[Chore] add `LoraLoaderMixin` to the inits (#8981)

* introduce  to promote reusability.

* up

* add more tests

* up

* remove comments.

* fix fuse_nan test

* clarify the scope of fuse_lora and unfuse_lora

* remove space

* rewrite fuse_lora a bit.

* feedback

* copy over load_lora_into_text_encoder.

* address dhruv's feedback.

* fix-copies

* fix issubclass.

* num_fused_loras

* fix

* fix

* remove mapping

* up

* fix

* style

* fix-copies

* change to SD3TransformerLoRALoadersMixin

* Apply suggestions from code review
Co-authored-by: Dhruv Nair <dhruv.nair@gmail.com>

* up

* handle wuerstchen

* up

* move lora to lora_pipeline.py

* up

* fix-copies

* fix documentation.

* comment set_adapters().

* fix-copies

* fix set_adapters() at the model level.

* fix?

* fix

* loraloadermixin.

---------
Co-authored-by: Dhruv Nair <dhruv.nair@gmail.com>

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_img2img.py

@@ -24,7 +24,7 @@ from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer, CLIPV
 from ...callbacks import MultiPipelineCallbacks, PipelineCallback
 from ...configuration_utils import FrozenDict
 from ...image_processor import PipelineImageInput, VaeImageProcessor
-from ...loaders import FromSingleFileMixin, IPAdapterMixin, LoraLoaderMixin, TextualInversionLoaderMixin
+from ...loaders import FromSingleFileMixin, IPAdapterMixin, StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
 from ...models import AutoencoderKL, ImageProjection, UNet2DConditionModel
 from ...models.lora import adjust_lora_scale_text_encoder
 from ...schedulers import KarrasDiffusionSchedulers
@@ -175,7 +175,7 @@ class StableDiffusionImg2ImgPipeline(
     StableDiffusionMixin,
     TextualInversionLoaderMixin,
     IPAdapterMixin,
-    LoraLoaderMixin,
+    StableDiffusionLoraLoaderMixin,
     FromSingleFileMixin,
 ):
     r"""
@@ -186,8 +186,8 @@ class StableDiffusionImg2ImgPipeline(
 
     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.StableDiffusionLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+        - [`~loaders.StableDiffusionLoraLoaderMixin.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
 
@@ -385,7 +385,7 @@ class StableDiffusionImg2ImgPipeline(
         """
         # set lora scale so that monkey patched LoRA
         # function of text encoder can correctly access it
-        if lora_scale is not None and isinstance(self, LoraLoaderMixin):
+        if lora_scale is not None and isinstance(self, StableDiffusionLoraLoaderMixin):
             self._lora_scale = lora_scale
 
             # dynamically adjust the LoRA scale
@@ -518,7 +518,7 @@ class StableDiffusionImg2ImgPipeline(
             negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
 
         if self.text_encoder is not None:
-            if isinstance(self, LoraLoaderMixin) and USE_PEFT_BACKEND:
+            if isinstance(self, StableDiffusionLoraLoaderMixin) and USE_PEFT_BACKEND:
                 # Retrieve the original scale by scaling back the LoRA layers
                 unscale_lora_layers(self.text_encoder, lora_scale)
 

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint.py

@@ -23,7 +23,7 @@ from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer, CLIPV
 from ...callbacks import MultiPipelineCallbacks, PipelineCallback
 from ...configuration_utils import FrozenDict
 from ...image_processor import PipelineImageInput, VaeImageProcessor
-from ...loaders import FromSingleFileMixin, IPAdapterMixin, LoraLoaderMixin, TextualInversionLoaderMixin
+from ...loaders import FromSingleFileMixin, IPAdapterMixin, StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
 from ...models import AsymmetricAutoencoderKL, AutoencoderKL, ImageProjection, UNet2DConditionModel
 from ...models.lora import adjust_lora_scale_text_encoder
 from ...schedulers import KarrasDiffusionSchedulers
@@ -116,7 +116,7 @@ class StableDiffusionInpaintPipeline(
     StableDiffusionMixin,
     TextualInversionLoaderMixin,
     IPAdapterMixin,
-    LoraLoaderMixin,
+    StableDiffusionLoraLoaderMixin,
     FromSingleFileMixin,
 ):
     r"""
@@ -127,8 +127,8 @@ class StableDiffusionInpaintPipeline(
 
     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.StableDiffusionLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+        - [`~loaders.StableDiffusionLoraLoaderMixin.save_lora_weights`] for saving LoRA weights
         - [`~loaders.IPAdapterMixin.load_ip_adapter`] for loading IP Adapters
         - [`~loaders.FromSingleFileMixin.from_single_file`] for loading `.ckpt` files
 
@@ -334,7 +334,7 @@ class StableDiffusionInpaintPipeline(
         """
         # set lora scale so that monkey patched LoRA
         # function of text encoder can correctly access it
-        if lora_scale is not None and isinstance(self, LoraLoaderMixin):
+        if lora_scale is not None and isinstance(self, StableDiffusionLoraLoaderMixin):
             self._lora_scale = lora_scale
 
             # dynamically adjust the LoRA scale
@@ -467,7 +467,7 @@ class StableDiffusionInpaintPipeline(
             negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
 
         if self.text_encoder is not None:
-            if isinstance(self, LoraLoaderMixin) and USE_PEFT_BACKEND:
+            if isinstance(self, StableDiffusionLoraLoaderMixin) and USE_PEFT_BACKEND:
                 # Retrieve the original scale by scaling back the LoRA layers
                 unscale_lora_layers(self.text_encoder, lora_scale)
 

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_instruct_pix2pix.py

@@ -22,7 +22,7 @@ from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer, CLIPV
 
 from ...callbacks import MultiPipelineCallbacks, PipelineCallback
 from ...image_processor import PipelineImageInput, VaeImageProcessor
-from ...loaders import IPAdapterMixin, LoraLoaderMixin, TextualInversionLoaderMixin
+from ...loaders import IPAdapterMixin, StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
 from ...models import AutoencoderKL, ImageProjection, UNet2DConditionModel
 from ...schedulers import KarrasDiffusionSchedulers
 from ...utils import PIL_INTERPOLATION, deprecate, logging
@@ -74,7 +74,11 @@ def retrieve_latents(
 
 
 class StableDiffusionInstructPix2PixPipeline(
-    DiffusionPipeline, StableDiffusionMixin, TextualInversionLoaderMixin, LoraLoaderMixin, IPAdapterMixin
+    DiffusionPipeline,
+    StableDiffusionMixin,
+    TextualInversionLoaderMixin,
+    StableDiffusionLoraLoaderMixin,
+    IPAdapterMixin,
 ):
     r"""
     Pipeline for pixel-level image editing by following text instructions (based on Stable Diffusion).
@@ -84,8 +88,8 @@ class StableDiffusionInstructPix2PixPipeline(
 
     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.StableDiffusionLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+        - [`~loaders.StableDiffusionLoraLoaderMixin.save_lora_weights`] for saving LoRA weights
         - [`~loaders.IPAdapterMixin.load_ip_adapter`] for loading IP Adapters
 
     Args:

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_upscale.py

@@ -22,7 +22,7 @@ import torch
 from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer
 
 from ...image_processor import PipelineImageInput, VaeImageProcessor
-from ...loaders import FromSingleFileMixin, LoraLoaderMixin, TextualInversionLoaderMixin
+from ...loaders import FromSingleFileMixin, StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
 from ...models import AutoencoderKL, UNet2DConditionModel
 from ...models.attention_processor import (
     AttnProcessor2_0,
@@ -66,7 +66,11 @@ def preprocess(image):
 
 
 class StableDiffusionUpscalePipeline(
-    DiffusionPipeline, StableDiffusionMixin, TextualInversionLoaderMixin, LoraLoaderMixin, FromSingleFileMixin
+    DiffusionPipeline,
+    StableDiffusionMixin,
+    TextualInversionLoaderMixin,
+    StableDiffusionLoraLoaderMixin,
+    FromSingleFileMixin,
 ):
     r"""
     Pipeline for text-guided image super-resolution using Stable Diffusion 2.
@@ -76,8 +80,8 @@ class StableDiffusionUpscalePipeline(
 
     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.StableDiffusionLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+        - [`~loaders.StableDiffusionLoraLoaderMixin.save_lora_weights`] for saving LoRA weights
         - [`~loaders.FromSingleFileMixin.from_single_file`] for loading `.ckpt` files
 
     Args:
@@ -243,7 +247,7 @@ class StableDiffusionUpscalePipeline(
         """
         # set lora scale so that monkey patched LoRA
         # function of text encoder can correctly access it
-        if lora_scale is not None and isinstance(self, LoraLoaderMixin):
+        if lora_scale is not None and isinstance(self, StableDiffusionLoraLoaderMixin):
             self._lora_scale = lora_scale
 
             # dynamically adjust the LoRA scale
@@ -376,7 +380,7 @@ class StableDiffusionUpscalePipeline(
             negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
 
         if self.text_encoder is not None:
-            if isinstance(self, LoraLoaderMixin) and USE_PEFT_BACKEND:
+            if isinstance(self, StableDiffusionLoraLoaderMixin) and USE_PEFT_BACKEND:
                 # Retrieve the original scale by scaling back the LoRA layers
                 unscale_lora_layers(self.text_encoder, lora_scale)
 

src/diffusers/pipelines/stable_diffusion/pipeline_stable_unclip.py

@@ -20,7 +20,7 @@ from transformers import CLIPTextModel, CLIPTextModelWithProjection, CLIPTokeniz
 from transformers.models.clip.modeling_clip import CLIPTextModelOutput
 
 from ...image_processor import VaeImageProcessor
-from ...loaders import LoraLoaderMixin, TextualInversionLoaderMixin
+from ...loaders import StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
 from ...models import AutoencoderKL, PriorTransformer, UNet2DConditionModel
 from ...models.embeddings import get_timestep_embedding
 from ...models.lora import adjust_lora_scale_text_encoder
@@ -58,7 +58,9 @@ EXAMPLE_DOC_STRING = """
 """
 
 
-class StableUnCLIPPipeline(DiffusionPipeline, StableDiffusionMixin, TextualInversionLoaderMixin, LoraLoaderMixin):
+class StableUnCLIPPipeline(
+    DiffusionPipeline, StableDiffusionMixin, TextualInversionLoaderMixin, StableDiffusionLoraLoaderMixin
+):
     """
     Pipeline for text-to-image generation using stable unCLIP.
 
@@ -67,8 +69,8 @@ class StableUnCLIPPipeline(DiffusionPipeline, StableDiffusionMixin, TextualInver
 
     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.StableDiffusionLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+        - [`~loaders.StableDiffusionLoraLoaderMixin.save_lora_weights`] for saving LoRA weights
 
     Args:
         prior_tokenizer ([`CLIPTokenizer`]):
@@ -326,7 +328,7 @@ class StableUnCLIPPipeline(DiffusionPipeline, StableDiffusionMixin, TextualInver
         """
         # set lora scale so that monkey patched LoRA
         # function of text encoder can correctly access it
-        if lora_scale is not None and isinstance(self, LoraLoaderMixin):
+        if lora_scale is not None and isinstance(self, StableDiffusionLoraLoaderMixin):
             self._lora_scale = lora_scale
 
             # dynamically adjust the LoRA scale
@@ -459,7 +461,7 @@ class StableUnCLIPPipeline(DiffusionPipeline, StableDiffusionMixin, TextualInver
             negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
 
         if self.text_encoder is not None:
-            if isinstance(self, LoraLoaderMixin) and USE_PEFT_BACKEND:
+            if isinstance(self, StableDiffusionLoraLoaderMixin) and USE_PEFT_BACKEND:
                 # Retrieve the original scale by scaling back the LoRA layers
                 unscale_lora_layers(self.text_encoder, lora_scale)
 

src/diffusers/pipelines/stable_diffusion/pipeline_stable_unclip_img2img.py

@@ -20,7 +20,7 @@ import torch
 from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer, CLIPVisionModelWithProjection
 
 from ...image_processor import VaeImageProcessor
-from ...loaders import LoraLoaderMixin, TextualInversionLoaderMixin
+from ...loaders import StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
 from ...models import AutoencoderKL, UNet2DConditionModel
 from ...models.embeddings import get_timestep_embedding
 from ...models.lora import adjust_lora_scale_text_encoder
@@ -70,7 +70,7 @@ EXAMPLE_DOC_STRING = """
 
 
 class StableUnCLIPImg2ImgPipeline(
-    DiffusionPipeline, StableDiffusionMixin, TextualInversionLoaderMixin, LoraLoaderMixin
+    DiffusionPipeline, StableDiffusionMixin, TextualInversionLoaderMixin, StableDiffusionLoraLoaderMixin
 ):
     """
     Pipeline for text-guided image-to-image generation using stable unCLIP.
@@ -80,8 +80,8 @@ class StableUnCLIPImg2ImgPipeline(
 
     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.StableDiffusionLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+        - [`~loaders.StableDiffusionLoraLoaderMixin.save_lora_weights`] for saving LoRA weights
 
     Args:
         feature_extractor ([`CLIPImageProcessor`]):
@@ -290,7 +290,7 @@ class StableUnCLIPImg2ImgPipeline(
         """
         # set lora scale so that monkey patched LoRA
         # function of text encoder can correctly access it
-        if lora_scale is not None and isinstance(self, LoraLoaderMixin):
+        if lora_scale is not None and isinstance(self, StableDiffusionLoraLoaderMixin):
             self._lora_scale = lora_scale
 
             # dynamically adjust the LoRA scale
@@ -423,7 +423,7 @@ class StableUnCLIPImg2ImgPipeline(
             negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
 
         if self.text_encoder is not None:
-            if isinstance(self, LoraLoaderMixin) and USE_PEFT_BACKEND:
+            if isinstance(self, StableDiffusionLoraLoaderMixin) and USE_PEFT_BACKEND:
                 # Retrieve the original scale by scaling back the LoRA layers
                 unscale_lora_layers(self.text_encoder, lora_scale)
 

src/diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3.py

@@ -29,9 +29,12 @@ from ...models.autoencoders import AutoencoderKL
 from ...models.transformers import SD3Transformer2DModel
 from ...schedulers import FlowMatchEulerDiscreteScheduler
 from ...utils import (
+    USE_PEFT_BACKEND,
     is_torch_xla_available,
     logging,
     replace_example_docstring,
+    scale_lora_layers,
+    unscale_lora_layers,
 )
 from ...utils.torch_utils import randn_tensor
 from ..pipeline_utils import DiffusionPipeline
@@ -329,6 +332,7 @@ class StableDiffusion3Pipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingle
         negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
         clip_skip: Optional[int] = None,
         max_sequence_length: int = 256,
+        lora_scale: Optional[float] = None,
     ):
         r"""
 
@@ -374,9 +378,22 @@ class StableDiffusion3Pipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingle
             clip_skip (`int`, *optional*):
                 Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
                 the output of the pre-final layer will be used for computing the prompt embeddings.
+            lora_scale (`float`, *optional*):
+                A lora scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded.
         """
         device = device or self._execution_device
 
+        # set lora scale so that monkey patched LoRA
+        # function of text encoder can correctly access it
+        if lora_scale is not None and isinstance(self, SD3LoraLoaderMixin):
+            self._lora_scale = lora_scale
+
+            # dynamically adjust the LoRA scale
+            if self.text_encoder is not None and USE_PEFT_BACKEND:
+                scale_lora_layers(self.text_encoder, lora_scale)
+            if self.text_encoder_2 is not None and USE_PEFT_BACKEND:
+                scale_lora_layers(self.text_encoder_2, lora_scale)
+
         prompt = [prompt] if isinstance(prompt, str) else prompt
         if prompt is not None:
             batch_size = len(prompt)
@@ -479,6 +496,16 @@ class StableDiffusion3Pipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingle
                 [negative_pooled_prompt_embed, negative_pooled_prompt_2_embed], dim=-1
             )
 
+        if self.text_encoder is not None:
+            if isinstance(self, SD3LoraLoaderMixin) and USE_PEFT_BACKEND:
+                # Retrieve the original scale by scaling back the LoRA layers
+                unscale_lora_layers(self.text_encoder, lora_scale)
+
+        if self.text_encoder_2 is not None:
+            if isinstance(self, SD3LoraLoaderMixin) and USE_PEFT_BACKEND:
+                # Retrieve the original scale by scaling back the LoRA layers
+                unscale_lora_layers(self.text_encoder_2, lora_scale)
+
         return prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds
 
     def check_inputs(
@@ -787,6 +814,9 @@ class StableDiffusion3Pipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingle
 
         device = self._execution_device
 
+        lora_scale = (
+            self.joint_attention_kwargs.get("scale", None) if self.joint_attention_kwargs is not None else None
+        )
         (
             prompt_embeds,
             negative_prompt_embeds,
@@ -808,6 +838,7 @@ class StableDiffusion3Pipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingle
             clip_skip=self.clip_skip,
             num_images_per_prompt=num_images_per_prompt,
             max_sequence_length=max_sequence_length,
+            lora_scale=lora_scale,
         )
 
         if self.do_classifier_free_guidance:

src/diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3_img2img.py

@@ -25,13 +25,17 @@ from transformers import (
 )
 
 from ...image_processor import PipelineImageInput, VaeImageProcessor
+from ...loaders import SD3LoraLoaderMixin
 from ...models.autoencoders import AutoencoderKL
 from ...models.transformers import SD3Transformer2DModel
 from ...schedulers import FlowMatchEulerDiscreteScheduler
 from ...utils import (
+    USE_PEFT_BACKEND,
     is_torch_xla_available,
     logging,
     replace_example_docstring,
+    scale_lora_layers,
+    unscale_lora_layers,
 )
 from ...utils.torch_utils import randn_tensor
 from ..pipeline_utils import DiffusionPipeline
@@ -346,6 +350,7 @@ class StableDiffusion3Img2ImgPipeline(DiffusionPipeline):
         negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
         clip_skip: Optional[int] = None,
         max_sequence_length: int = 256,
+        lora_scale: Optional[float] = None,
     ):
         r"""
 
@@ -391,9 +396,22 @@ class StableDiffusion3Img2ImgPipeline(DiffusionPipeline):
             clip_skip (`int`, *optional*):
                 Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
                 the output of the pre-final layer will be used for computing the prompt embeddings.
+            lora_scale (`float`, *optional*):
+                A lora scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded.
         """
         device = device or self._execution_device
 
+        # set lora scale so that monkey patched LoRA
+        # function of text encoder can correctly access it
+        if lora_scale is not None and isinstance(self, SD3LoraLoaderMixin):
+            self._lora_scale = lora_scale
+
+            # dynamically adjust the LoRA scale
+            if self.text_encoder is not None and USE_PEFT_BACKEND:
+                scale_lora_layers(self.text_encoder, lora_scale)
+            if self.text_encoder_2 is not None and USE_PEFT_BACKEND:
+                scale_lora_layers(self.text_encoder_2, lora_scale)
+
         prompt = [prompt] if isinstance(prompt, str) else prompt
         if prompt is not None:
             batch_size = len(prompt)
@@ -496,6 +514,16 @@ class StableDiffusion3Img2ImgPipeline(DiffusionPipeline):
                 [negative_pooled_prompt_embed, negative_pooled_prompt_2_embed], dim=-1
             )
 
+        if self.text_encoder is not None:
+            if isinstance(self, SD3LoraLoaderMixin) and USE_PEFT_BACKEND:
+                # Retrieve the original scale by scaling back the LoRA layers
+                unscale_lora_layers(self.text_encoder, lora_scale)
+
+        if self.text_encoder_2 is not None:
+            if isinstance(self, SD3LoraLoaderMixin) and USE_PEFT_BACKEND:
+                # Retrieve the original scale by scaling back the LoRA layers
+                unscale_lora_layers(self.text_encoder_2, lora_scale)
+
         return prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds
 
     def check_inputs(

src/diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3_inpaint.py

@@ -25,13 +25,17 @@ from transformers import (
 
 from ...callbacks import MultiPipelineCallbacks, PipelineCallback
 from ...image_processor import PipelineImageInput, VaeImageProcessor
+from ...loaders import SD3LoraLoaderMixin
 from ...models.autoencoders import AutoencoderKL
 from ...models.transformers import SD3Transformer2DModel
 from ...schedulers import FlowMatchEulerDiscreteScheduler
 from ...utils import (
+    USE_PEFT_BACKEND,
     is_torch_xla_available,
     logging,
     replace_example_docstring,
+    scale_lora_layers,
+    unscale_lora_layers,
 )
 from ...utils.torch_utils import randn_tensor
 from ..pipeline_utils import DiffusionPipeline
@@ -352,6 +356,7 @@ class StableDiffusion3InpaintPipeline(DiffusionPipeline):
         negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
         clip_skip: Optional[int] = None,
         max_sequence_length: int = 256,
+        lora_scale: Optional[float] = None,
     ):
         r"""
 
@@ -397,9 +402,22 @@ class StableDiffusion3InpaintPipeline(DiffusionPipeline):
             clip_skip (`int`, *optional*):
                 Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
                 the output of the pre-final layer will be used for computing the prompt embeddings.
+            lora_scale (`float`, *optional*):
+                A lora scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded.
         """
         device = device or self._execution_device
 
+        # set lora scale so that monkey patched LoRA
+        # function of text encoder can correctly access it
+        if lora_scale is not None and isinstance(self, SD3LoraLoaderMixin):
+            self._lora_scale = lora_scale
+
+            # dynamically adjust the LoRA scale
+            if self.text_encoder is not None and USE_PEFT_BACKEND:
+                scale_lora_layers(self.text_encoder, lora_scale)
+            if self.text_encoder_2 is not None and USE_PEFT_BACKEND:
+                scale_lora_layers(self.text_encoder_2, lora_scale)
+
         prompt = [prompt] if isinstance(prompt, str) else prompt
         if prompt is not None:
             batch_size = len(prompt)
@@ -502,6 +520,16 @@ class StableDiffusion3InpaintPipeline(DiffusionPipeline):
                 [negative_pooled_prompt_embed, negative_pooled_prompt_2_embed], dim=-1
             )
 
+        if self.text_encoder is not None:
+            if isinstance(self, SD3LoraLoaderMixin) and USE_PEFT_BACKEND:
+                # Retrieve the original scale by scaling back the LoRA layers
+                unscale_lora_layers(self.text_encoder, lora_scale)
+
+        if self.text_encoder_2 is not None:
+            if isinstance(self, SD3LoraLoaderMixin) and USE_PEFT_BACKEND:
+                # Retrieve the original scale by scaling back the LoRA layers
+                unscale_lora_layers(self.text_encoder_2, lora_scale)
+
         return prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds
 
     # Copied from diffusers.pipelines.stable_diffusion_3.pipeline_stable_diffusion_3_img2img.StableDiffusion3Img2ImgPipeline.check_inputs

src/diffusers/pipelines/stable_diffusion_attend_and_excite/pipeline_stable_diffusion_attend_and_excite.py

@@ -22,7 +22,7 @@ from torch.nn import functional as F
 from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer
 
 from ...image_processor import VaeImageProcessor
-from ...loaders import LoraLoaderMixin, TextualInversionLoaderMixin
+from ...loaders import StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
 from ...models import AutoencoderKL, UNet2DConditionModel
 from ...models.attention_processor import Attention
 from ...models.lora import adjust_lora_scale_text_encoder
@@ -323,7 +323,7 @@ class StableDiffusionAttendAndExcitePipeline(DiffusionPipeline, StableDiffusionM
         """
         # set lora scale so that monkey patched LoRA
         # function of text encoder can correctly access it
-        if lora_scale is not None and isinstance(self, LoraLoaderMixin):
+        if lora_scale is not None and isinstance(self, StableDiffusionLoraLoaderMixin):
             self._lora_scale = lora_scale
 
             # dynamically adjust the LoRA scale
@@ -456,7 +456,7 @@ class StableDiffusionAttendAndExcitePipeline(DiffusionPipeline, StableDiffusionM
             negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
 
         if self.text_encoder is not None:
-            if isinstance(self, LoraLoaderMixin) and USE_PEFT_BACKEND:
+            if isinstance(self, StableDiffusionLoraLoaderMixin) and USE_PEFT_BACKEND:
                 # Retrieve the original scale by scaling back the LoRA layers
                 unscale_lora_layers(self.text_encoder, lora_scale)
 

src/diffusers/pipelines/stable_diffusion_diffedit/pipeline_stable_diffusion_diffedit.py

@@ -24,7 +24,7 @@ from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer
 
 from ...configuration_utils import FrozenDict
 from ...image_processor import VaeImageProcessor
-from ...loaders import LoraLoaderMixin, TextualInversionLoaderMixin
+from ...loaders import StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
 from ...models import AutoencoderKL, UNet2DConditionModel
 from ...models.lora import adjust_lora_scale_text_encoder
 from ...schedulers import DDIMInverseScheduler, KarrasDiffusionSchedulers
@@ -234,7 +234,7 @@ def preprocess_mask(mask, batch_size: int = 1):
 
 
 class StableDiffusionDiffEditPipeline(
-    DiffusionPipeline, StableDiffusionMixin, TextualInversionLoaderMixin, LoraLoaderMixin
+    DiffusionPipeline, StableDiffusionMixin, TextualInversionLoaderMixin, StableDiffusionLoraLoaderMixin
 ):
     r"""
     <Tip warning={true}>
@@ -250,8 +250,8 @@ class StableDiffusionDiffEditPipeline(
 
     The pipeline also inherits the following loading and saving 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.StableDiffusionLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+        - [`~loaders.StableDiffusionLoraLoaderMixin.save_lora_weights`] for saving LoRA weights
 
     Args:
         vae ([`AutoencoderKL`]):
@@ -448,7 +448,7 @@ class StableDiffusionDiffEditPipeline(
         """
         # set lora scale so that monkey patched LoRA
         # function of text encoder can correctly access it
-        if lora_scale is not None and isinstance(self, LoraLoaderMixin):
+        if lora_scale is not None and isinstance(self, StableDiffusionLoraLoaderMixin):
             self._lora_scale = lora_scale
 
             # dynamically adjust the LoRA scale
@@ -581,7 +581,7 @@ class StableDiffusionDiffEditPipeline(
             negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
 
         if self.text_encoder is not None:
-            if isinstance(self, LoraLoaderMixin) and USE_PEFT_BACKEND:
+            if isinstance(self, StableDiffusionLoraLoaderMixin) and USE_PEFT_BACKEND:
                 # Retrieve the original scale by scaling back the LoRA layers
                 unscale_lora_layers(self.text_encoder, lora_scale)
 

src/diffusers/pipelines/stable_diffusion_gligen/pipeline_stable_diffusion_gligen.py

@@ -21,7 +21,7 @@ import torch
 from transformers import CLIPFeatureExtractor, CLIPTextModel, CLIPTokenizer
 
 from ...image_processor import VaeImageProcessor
-from ...loaders import LoraLoaderMixin, TextualInversionLoaderMixin
+from ...loaders import StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
 from ...models import AutoencoderKL, UNet2DConditionModel
 from ...models.attention import GatedSelfAttentionDense
 from ...models.lora import adjust_lora_scale_text_encoder
@@ -249,7 +249,7 @@ class StableDiffusionGLIGENPipeline(DiffusionPipeline, StableDiffusionMixin):
         """
         # set lora scale so that monkey patched LoRA
         # function of text encoder can correctly access it
-        if lora_scale is not None and isinstance(self, LoraLoaderMixin):
+        if lora_scale is not None and isinstance(self, StableDiffusionLoraLoaderMixin):
             self._lora_scale = lora_scale
 
             # dynamically adjust the LoRA scale
@@ -382,7 +382,7 @@ class StableDiffusionGLIGENPipeline(DiffusionPipeline, StableDiffusionMixin):
             negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
 
         if self.text_encoder is not None:
-            if isinstance(self, LoraLoaderMixin) and USE_PEFT_BACKEND:
+            if isinstance(self, StableDiffusionLoraLoaderMixin) and USE_PEFT_BACKEND:
                 # Retrieve the original scale by scaling back the LoRA layers
                 unscale_lora_layers(self.text_encoder, lora_scale)
 

src/diffusers/pipelines/stable_diffusion_gligen/pipeline_stable_diffusion_gligen_text_image.py

@@ -27,7 +27,7 @@ from transformers import (
 )
 
 from ...image_processor import VaeImageProcessor
-from ...loaders import LoraLoaderMixin, TextualInversionLoaderMixin
+from ...loaders import StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
 from ...models import AutoencoderKL, UNet2DConditionModel
 from ...models.attention import GatedSelfAttentionDense
 from ...models.lora import adjust_lora_scale_text_encoder
@@ -274,7 +274,7 @@ class StableDiffusionGLIGENTextImagePipeline(DiffusionPipeline, StableDiffusionM
         """
         # set lora scale so that monkey patched LoRA
         # function of text encoder can correctly access it
-        if lora_scale is not None and isinstance(self, LoraLoaderMixin):
+        if lora_scale is not None and isinstance(self, StableDiffusionLoraLoaderMixin):
             self._lora_scale = lora_scale
 
             # dynamically adjust the LoRA scale
@@ -407,7 +407,7 @@ class StableDiffusionGLIGENTextImagePipeline(DiffusionPipeline, StableDiffusionM
             negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
 
         if self.text_encoder is not None:
-            if isinstance(self, LoraLoaderMixin) and USE_PEFT_BACKEND:
+            if isinstance(self, StableDiffusionLoraLoaderMixin) and USE_PEFT_BACKEND:
                 # Retrieve the original scale by scaling back the LoRA layers
                 unscale_lora_layers(self.text_encoder, lora_scale)
 

src/diffusers/pipelines/stable_diffusion_k_diffusion/pipeline_stable_diffusion_k_diffusion.py

@@ -21,7 +21,7 @@ from k_diffusion.external import CompVisDenoiser, CompVisVDenoiser
 from k_diffusion.sampling import BrownianTreeNoiseSampler, get_sigmas_karras
 
 from ...image_processor import VaeImageProcessor
-from ...loaders import LoraLoaderMixin, TextualInversionLoaderMixin
+from ...loaders import StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
 from ...models.lora import adjust_lora_scale_text_encoder
 from ...schedulers import LMSDiscreteScheduler
 from ...utils import USE_PEFT_BACKEND, deprecate, logging, scale_lora_layers, unscale_lora_layers
@@ -48,7 +48,7 @@ class ModelWrapper:
 
 
 class StableDiffusionKDiffusionPipeline(
-    DiffusionPipeline, StableDiffusionMixin, TextualInversionLoaderMixin, LoraLoaderMixin
+    DiffusionPipeline, StableDiffusionMixin, TextualInversionLoaderMixin, StableDiffusionLoraLoaderMixin
 ):
     r"""
     Pipeline for text-to-image generation using Stable Diffusion.
@@ -58,8 +58,8 @@ class StableDiffusionKDiffusionPipeline(
 
     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.StableDiffusionLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+        - [`~loaders.StableDiffusionLoraLoaderMixin.save_lora_weights`] for saving LoRA weights
 
     <Tip warning={true}>
 
@@ -223,7 +223,7 @@ class StableDiffusionKDiffusionPipeline(
         """
         # set lora scale so that monkey patched LoRA
         # function of text encoder can correctly access it
-        if lora_scale is not None and isinstance(self, LoraLoaderMixin):
+        if lora_scale is not None and isinstance(self, StableDiffusionLoraLoaderMixin):
             self._lora_scale = lora_scale
 
             # dynamically adjust the LoRA scale
@@ -356,7 +356,7 @@ class StableDiffusionKDiffusionPipeline(
             negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
 
         if self.text_encoder is not None:
-            if isinstance(self, LoraLoaderMixin) and USE_PEFT_BACKEND:
+            if isinstance(self, StableDiffusionLoraLoaderMixin) and USE_PEFT_BACKEND:
                 # Retrieve the original scale by scaling back the LoRA layers
                 unscale_lora_layers(self.text_encoder, lora_scale)
 

src/diffusers/pipelines/stable_diffusion_ldm3d/pipeline_stable_diffusion_ldm3d.py

@@ -22,7 +22,7 @@ import torch
 from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer, CLIPVisionModelWithProjection
 
 from ...image_processor import PipelineImageInput, VaeImageProcessorLDM3D
-from ...loaders import FromSingleFileMixin, IPAdapterMixin, LoraLoaderMixin, TextualInversionLoaderMixin
+from ...loaders import FromSingleFileMixin, IPAdapterMixin, StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
 from ...models import AutoencoderKL, ImageProjection, UNet2DConditionModel
 from ...models.lora import adjust_lora_scale_text_encoder
 from ...schedulers import KarrasDiffusionSchedulers
@@ -161,7 +161,7 @@ class StableDiffusionLDM3DPipeline(
     StableDiffusionMixin,
     TextualInversionLoaderMixin,
     IPAdapterMixin,
-    LoraLoaderMixin,
+    StableDiffusionLoraLoaderMixin,
     FromSingleFileMixin,
 ):
     r"""
@@ -172,8 +172,8 @@ class StableDiffusionLDM3DPipeline(
 
     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.StableDiffusionLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+        - [`~loaders.StableDiffusionLoraLoaderMixin.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
 
@@ -323,7 +323,7 @@ class StableDiffusionLDM3DPipeline(
         """
         # set lora scale so that monkey patched LoRA
         # function of text encoder can correctly access it
-        if lora_scale is not None and isinstance(self, LoraLoaderMixin):
+        if lora_scale is not None and isinstance(self, StableDiffusionLoraLoaderMixin):
             self._lora_scale = lora_scale
 
             # dynamically adjust the LoRA scale
@@ -456,7 +456,7 @@ class StableDiffusionLDM3DPipeline(
             negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
 
         if self.text_encoder is not None:
-            if isinstance(self, LoraLoaderMixin) and USE_PEFT_BACKEND:
+            if isinstance(self, StableDiffusionLoraLoaderMixin) and USE_PEFT_BACKEND:
                 # Retrieve the original scale by scaling back the LoRA layers
                 unscale_lora_layers(self.text_encoder, lora_scale)
 

src/diffusers/pipelines/stable_diffusion_panorama/pipeline_stable_diffusion_panorama.py

@@ -19,7 +19,7 @@ import torch
 from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer, CLIPVisionModelWithProjection
 
 from ...image_processor import PipelineImageInput, VaeImageProcessor
-from ...loaders import IPAdapterMixin, LoraLoaderMixin, TextualInversionLoaderMixin
+from ...loaders import IPAdapterMixin, StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
 from ...models import AutoencoderKL, ImageProjection, UNet2DConditionModel
 from ...models.lora import adjust_lora_scale_text_encoder
 from ...schedulers import DDIMScheduler
@@ -135,7 +135,11 @@ def retrieve_timesteps(
 
 
 class StableDiffusionPanoramaPipeline(
-    DiffusionPipeline, StableDiffusionMixin, TextualInversionLoaderMixin, LoraLoaderMixin, IPAdapterMixin
+    DiffusionPipeline,
+    StableDiffusionMixin,
+    TextualInversionLoaderMixin,
+    StableDiffusionLoraLoaderMixin,
+    IPAdapterMixin,
 ):
     r"""
     Pipeline for text-to-image generation using MultiDiffusion.
@@ -145,8 +149,8 @@ class StableDiffusionPanoramaPipeline(
 
     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.StableDiffusionLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+        - [`~loaders.StableDiffusionLoraLoaderMixin.save_lora_weights`] for saving LoRA weights
         - [`~loaders.IPAdapterMixin.load_ip_adapter`] for loading IP Adapters
 
     Args:
@@ -295,7 +299,7 @@ class StableDiffusionPanoramaPipeline(
         """
         # set lora scale so that monkey patched LoRA
         # function of text encoder can correctly access it
-        if lora_scale is not None and isinstance(self, LoraLoaderMixin):
+        if lora_scale is not None and isinstance(self, StableDiffusionLoraLoaderMixin):
             self._lora_scale = lora_scale
 
             # dynamically adjust the LoRA scale
@@ -428,7 +432,7 @@ class StableDiffusionPanoramaPipeline(
             negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
 
         if self.text_encoder is not None:
-            if isinstance(self, LoraLoaderMixin) and USE_PEFT_BACKEND:
+            if isinstance(self, StableDiffusionLoraLoaderMixin) and USE_PEFT_BACKEND:
                 # Retrieve the original scale by scaling back the LoRA layers
                 unscale_lora_layers(self.text_encoder, lora_scale)
 

src/diffusers/pipelines/stable_diffusion_sag/pipeline_stable_diffusion_sag.py

@@ -20,7 +20,7 @@ import torch.nn.functional as F
 from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer, CLIPVisionModelWithProjection
 
 from ...image_processor import PipelineImageInput, VaeImageProcessor
-from ...loaders import IPAdapterMixin, LoraLoaderMixin, TextualInversionLoaderMixin
+from ...loaders import IPAdapterMixin, StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
 from ...models import AutoencoderKL, ImageProjection, UNet2DConditionModel
 from ...models.lora import adjust_lora_scale_text_encoder
 from ...schedulers import KarrasDiffusionSchedulers
@@ -238,7 +238,7 @@ class StableDiffusionSAGPipeline(DiffusionPipeline, StableDiffusionMixin, Textua
         """
         # set lora scale so that monkey patched LoRA
         # function of text encoder can correctly access it
-        if lora_scale is not None and isinstance(self, LoraLoaderMixin):
+        if lora_scale is not None and isinstance(self, StableDiffusionLoraLoaderMixin):
             self._lora_scale = lora_scale
 
             # dynamically adjust the LoRA scale
@@ -371,7 +371,7 @@ class StableDiffusionSAGPipeline(DiffusionPipeline, StableDiffusionMixin, Textua
             negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
 
         if self.text_encoder is not None:
-            if isinstance(self, LoraLoaderMixin) and USE_PEFT_BACKEND:
+            if isinstance(self, StableDiffusionLoraLoaderMixin) and USE_PEFT_BACKEND:
                 # Retrieve the original scale by scaling back the LoRA layers
                 unscale_lora_layers(self.text_encoder, lora_scale)
 

src/diffusers/pipelines/t2i_adapter/pipeline_stable_diffusion_adapter.py

@@ -22,7 +22,7 @@ import torch
 from transformers import CLIPFeatureExtractor, CLIPTextModel, CLIPTokenizer
 
 from ...image_processor import VaeImageProcessor
-from ...loaders import LoraLoaderMixin, TextualInversionLoaderMixin
+from ...loaders import StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
 from ...models import AutoencoderKL, MultiAdapter, T2IAdapter, UNet2DConditionModel
 from ...models.lora import adjust_lora_scale_text_encoder
 from ...schedulers import KarrasDiffusionSchedulers
@@ -340,7 +340,7 @@ class StableDiffusionAdapterPipeline(DiffusionPipeline, StableDiffusionMixin):
         """
         # set lora scale so that monkey patched LoRA
         # function of text encoder can correctly access it
-        if lora_scale is not None and isinstance(self, LoraLoaderMixin):
+        if lora_scale is not None and isinstance(self, StableDiffusionLoraLoaderMixin):
             self._lora_scale = lora_scale
 
             # dynamically adjust the LoRA scale
@@ -473,7 +473,7 @@ class StableDiffusionAdapterPipeline(DiffusionPipeline, StableDiffusionMixin):
             negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
 
         if self.text_encoder is not None:
-            if isinstance(self, LoraLoaderMixin) and USE_PEFT_BACKEND:
+            if isinstance(self, StableDiffusionLoraLoaderMixin) and USE_PEFT_BACKEND:
                 # Retrieve the original scale by scaling back the LoRA layers
                 unscale_lora_layers(self.text_encoder, lora_scale)
 

src/diffusers/pipelines/text_to_video_synthesis/pipeline_text_to_video_synth.py

@@ -18,7 +18,7 @@ from typing import Any, Callable, Dict, List, Optional, Union
 import torch
 from transformers import CLIPTextModel, CLIPTokenizer
 
-from ...loaders import LoraLoaderMixin, TextualInversionLoaderMixin
+from ...loaders import StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
 from ...models import AutoencoderKL, UNet3DConditionModel
 from ...models.lora import adjust_lora_scale_text_encoder
 from ...schedulers import KarrasDiffusionSchedulers
@@ -58,7 +58,9 @@ EXAMPLE_DOC_STRING = """
 """
 
 
-class TextToVideoSDPipeline(DiffusionPipeline, StableDiffusionMixin, TextualInversionLoaderMixin, LoraLoaderMixin):
+class TextToVideoSDPipeline(
+    DiffusionPipeline, StableDiffusionMixin, TextualInversionLoaderMixin, StableDiffusionLoraLoaderMixin
+):
     r"""
     Pipeline for text-to-video generation.
 
@@ -67,8 +69,8 @@ class TextToVideoSDPipeline(DiffusionPipeline, StableDiffusionMixin, TextualInve
 
     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.StableDiffusionLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+        - [`~loaders.StableDiffusionLoraLoaderMixin.save_lora_weights`] for saving LoRA weights
 
     Args:
         vae ([`AutoencoderKL`]):
@@ -183,7 +185,7 @@ class TextToVideoSDPipeline(DiffusionPipeline, StableDiffusionMixin, TextualInve
         """
         # set lora scale so that monkey patched LoRA
         # function of text encoder can correctly access it
-        if lora_scale is not None and isinstance(self, LoraLoaderMixin):
+        if lora_scale is not None and isinstance(self, StableDiffusionLoraLoaderMixin):
             self._lora_scale = lora_scale
 
             # dynamically adjust the LoRA scale
@@ -316,7 +318,7 @@ class TextToVideoSDPipeline(DiffusionPipeline, StableDiffusionMixin, TextualInve
             negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
 
         if self.text_encoder is not None:
-            if isinstance(self, LoraLoaderMixin) and USE_PEFT_BACKEND:
+            if isinstance(self, StableDiffusionLoraLoaderMixin) and USE_PEFT_BACKEND:
                 # Retrieve the original scale by scaling back the LoRA layers
                 unscale_lora_layers(self.text_encoder, lora_scale)
 

src/diffusers/pipelines/text_to_video_synthesis/pipeline_text_to_video_synth_img2img.py

@@ -19,7 +19,7 @@ import numpy as np
 import torch
 from transformers import CLIPTextModel, CLIPTokenizer
 
-from ...loaders import LoraLoaderMixin, TextualInversionLoaderMixin
+from ...loaders import StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
 from ...models import AutoencoderKL, UNet3DConditionModel
 from ...models.lora import adjust_lora_scale_text_encoder
 from ...schedulers import KarrasDiffusionSchedulers
@@ -93,7 +93,9 @@ def retrieve_latents(
         raise AttributeError("Could not access latents of provided encoder_output")
 
 
-class VideoToVideoSDPipeline(DiffusionPipeline, StableDiffusionMixin, TextualInversionLoaderMixin, LoraLoaderMixin):
+class VideoToVideoSDPipeline(
+    DiffusionPipeline, StableDiffusionMixin, TextualInversionLoaderMixin, StableDiffusionLoraLoaderMixin
+):
     r"""
     Pipeline for text-guided video-to-video generation.
 
@@ -102,8 +104,8 @@ class VideoToVideoSDPipeline(DiffusionPipeline, StableDiffusionMixin, TextualInv
 
     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.StableDiffusionLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+        - [`~loaders.StableDiffusionLoraLoaderMixin.save_lora_weights`] for saving LoRA weights
 
     Args:
         vae ([`AutoencoderKL`]):
@@ -218,7 +220,7 @@ class VideoToVideoSDPipeline(DiffusionPipeline, StableDiffusionMixin, TextualInv
         """
         # set lora scale so that monkey patched LoRA
         # function of text encoder can correctly access it
-        if lora_scale is not None and isinstance(self, LoraLoaderMixin):
+        if lora_scale is not None and isinstance(self, StableDiffusionLoraLoaderMixin):
             self._lora_scale = lora_scale
 
             # dynamically adjust the LoRA scale
@@ -351,7 +353,7 @@ class VideoToVideoSDPipeline(DiffusionPipeline, StableDiffusionMixin, TextualInv
             negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
 
         if self.text_encoder is not None:
-            if isinstance(self, LoraLoaderMixin) and USE_PEFT_BACKEND:
+            if isinstance(self, StableDiffusionLoraLoaderMixin) and USE_PEFT_BACKEND:
                 # Retrieve the original scale by scaling back the LoRA layers
                 unscale_lora_layers(self.text_encoder, lora_scale)