPipeline(DiffusionPipeline, TextualInversionLoa
         self.control_image_processor = VaeImageProcessor(
             vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True, do_normalize=False
         )
-        self.watermark = StableDiffusionXLWatermarker()
+        add_watermarker = add_watermarker if add_watermarker is not None else is_invisible_watermark_available()
+
+        if add_watermarker:
+            self.watermark = StableDiffusionXLWatermarker()
+        else:
+            self.watermark = None
+
         self.register_to_config(force_zeros_for_empty_prompt=force_zeros_for_empty_prompt)
 
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_vae_slicing
@@ -995,7 +1008,10 @@ class StableDiffusionXLControlNetPipeline(DiffusionPipeline, TextualInversionLoa
             image = latents
             return StableDiffusionXLPipelineOutput(images=image)
 
-        image = self.watermark.apply_watermark(image)
+        # apply watermark if available
+        if self.watermark is not None:
+            image = self.watermark.apply_watermark(image)
+
         image = self.image_processor.postprocess(image, output_type=output_type)
 
         # Offload last model to CPU

src/diffusers/pipelines/stable_diffusion_xl/__init__.py

@@ -7,7 +7,6 @@ import PIL
 from ...utils import (
     BaseOutput,
     OptionalDependencyNotAvailable,
-    is_invisible_watermark_available,
     is_torch_available,
     is_transformers_available,
 )
@@ -28,10 +27,10 @@ class StableDiffusionXLPipelineOutput(BaseOutput):
 
 
 try:
-    if not (is_transformers_available() and is_torch_available() and is_invisible_watermark_available()):
+    if not (is_transformers_available() and is_torch_available()):
         raise OptionalDependencyNotAvailable()
 except OptionalDependencyNotAvailable:
-    from ...utils.dummy_torch_and_transformers_and_invisible_watermark_objects import *  # noqa F403
+    from ...utils.dummy_torch_and_transformers_and_objects import *  # noqa F403
 else:
     from .pipeline_stable_diffusion_xl import StableDiffusionXLPipeline
     from .pipeline_stable_diffusion_xl_img2img import StableDiffusionXLImg2ImgPipeline

src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py

@@ -32,13 +32,17 @@ from ...schedulers import KarrasDiffusionSchedulers
 from ...utils import (
     is_accelerate_available,
     is_accelerate_version,
+    is_invisible_watermark_available,
     logging,
     randn_tensor,
     replace_example_docstring,
 )
 from ..pipeline_utils import DiffusionPipeline
 from . import StableDiffusionXLPipelineOutput
-from .watermark import StableDiffusionXLWatermarker
+
+
+if is_invisible_watermark_available():
+    from .watermark import StableDiffusionXLWatermarker
 
 
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
@@ -125,6 +129,7 @@ class StableDiffusionXLPipeline(DiffusionPipeline, FromSingleFileMixin, LoraLoad
         unet: UNet2DConditionModel,
         scheduler: KarrasDiffusionSchedulers,
         force_zeros_for_empty_prompt: bool = True,
+        add_watermarker: Optional[bool] = None,
     ):
         super().__init__()
 
@@ -142,7 +147,12 @@ class StableDiffusionXLPipeline(DiffusionPipeline, FromSingleFileMixin, LoraLoad
         self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
         self.default_sample_size = self.unet.config.sample_size
 
-        self.watermark = StableDiffusionXLWatermarker()
+        add_watermarker = add_watermarker if add_watermarker is not None else is_invisible_watermark_available()
+
+        if add_watermarker:
+            self.watermark = StableDiffusionXLWatermarker()
+        else:
+            self.watermark = None
 
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_vae_slicing
     def enable_vae_slicing(self):
@@ -839,7 +849,10 @@ class StableDiffusionXLPipeline(DiffusionPipeline, FromSingleFileMixin, LoraLoad
             image = latents
             return StableDiffusionXLPipelineOutput(images=image)
 
-        image = self.watermark.apply_watermark(image)
+        # apply watermark if available
+        if self.watermark is not None:
+            image = self.watermark.apply_watermark(image)
+
         image = self.image_processor.postprocess(image, output_type=output_type)
 
         # Offload last model to CPU

src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_img2img.py

@@ -33,13 +33,17 @@ from ...schedulers import KarrasDiffusionSchedulers
 from ...utils import (
     is_accelerate_available,
     is_accelerate_version,
+    is_invisible_watermark_available,
     logging,
     randn_tensor,
     replace_example_docstring,
 )
 from ..pipeline_utils import DiffusionPipeline
 from . import StableDiffusionXLPipelineOutput
-from .watermark import StableDiffusionXLWatermarker
+
+
+if is_invisible_watermark_available():
+    from .watermark import StableDiffusionXLWatermarker
 
 
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
@@ -131,6 +135,7 @@ class StableDiffusionXLImg2ImgPipeline(DiffusionPipeline, FromSingleFileMixin, L
         scheduler: KarrasDiffusionSchedulers,
         requires_aesthetics_score: bool = False,
         force_zeros_for_empty_prompt: bool = True,
+        add_watermarker: Optional[bool] = None,
     ):
         super().__init__()
 
@@ -148,7 +153,12 @@ class StableDiffusionXLImg2ImgPipeline(DiffusionPipeline, FromSingleFileMixin, L
         self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)
         self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
 
-        self.watermark = StableDiffusionXLWatermarker()
+        add_watermarker = add_watermarker if add_watermarker is not None else is_invisible_watermark_available()
+
+        if add_watermarker:
+            self.watermark = StableDiffusionXLWatermarker()
+        else:
+            self.watermark = None
 
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_vae_slicing
     def enable_vae_slicing(self):
@@ -990,7 +1000,10 @@ class StableDiffusionXLImg2ImgPipeline(DiffusionPipeline, FromSingleFileMixin, L
             image = latents
             return StableDiffusionXLPipelineOutput(images=image)
 
-        image = self.watermark.apply_watermark(image)
+        # apply watermark if available
+        if self.watermark is not None:
+            image = self.watermark.apply_watermark(image)
+
         image = self.image_processor.postprocess(image, output_type=output_type)
 
         # Offload last model to CPU

src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_inpaint.py

@@ -30,10 +30,20 @@ from ...models.attention_processor import (
     XFormersAttnProcessor,
 )
 from ...schedulers import KarrasDiffusionSchedulers
-from ...utils import is_accelerate_available, is_accelerate_version, logging, randn_tensor, replace_example_docstring
+from ...utils import (
+    is_accelerate_available,
+    is_accelerate_version,
+    is_invisible_watermark_available,
+    logging,
+    randn_tensor,
+    replace_example_docstring,
+)
 from ..pipeline_utils import DiffusionPipeline
 from . import StableDiffusionXLPipelineOutput
-from .watermark import StableDiffusionXLWatermarker
+
+
+if is_invisible_watermark_available():
+    from .watermark import StableDiffusionXLWatermarker
 
 
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
@@ -265,6 +275,7 @@ class StableDiffusionXLInpaintPipeline(
         scheduler: KarrasDiffusionSchedulers,
         requires_aesthetics_score: bool = False,
         force_zeros_for_empty_prompt: bool = True,
+        add_watermarker: Optional[bool] = None,
     ):
         super().__init__()
 
@@ -282,7 +293,12 @@ class StableDiffusionXLInpaintPipeline(
         self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)
         self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
 
-        self.watermark = StableDiffusionXLWatermarker()
+        add_watermarker = add_watermarker if add_watermarker is not None else is_invisible_watermark_available()
+
+        if add_watermarker:
+            self.watermark = StableDiffusionXLWatermarker()
+        else:
+            self.watermark = None
 
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_vae_slicing
     def enable_vae_slicing(self):
@@ -1266,6 +1282,10 @@ class StableDiffusionXLInpaintPipeline(
         else:
             return StableDiffusionXLPipelineOutput(images=latents)
 
+        # apply watermark if available
+        if self.watermark is not None:
+            image = self.watermark.apply_watermark(image)
+
         image = self.image_processor.postprocess(image, output_type=output_type)
 
         # Offload last model to CPU

src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_instruct_pix2pix.py

@@ -34,12 +34,16 @@ from ...utils import (
     deprecate,
     is_accelerate_available,
     is_accelerate_version,
+    is_invisible_watermark_available,
     logging,
     randn_tensor,
 )
 from ..pipeline_utils import DiffusionPipeline
 from . import StableDiffusionXLPipelineOutput
-from .watermark import StableDiffusionXLWatermarker
+
+
+if is_invisible_watermark_available():
+    from .watermark import StableDiffusionXLWatermarker
 
 
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
@@ -109,6 +113,7 @@ class StableDiffusionXLInstructPix2PixPipeline(DiffusionPipeline, FromSingleFile
         scheduler: KarrasDiffusionSchedulers,
         requires_aesthetics_score: bool = False,
         force_zeros_for_empty_prompt: bool = True,
+        add_watermarker: Optional[bool] = None,
     ):
         super().__init__()
 
@@ -128,7 +133,12 @@ class StableDiffusionXLInstructPix2PixPipeline(DiffusionPipeline, FromSingleFile
 
         self.vae.config.force_upcast = True  # force the VAE to be in float32 mode, as it overflows in float16
 
-        self.watermark = StableDiffusionXLWatermarker()
+        add_watermarker = add_watermarker if add_watermarker is not None else is_invisible_watermark_available()
+
+        if add_watermarker:
+            self.watermark = StableDiffusionXLWatermarker()
+        else:
+            self.watermark = None
 
     def enable_vae_slicing(self):
         r"""
@@ -908,7 +918,10 @@ class StableDiffusionXLInstructPix2PixPipeline(DiffusionPipeline, FromSingleFile
             image = latents
             return StableDiffusionXLPipelineOutput(images=image)
 
-        image = self.watermark.apply_watermark(image)
+        # apply watermark if available
+        if self.watermark is not None:
+            image = self.watermark.apply_watermark(image)
+
         image = self.image_processor.postprocess(image, output_type=output_type)
 
         # Offload last model to CPU

src/diffusers/pipelines/stable_diffusion_xl/watermark.py

@@ -1,6 +1,11 @@
 import numpy as np
 import torch
-from imwatermark import WatermarkEncoder
+
+from ...utils import is_invisible_watermark_available
+
+
+if is_invisible_watermark_available():
+    from imwatermark import WatermarkEncoder
 
 
 # Copied from https://github.com/Stability-AI/generative-models/blob/613af104c6b85184091d42d374fef420eddb356d/scripts/demo/streamlit_helpers.py#L66

src/diffusers/utils/dummy_torch_and_transformers_and_invisible_watermark_objects.py

@@ -1,77 +0,0 @@
-# This file is autogenerated by the command `make fix-copies`, do not edit.
-from ..utils import DummyObject, requires_backends
-
-
-class StableDiffusionXLControlNetPipeline(metaclass=DummyObject):
-    _backends = ["torch", "transformers", "invisible_watermark"]
-
-    def __init__(self, *args, **kwargs):
-        requires_backends(self, ["torch", "transformers", "invisible_watermark"])
-
-    @classmethod
-    def from_config(cls, *args, **kwargs):
-        requires_backends(cls, ["torch", "transformers", "invisible_watermark"])
-
-    @classmethod
-    def from_pretrained(cls, *args, **kwargs):
-        requires_backends(cls, ["torch", "transformers", "invisible_watermark"])
-
-
-class StableDiffusionXLImg2ImgPipeline(metaclass=DummyObject):
-    _backends = ["torch", "transformers", "invisible_watermark"]
-
-    def __init__(self, *args, **kwargs):
-        requires_backends(self, ["torch", "transformers", "invisible_watermark"])
-
-    @classmethod
-    def from_config(cls, *args, **kwargs):
-        requires_backends(cls, ["torch", "transformers", "invisible_watermark"])
-
-    @classmethod
-    def from_pretrained(cls, *args, **kwargs):
-        requires_backends(cls, ["torch", "transformers", "invisible_watermark"])
-
-
-class StableDiffusionXLInpaintPipeline(metaclass=DummyObject):
-    _backends = ["torch", "transformers", "invisible_watermark"]
-
-    def __init__(self, *args, **kwargs):
-        requires_backends(self, ["torch", "transformers", "invisible_watermark"])
-
-    @classmethod
-    def from_config(cls, *args, **kwargs):
-        requires_backends(cls, ["torch", "transformers", "invisible_watermark"])
-
-    @classmethod
-    def from_pretrained(cls, *args, **kwargs):
-        requires_backends(cls, ["torch", "transformers", "invisible_watermark"])
-
-
-class StableDiffusionXLInstructPix2PixPipeline(metaclass=DummyObject):
-    _backends = ["torch", "transformers", "invisible_watermark"]
-
-    def __init__(self, *args, **kwargs):
-        requires_backends(self, ["torch", "transformers", "invisible_watermark"])
-
-    @classmethod
-    def from_config(cls, *args, **kwargs):
-        requires_backends(cls, ["torch", "transformers", "invisible_watermark"])
-
-    @classmethod
-    def from_pretrained(cls, *args, **kwargs):
-        requires_backends(cls, ["torch", "transformers", "invisible_watermark"])
-
-
-class StableDiffusionXLPipeline(metaclass=DummyObject):
-    _backends = ["torch", "transformers", "invisible_watermark"]
-
-    def __init__(self, *args, **kwargs):
-        requires_backends(self, ["torch", "transformers", "invisible_watermark"])
-
-    @classmethod
-    def from_config(cls, *args, **kwargs):
-        requires_backends(cls, ["torch", "transformers", "invisible_watermark"])
-
-    @classmethod
-    def from_pretrained(cls, *args, **kwargs):
-        requires_backends(cls, ["torch", "transformers", "invisible_watermark"])

src/diffusers/utils/dummy_torch_and_transformers_objects.py

@@ -827,6 +827,81 @@ class StableDiffusionUpscalePipeline(metaclass=DummyObject):
         requires_backends(cls, ["torch", "transformers"])
 
 
+class StableDiffusionXLControlNetPipeline(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 StableDiffusionXLImg2ImgPipeline(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 StableDiffusionXLInpaintPipeline(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 StableDiffusionXLInstructPix2PixPipeline(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 StableDiffusionXLPipeline(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 StableUnCLIPImg2ImgPipeline(metaclass=DummyObject):
     _backends = ["torch", "transformers"]