diffusers 0.24.0__py3-none-any.whl → 0.25.1__py3-none-any.whl

diffusers/pipelines/controlnet/pipeline_controlnet.py

@@ -24,7 +24,7 @@ from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer, CLIPV
 
 from ...image_processor import PipelineImageInput, VaeImageProcessor
 from ...loaders import FromSingleFileMixin, IPAdapterMixin, LoraLoaderMixin, TextualInversionLoaderMixin
-from ...models import AutoencoderKL, ControlNetModel, UNet2DConditionModel
+from ...models import AutoencoderKL, ControlNetModel, ImageProjection, UNet2DConditionModel
 from ...models.lora import adjust_lora_scale_text_encoder
 from ...schedulers import KarrasDiffusionSchedulers
 from ...utils import (
@@ -147,6 +147,9 @@ class StableDiffusionControlNetPipeline(
 
     The pipeline also inherits the following loading methods:
         - [`~loaders.TextualInversionLoaderMixin.load_textual_inversion`] for loading textual inversion embeddings
+        - [`~loaders.LoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+        - [`~loaders.LoraLoaderMixin.save_lora_weights`] for saving LoRA weights
+        - [`~loaders.FromSingleFileMixin.from_single_file`] for loading `.ckpt` files
         - [`~loaders.IPAdapterMixin.load_ip_adapter`] for loading IP Adapters
 
     Args:
@@ -173,7 +176,7 @@ class StableDiffusionControlNetPipeline(
             A `CLIPImageProcessor` to extract features from generated images; used as inputs to the `safety_checker`.
     """
 
-    model_cpu_offload_seq = "text_encoder->unet->vae"
+    model_cpu_offload_seq = "text_encoder->image_encoder->unet->vae"
     _optional_components = ["safety_checker", "feature_extractor", "image_encoder"]
     _exclude_from_cpu_offload = ["safety_checker"]
     _callback_tensor_inputs = ["latents", "prompt_embeds", "negative_prompt_embeds"]
@@ -479,18 +482,29 @@ class StableDiffusionControlNetPipeline(
         return prompt_embeds, negative_prompt_embeds
 
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.encode_image
-    def encode_image(self, image, device, num_images_per_prompt):
+    def encode_image(self, image, device, num_images_per_prompt, output_hidden_states=None):
         dtype = next(self.image_encoder.parameters()).dtype
 
         if not isinstance(image, torch.Tensor):
             image = self.feature_extractor(image, return_tensors="pt").pixel_values
 
         image = image.to(device=device, dtype=dtype)
-        image_embeds = self.image_encoder(image).image_embeds
-        image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0)
+        if output_hidden_states:
+            image_enc_hidden_states = self.image_encoder(image, output_hidden_states=True).hidden_states[-2]
+            image_enc_hidden_states = image_enc_hidden_states.repeat_interleave(num_images_per_prompt, dim=0)
+            uncond_image_enc_hidden_states = self.image_encoder(
+                torch.zeros_like(image), output_hidden_states=True
+            ).hidden_states[-2]
+            uncond_image_enc_hidden_states = uncond_image_enc_hidden_states.repeat_interleave(
+                num_images_per_prompt, dim=0
+            )
+            return image_enc_hidden_states, uncond_image_enc_hidden_states
+        else:
+            image_embeds = self.image_encoder(image).image_embeds
+            image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0)
+            uncond_image_embeds = torch.zeros_like(image_embeds)
 
-        uncond_image_embeds = torch.zeros_like(image_embeds)
-        return image_embeds, uncond_image_embeds
+            return image_embeds, uncond_image_embeds
 
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.run_safety_checker
     def run_safety_checker(self, image, device, dtype):
@@ -619,7 +633,7 @@ class StableDiffusionControlNetPipeline(
             # When `image` is a nested list:
             # (e.g. [[canny_image_1, pose_image_1], [canny_image_2, pose_image_2]])
             elif any(isinstance(i, list) for i in image):
-                raise ValueError("A single batch of multiple conditionings are supported at the moment.")
+                raise ValueError("A single batch of multiple conditionings is not supported at the moment.")
             elif len(image) != len(self.controlnet.nets):
                 raise ValueError(
                     f"For multiple controlnets: `image` must have the same length as the number of controlnets, but got {len(image)} images and {len(self.controlnet.nets)} ControlNets."
@@ -645,7 +659,7 @@ class StableDiffusionControlNetPipeline(
         ):
             if isinstance(controlnet_conditioning_scale, list):
                 if any(isinstance(i, list) for i in controlnet_conditioning_scale):
-                    raise ValueError("A single batch of multiple conditionings are supported at the moment.")
+                    raise ValueError("A single batch of multiple conditionings is not supported at the moment.")
             elif isinstance(controlnet_conditioning_scale, list) and len(controlnet_conditioning_scale) != len(
                 self.controlnet.nets
             ):
@@ -1067,7 +1081,10 @@ class StableDiffusionControlNetPipeline(
             prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds])
 
         if ip_adapter_image is not None:
-            image_embeds, negative_image_embeds = self.encode_image(ip_adapter_image, device, num_images_per_prompt)
+            output_hidden_state = False if isinstance(self.unet.encoder_hid_proj, ImageProjection) else True
+            image_embeds, negative_image_embeds = self.encode_image(
+                ip_adapter_image, device, num_images_per_prompt, output_hidden_state
+            )
             if self.do_classifier_free_guidance:
                 image_embeds = torch.cat([negative_image_embeds, image_embeds])
 

diffusers/pipelines/controlnet/pipeline_controlnet_img2img.py

@@ -19,10 +19,10 @@ import numpy as np
 import PIL.Image
 import torch
 import torch.nn.functional as F
-from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer
+from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer, CLIPVisionModelWithProjection
 
 from ...image_processor import PipelineImageInput, VaeImageProcessor
-from ...loaders import FromSingleFileMixin, LoraLoaderMixin, TextualInversionLoaderMixin
+from ...loaders import FromSingleFileMixin, IPAdapterMixin, LoraLoaderMixin, TextualInversionLoaderMixin
 from ...models import AutoencoderKL, ControlNetModel, UNet2DConditionModel
 from ...models.lora import adjust_lora_scale_text_encoder
 from ...schedulers import KarrasDiffusionSchedulers
@@ -130,7 +130,7 @@ def prepare_image(image):
 
 
 class StableDiffusionControlNetImg2ImgPipeline(
-    DiffusionPipeline, TextualInversionLoaderMixin, LoraLoaderMixin, FromSingleFileMixin
+    DiffusionPipeline, TextualInversionLoaderMixin, LoraLoaderMixin, IPAdapterMixin, FromSingleFileMixin
 ):
     r"""
     Pipeline for image-to-image generation using Stable Diffusion with ControlNet guidance.
@@ -140,6 +140,10 @@ class StableDiffusionControlNetImg2ImgPipeline(
 
     The pipeline also inherits the following loading methods:
         - [`~loaders.TextualInversionLoaderMixin.load_textual_inversion`] for loading textual inversion embeddings
+        - [`~loaders.LoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+        - [`~loaders.LoraLoaderMixin.save_lora_weights`] for saving LoRA weights
+        - [`~loaders.FromSingleFileMixin.from_single_file`] for loading `.ckpt` files
+        - [`~loaders.IPAdapterMixin.load_ip_adapter`] for loading IP Adapters
 
     Args:
         vae ([`AutoencoderKL`]):
@@ -166,7 +170,7 @@ class StableDiffusionControlNetImg2ImgPipeline(
     """
 
     model_cpu_offload_seq = "text_encoder->unet->vae"
-    _optional_components = ["safety_checker", "feature_extractor"]
+    _optional_components = ["safety_checker", "feature_extractor", "image_encoder"]
     _exclude_from_cpu_offload = ["safety_checker"]
     _callback_tensor_inputs = ["latents", "prompt_embeds", "negative_prompt_embeds"]
 
@@ -180,6 +184,7 @@ class StableDiffusionControlNetImg2ImgPipeline(
         scheduler: KarrasDiffusionSchedulers,
         safety_checker: StableDiffusionSafetyChecker,
         feature_extractor: CLIPImageProcessor,
+        image_encoder: CLIPVisionModelWithProjection = None,
         requires_safety_checker: bool = True,
     ):
         super().__init__()
@@ -212,6 +217,7 @@ class StableDiffusionControlNetImg2ImgPipeline(
             scheduler=scheduler,
             safety_checker=safety_checker,
             feature_extractor=feature_extractor,
+            image_encoder=image_encoder,
         )
         self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)
         self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True)
@@ -468,6 +474,31 @@ class StableDiffusionControlNetImg2ImgPipeline(
 
         return prompt_embeds, negative_prompt_embeds
 
+    # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.encode_image
+    def encode_image(self, image, device, num_images_per_prompt, output_hidden_states=None):
+        dtype = next(self.image_encoder.parameters()).dtype
+
+        if not isinstance(image, torch.Tensor):
+            image = self.feature_extractor(image, return_tensors="pt").pixel_values
+
+        image = image.to(device=device, dtype=dtype)
+        if output_hidden_states:
+            image_enc_hidden_states = self.image_encoder(image, output_hidden_states=True).hidden_states[-2]
+            image_enc_hidden_states = image_enc_hidden_states.repeat_interleave(num_images_per_prompt, dim=0)
+            uncond_image_enc_hidden_states = self.image_encoder(
+                torch.zeros_like(image), output_hidden_states=True
+            ).hidden_states[-2]
+            uncond_image_enc_hidden_states = uncond_image_enc_hidden_states.repeat_interleave(
+                num_images_per_prompt, dim=0
+            )
+            return image_enc_hidden_states, uncond_image_enc_hidden_states
+        else:
+            image_embeds = self.image_encoder(image).image_embeds
+            image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0)
+            uncond_image_embeds = torch.zeros_like(image_embeds)
+
+            return image_embeds, uncond_image_embeds
+
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.run_safety_checker
     def run_safety_checker(self, image, device, dtype):
         if self.safety_checker is None:
@@ -861,6 +892,7 @@ class StableDiffusionControlNetImg2ImgPipeline(
         latents: Optional[torch.FloatTensor] = None,
         prompt_embeds: Optional[torch.FloatTensor] = None,
         negative_prompt_embeds: Optional[torch.FloatTensor] = None,
+        ip_adapter_image: Optional[PipelineImageInput] = None,
         output_type: Optional[str] = "pil",
         return_dict: bool = True,
         cross_attention_kwargs: Optional[Dict[str, Any]] = None,
@@ -922,6 +954,7 @@ class StableDiffusionControlNetImg2ImgPipeline(
             negative_prompt_embeds (`torch.FloatTensor`, *optional*):
                 Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If
                 not provided, `negative_prompt_embeds` are generated from the `negative_prompt` input argument.
+            ip_adapter_image: (`PipelineImageInput`, *optional*): Optional image input to work with IP Adapters.
             output_type (`str`, *optional*, defaults to `"pil"`):
                 The output format of the generated image. Choose between `PIL.Image` or `np.array`.
             return_dict (`bool`, *optional*, defaults to `True`):
@@ -1053,6 +1086,11 @@ class StableDiffusionControlNetImg2ImgPipeline(
         if self.do_classifier_free_guidance:
             prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds])
 
+        if ip_adapter_image is not None:
+            image_embeds, negative_image_embeds = self.encode_image(ip_adapter_image, device, num_images_per_prompt)
+            if self.do_classifier_free_guidance:
+                image_embeds = torch.cat([negative_image_embeds, image_embeds])
+
         # 4. Prepare image
         image = self.image_processor.preprocess(image, height=height, width=width).to(dtype=torch.float32)
 
@@ -1111,7 +1149,10 @@ class StableDiffusionControlNetImg2ImgPipeline(
         # 7. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
         extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
 
-        # 7.1 Create tensor stating which controlnets to keep
+        # 7.1 Add image embeds for IP-Adapter
+        added_cond_kwargs = {"image_embeds": image_embeds} if ip_adapter_image is not None else None
+
+        # 7.2 Create tensor stating which controlnets to keep
         controlnet_keep = []
         for i in range(len(timesteps)):
             keeps = [
@@ -1171,6 +1212,7 @@ class StableDiffusionControlNetImg2ImgPipeline(
                     cross_attention_kwargs=self.cross_attention_kwargs,
                     down_block_additional_residuals=down_block_res_samples,
                     mid_block_additional_residual=mid_block_res_sample,
+                    added_cond_kwargs=added_cond_kwargs,
                     return_dict=False,
                 )[0]
 

diffusers/pipelines/controlnet/pipeline_controlnet_inpaint.py

@@ -25,7 +25,7 @@ from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer, CLIPV
 
 from ...image_processor import PipelineImageInput, VaeImageProcessor
 from ...loaders import FromSingleFileMixin, IPAdapterMixin, LoraLoaderMixin, TextualInversionLoaderMixin
-from ...models import AutoencoderKL, ControlNetModel, UNet2DConditionModel
+from ...models import AutoencoderKL, ControlNetModel, ImageProjection, UNet2DConditionModel
 from ...models.lora import adjust_lora_scale_text_encoder
 from ...schedulers import KarrasDiffusionSchedulers
 from ...utils import (
@@ -251,6 +251,9 @@ class StableDiffusionControlNetInpaintPipeline(
 
     The pipeline also inherits the following loading methods:
         - [`~loaders.TextualInversionLoaderMixin.load_textual_inversion`] for loading textual inversion embeddings
+        - [`~loaders.LoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+        - [`~loaders.LoraLoaderMixin.save_lora_weights`] for saving LoRA weights
+        - [`~loaders.FromSingleFileMixin.from_single_file`] for loading `.ckpt` files
         - [`~loaders.IPAdapterMixin.load_ip_adapter`] for loading IP Adapters
 
     <Tip>
@@ -288,7 +291,7 @@ class StableDiffusionControlNetInpaintPipeline(
             A `CLIPImageProcessor` to extract features from generated images; used as inputs to the `safety_checker`.
     """
 
-    model_cpu_offload_seq = "text_encoder->unet->vae"
+    model_cpu_offload_seq = "text_encoder->image_encoder->unet->vae"
     _optional_components = ["safety_checker", "feature_extractor", "image_encoder"]
     _exclude_from_cpu_offload = ["safety_checker"]
     _callback_tensor_inputs = ["latents", "prompt_embeds", "negative_prompt_embeds"]
@@ -597,18 +600,29 @@ class StableDiffusionControlNetInpaintPipeline(
         return prompt_embeds, negative_prompt_embeds
 
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.encode_image
-    def encode_image(self, image, device, num_images_per_prompt):
+    def encode_image(self, image, device, num_images_per_prompt, output_hidden_states=None):
         dtype = next(self.image_encoder.parameters()).dtype
 
         if not isinstance(image, torch.Tensor):
             image = self.feature_extractor(image, return_tensors="pt").pixel_values
 
         image = image.to(device=device, dtype=dtype)
-        image_embeds = self.image_encoder(image).image_embeds
-        image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0)
+        if output_hidden_states:
+            image_enc_hidden_states = self.image_encoder(image, output_hidden_states=True).hidden_states[-2]
+            image_enc_hidden_states = image_enc_hidden_states.repeat_interleave(num_images_per_prompt, dim=0)
+            uncond_image_enc_hidden_states = self.image_encoder(
+                torch.zeros_like(image), output_hidden_states=True
+            ).hidden_states[-2]
+            uncond_image_enc_hidden_states = uncond_image_enc_hidden_states.repeat_interleave(
+                num_images_per_prompt, dim=0
+            )
+            return image_enc_hidden_states, uncond_image_enc_hidden_states
+        else:
+            image_embeds = self.image_encoder(image).image_embeds
+            image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0)
+            uncond_image_embeds = torch.zeros_like(image_embeds)
 
-        uncond_image_embeds = torch.zeros_like(image_embeds)
-        return image_embeds, uncond_image_embeds
+            return image_embeds, uncond_image_embeds
 
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.run_safety_checker
     def run_safety_checker(self, image, device, dtype):
@@ -1284,7 +1298,10 @@ class StableDiffusionControlNetInpaintPipeline(
             prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds])
 
         if ip_adapter_image is not None:
-            image_embeds, negative_image_embeds = self.encode_image(ip_adapter_image, device, num_images_per_prompt)
+            output_hidden_state = False if isinstance(self.unet.encoder_hid_proj, ImageProjection) else True
+            image_embeds, negative_image_embeds = self.encode_image(
+                ip_adapter_image, device, num_images_per_prompt, output_hidden_state
+            )
             if self.do_classifier_free_guidance:
                 image_embeds = torch.cat([negative_image_embeds, image_embeds])
 

diffusers/pipelines/controlnet/pipeline_controlnet_inpaint_sd_xl.py

@@ -148,12 +148,10 @@ class StableDiffusionXLControlNetInpaintPipeline(
     This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the
     library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)
 
-    In addition the pipeline inherits the following loading methods:
-        - *LoRA*: [`loaders.StableDiffusionXLLoraLoaderMixin.load_lora_weights`]
-        - *Ckpt*: [`loaders.FromSingleFileMixin.from_single_file`]
-
-    as well as the following saving methods:
-        - *LoRA*: [`loaders.StableDiffusionXLLoraLoaderMixin.save_lora_weights`]
+    The pipeline also inherits the following loading methods:
+        - [`~loaders.StableDiffusionXLLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+        - [`~loaders.StableDiffusionXLLoraLoaderMixin.save_lora_weights`] for saving LoRA weights
+        - [`~loaders.FromSingleFileMixin.from_single_file`] for loading `.ckpt` files
 
     Args:
         vae ([`AutoencoderKL`]):

diffusers/pipelines/controlnet/pipeline_controlnet_sd_xl.py

@@ -37,7 +37,7 @@ from ...loaders import (
     StableDiffusionXLLoraLoaderMixin,
     TextualInversionLoaderMixin,
 )
-from ...models import AutoencoderKL, ControlNetModel, UNet2DConditionModel
+from ...models import AutoencoderKL, ControlNetModel, ImageProjection, UNet2DConditionModel
 from ...models.attention_processor import (
     AttnProcessor2_0,
     LoRAAttnProcessor2_0,
@@ -129,8 +129,10 @@ class StableDiffusionXLControlNetPipeline(
 
     The pipeline also inherits the following loading methods:
         - [`~loaders.TextualInversionLoaderMixin.load_textual_inversion`] for loading textual inversion embeddings
-        - [`loaders.StableDiffusionXLLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
-        - [`loaders.FromSingleFileMixin.from_single_file`] for loading `.ckpt` files
+        - [`~loaders.StableDiffusionXLLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+        - [`~loaders.StableDiffusionXLLoraLoaderMixin.save_lora_weights`] for saving LoRA weights
+        - [`~loaders.FromSingleFileMixin.from_single_file`] for loading `.ckpt` files
+        - [`~loaders.IPAdapterMixin.load_ip_adapter`] for loading IP Adapters
 
     Args:
         vae ([`AutoencoderKL`]):
@@ -163,7 +165,7 @@ class StableDiffusionXLControlNetPipeline(
     """
 
     # leave controlnet out on purpose because it iterates with unet
-    model_cpu_offload_seq = "text_encoder->text_encoder_2->unet->vae"
+    model_cpu_offload_seq = "text_encoder->text_encoder_2->image_encoder->unet->vae"
     _optional_components = [
         "tokenizer",
         "tokenizer_2",
@@ -489,18 +491,29 @@ class StableDiffusionXLControlNetPipeline(
         return prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds
 
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.encode_image
-    def encode_image(self, image, device, num_images_per_prompt):
+    def encode_image(self, image, device, num_images_per_prompt, output_hidden_states=None):
         dtype = next(self.image_encoder.parameters()).dtype
 
         if not isinstance(image, torch.Tensor):
             image = self.feature_extractor(image, return_tensors="pt").pixel_values
 
         image = image.to(device=device, dtype=dtype)
-        image_embeds = self.image_encoder(image).image_embeds
-        image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0)
+        if output_hidden_states:
+            image_enc_hidden_states = self.image_encoder(image, output_hidden_states=True).hidden_states[-2]
+            image_enc_hidden_states = image_enc_hidden_states.repeat_interleave(num_images_per_prompt, dim=0)
+            uncond_image_enc_hidden_states = self.image_encoder(
+                torch.zeros_like(image), output_hidden_states=True
+            ).hidden_states[-2]
+            uncond_image_enc_hidden_states = uncond_image_enc_hidden_states.repeat_interleave(
+                num_images_per_prompt, dim=0
+            )
+            return image_enc_hidden_states, uncond_image_enc_hidden_states
+        else:
+            image_embeds = self.image_encoder(image).image_embeds
+            image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0)
+            uncond_image_embeds = torch.zeros_like(image_embeds)
 
-        uncond_image_embeds = torch.zeros_like(image_embeds)
-        return image_embeds, uncond_image_embeds
+            return image_embeds, uncond_image_embeds
 
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs
     def prepare_extra_step_kwargs(self, generator, eta):
@@ -1169,7 +1182,10 @@ class StableDiffusionXLControlNetPipeline(
 
         # 3.2 Encode ip_adapter_image
         if ip_adapter_image is not None:
-            image_embeds, negative_image_embeds = self.encode_image(ip_adapter_image, device, num_images_per_prompt)
+            output_hidden_state = False if isinstance(self.unet.encoder_hid_proj, ImageProjection) else True
+            image_embeds, negative_image_embeds = self.encode_image(
+                ip_adapter_image, device, num_images_per_prompt, output_hidden_state
+            )
             if self.do_classifier_free_guidance:
                 image_embeds = torch.cat([negative_image_embeds, image_embeds])
 

diffusers/pipelines/controlnet/pipeline_controlnet_sd_xl_img2img.py

@@ -155,9 +155,10 @@ class StableDiffusionXLControlNetImg2ImgPipeline(
     This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the
     library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)
 
-    In addition the pipeline inherits the following loading methods:
-        - *Textual-Inversion*: [`loaders.TextualInversionLoaderMixin.load_textual_inversion`]
-        - *LoRA*: [`loaders.StableDiffusionXLLoraLoaderMixin.load_lora_weights`]
+    The pipeline also inherits the following loading methods:
+        - [`~loaders.TextualInversionLoaderMixin.load_textual_inversion`] for loading textual inversion embeddings
+        - [`~loaders.StableDiffusionXLLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+        - [`~loaders.StableDiffusionXLLoraLoaderMixin.save_lora_weights`] for saving LoRA weights
 
     Args:
         vae ([`AutoencoderKL`]):

diffusers/pipelines/deprecated/__init__.py

@@ -0,0 +1,153 @@
+from typing import TYPE_CHECKING
+
+from ...utils import (
+    DIFFUSERS_SLOW_IMPORT,
+    OptionalDependencyNotAvailable,
+    _LazyModule,
+    get_objects_from_module,
+    is_librosa_available,
+    is_note_seq_available,
+    is_torch_available,
+    is_transformers_available,
+)
+
+
+_dummy_objects = {}
+_import_structure = {}
+
+try:
+    if not is_torch_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    from ...utils import dummy_pt_objects
+
+    _dummy_objects.update(get_objects_from_module(dummy_pt_objects))
+else:
+    _import_structure["latent_diffusion_uncond"] = ["LDMPipeline"]
+    _import_structure["pndm"] = ["PNDMPipeline"]
+    _import_structure["repaint"] = ["RePaintPipeline"]
+    _import_structure["score_sde_ve"] = ["ScoreSdeVePipeline"]
+    _import_structure["stochastic_karras_ve"] = ["KarrasVePipeline"]
+
+try:
+    if not (is_transformers_available() and is_torch_available()):
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    from ...utils import dummy_torch_and_transformers_objects
+
+    _dummy_objects.update(get_objects_from_module(dummy_torch_and_transformers_objects))
+else:
+    _import_structure["alt_diffusion"] = [
+        "AltDiffusionImg2ImgPipeline",
+        "AltDiffusionPipeline",
+        "AltDiffusionPipelineOutput",
+    ]
+    _import_structure["versatile_diffusion"] = [
+        "VersatileDiffusionDualGuidedPipeline",
+        "VersatileDiffusionImageVariationPipeline",
+        "VersatileDiffusionPipeline",
+        "VersatileDiffusionTextToImagePipeline",
+    ]
+    _import_structure["vq_diffusion"] = ["VQDiffusionPipeline"]
+    _import_structure["stable_diffusion_variants"] = [
+        "CycleDiffusionPipeline",
+        "StableDiffusionInpaintPipelineLegacy",
+        "StableDiffusionPix2PixZeroPipeline",
+        "StableDiffusionParadigmsPipeline",
+        "StableDiffusionModelEditingPipeline",
+    ]
+
+try:
+    if not (is_torch_available() and is_librosa_available()):
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    from ...utils import dummy_torch_and_librosa_objects  # noqa F403
+
+    _dummy_objects.update(get_objects_from_module(dummy_torch_and_librosa_objects))
+
+else:
+    _import_structure["audio_diffusion"] = ["AudioDiffusionPipeline", "Mel"]
+
+try:
+    if not (is_transformers_available() and is_torch_available() and is_note_seq_available()):
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    from ...utils import dummy_transformers_and_torch_and_note_seq_objects  # noqa F403
+
+    _dummy_objects.update(get_objects_from_module(dummy_transformers_and_torch_and_note_seq_objects))
+
+else:
+    _import_structure["spectrogram_diffusion"] = ["MidiProcessor", "SpectrogramDiffusionPipeline"]
+
+
+if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
+    try:
+        if not is_torch_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        from ...utils.dummy_pt_objects import *
+
+    else:
+        from .latent_diffusion_uncond import LDMPipeline
+        from .pndm import PNDMPipeline
+        from .repaint import RePaintPipeline
+        from .score_sde_ve import ScoreSdeVePipeline
+        from .stochastic_karras_ve import KarrasVePipeline
+
+    try:
+        if not (is_transformers_available() and is_torch_available()):
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        from ...utils.dummy_torch_and_transformers_objects import *
+
+    else:
+        from .alt_diffusion import AltDiffusionImg2ImgPipeline, AltDiffusionPipeline, AltDiffusionPipelineOutput
+        from .audio_diffusion import AudioDiffusionPipeline, Mel
+        from .spectrogram_diffusion import SpectrogramDiffusionPipeline
+        from .stable_diffusion_variants import (
+            CycleDiffusionPipeline,
+            StableDiffusionInpaintPipelineLegacy,
+            StableDiffusionModelEditingPipeline,
+            StableDiffusionParadigmsPipeline,
+            StableDiffusionPix2PixZeroPipeline,
+        )
+        from .stochastic_karras_ve import KarrasVePipeline
+        from .versatile_diffusion import (
+            VersatileDiffusionDualGuidedPipeline,
+            VersatileDiffusionImageVariationPipeline,
+            VersatileDiffusionPipeline,
+            VersatileDiffusionTextToImagePipeline,
+        )
+        from .vq_diffusion import VQDiffusionPipeline
+
+    try:
+        if not (is_torch_available() and is_librosa_available()):
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        from ...utils.dummy_torch_and_librosa_objects import *
+    else:
+        from .audio_diffusion import AudioDiffusionPipeline, Mel
+
+    try:
+        if not (is_transformers_available() and is_torch_available() and is_note_seq_available()):
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        from ...utils.dummy_transformers_and_torch_and_note_seq_objects import *  # noqa F403
+    else:
+        from .spectrogram_diffusion import (
+            MidiProcessor,
+            SpectrogramDiffusionPipeline,
+        )
+
+
+else:
+    import sys
+
+    sys.modules[__name__] = _LazyModule(
+        __name__,
+        globals()["__file__"],
+        _import_structure,
+        module_spec=__spec__,
+    )
+    for name, value in _dummy_objects.items():
+        setattr(sys.modules[__name__], name, value)

diffusers/pipelines/{alt_diffusion → deprecated/alt_diffusion}/__init__.py

@@ -1,6 +1,6 @@
 from typing import TYPE_CHECKING
 
-from ...utils import (
+from ....utils import (
     DIFFUSERS_SLOW_IMPORT,
     OptionalDependencyNotAvailable,
     _LazyModule,
@@ -17,7 +17,7 @@ try:
     if not (is_transformers_available() and is_torch_available()):
         raise OptionalDependencyNotAvailable()
 except OptionalDependencyNotAvailable:
-    from ...utils import dummy_torch_and_transformers_objects
+    from ....utils import dummy_torch_and_transformers_objects
 
     _dummy_objects.update(get_objects_from_module(dummy_torch_and_transformers_objects))
 else:
@@ -32,7 +32,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
         if not (is_transformers_available() and is_torch_available()):
             raise OptionalDependencyNotAvailable()
     except OptionalDependencyNotAvailable:
-        from ...utils.dummy_torch_and_transformers_objects import *
+        from ....utils.dummy_torch_and_transformers_objects import *
 
     else:
         from .modeling_roberta_series import RobertaSeriesModelWithTransformation