[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/deepfloyd_if/pipeline_if_inpainting.py

@@ -9,7 +9,7 @@ import PIL.Image
 import torch
 from transformers import CLIPImageProcessor, T5EncoderModel, T5Tokenizer
 
-from ...loaders import LoraLoaderMixin
+from ...loaders import StableDiffusionLoraLoaderMixin
 from ...models import UNet2DConditionModel
 from ...schedulers import DDPMScheduler
 from ...utils import (
@@ -111,7 +111,7 @@ EXAMPLE_DOC_STRING = """
 """
 
 
-class IFInpaintingPipeline(DiffusionPipeline, LoraLoaderMixin):
+class IFInpaintingPipeline(DiffusionPipeline, StableDiffusionLoraLoaderMixin):
     tokenizer: T5Tokenizer
     text_encoder: T5EncoderModel
 

src/diffusers/pipelines/deepfloyd_if/pipeline_if_inpainting_superresolution.py

@@ -10,7 +10,7 @@ import torch
 import torch.nn.functional as F
 from transformers import CLIPImageProcessor, T5EncoderModel, T5Tokenizer
 
-from ...loaders import LoraLoaderMixin
+from ...loaders import StableDiffusionLoraLoaderMixin
 from ...models import UNet2DConditionModel
 from ...schedulers import DDPMScheduler
 from ...utils import (
@@ -113,7 +113,7 @@ EXAMPLE_DOC_STRING = """
     """
 
 
-class IFInpaintingSuperResolutionPipeline(DiffusionPipeline, LoraLoaderMixin):
+class IFInpaintingSuperResolutionPipeline(DiffusionPipeline, StableDiffusionLoraLoaderMixin):
     tokenizer: T5Tokenizer
     text_encoder: T5EncoderModel
 

src/diffusers/pipelines/deepfloyd_if/pipeline_if_superresolution.py

@@ -10,7 +10,7 @@ import torch
 import torch.nn.functional as F
 from transformers import CLIPImageProcessor, T5EncoderModel, T5Tokenizer
 
-from ...loaders import LoraLoaderMixin
+from ...loaders import StableDiffusionLoraLoaderMixin
 from ...models import UNet2DConditionModel
 from ...schedulers import DDPMScheduler
 from ...utils import (
@@ -69,7 +69,7 @@ EXAMPLE_DOC_STRING = """
 """
 
 
-class IFSuperResolutionPipeline(DiffusionPipeline, LoraLoaderMixin):
+class IFSuperResolutionPipeline(DiffusionPipeline, StableDiffusionLoraLoaderMixin):
     tokenizer: T5Tokenizer
     text_encoder: T5EncoderModel
 

src/diffusers/pipelines/deprecated/alt_diffusion/pipeline_alt_diffusion.py

@@ -21,7 +21,12 @@ from transformers import CLIPImageProcessor, CLIPVisionModelWithProjection, XLMR
 
 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
@@ -137,7 +142,7 @@ class AltDiffusionPipeline(
     DiffusionPipeline,
     StableDiffusionMixin,
     TextualInversionLoaderMixin,
-    LoraLoaderMixin,
+    StableDiffusionLoraLoaderMixin,
     IPAdapterMixin,
     FromSingleFileMixin,
 ):
@@ -149,8 +154,8 @@ class AltDiffusionPipeline(
 
     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
 
@@ -346,7 +351,7 @@ class AltDiffusionPipeline(
         """
         # 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
@@ -478,7 +483,7 @@ class AltDiffusionPipeline(
             negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1)
             negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
 
-        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/deprecated/alt_diffusion/pipeline_alt_diffusion_img2img.py

@@ -23,7 +23,12 @@ from transformers import CLIPImageProcessor, CLIPVisionModelWithProjection, XLMR
 
 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
@@ -178,7 +183,7 @@ class AltDiffusionImg2ImgPipeline(
     StableDiffusionMixin,
     TextualInversionLoaderMixin,
     IPAdapterMixin,
-    LoraLoaderMixin,
+    StableDiffusionLoraLoaderMixin,
     FromSingleFileMixin,
 ):
     r"""
@@ -189,8 +194,8 @@ class AltDiffusionImg2ImgPipeline(
 
     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
 
@@ -386,7 +391,7 @@ class AltDiffusionImg2ImgPipeline(
         """
         # 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 +523,7 @@ class AltDiffusionImg2ImgPipeline(
             negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1)
             negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
 
-        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/deprecated/stable_diffusion_variants/pipeline_cycle_diffusion.py

@@ -23,7 +23,7 @@ from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer
 
 from ....configuration_utils import FrozenDict
 from ....image_processor import PipelineImageInput, 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 DDIMScheduler
@@ -136,7 +136,7 @@ def compute_noise(scheduler, prev_latents, latents, timestep, noise_pred, eta):
     return noise
 
 
-class CycleDiffusionPipeline(DiffusionPipeline, TextualInversionLoaderMixin, LoraLoaderMixin):
+class CycleDiffusionPipeline(DiffusionPipeline, TextualInversionLoaderMixin, StableDiffusionLoraLoaderMixin):
     r"""
     Pipeline for text-guided image to image generation using Stable Diffusion.
 
@@ -145,8 +145,8 @@ class CycleDiffusionPipeline(DiffusionPipeline, TextualInversionLoaderMixin, Lor
 
     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`]):
@@ -324,7 +324,7 @@ class CycleDiffusionPipeline(DiffusionPipeline, TextualInversionLoaderMixin, Lor
         """
         # 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
@@ -457,7 +457,7 @@ class CycleDiffusionPipeline(DiffusionPipeline, TextualInversionLoaderMixin, Lor
             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/deprecated/stable_diffusion_variants/pipeline_stable_diffusion_inpaint_legacy.py

@@ -23,7 +23,7 @@ from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer
 
 from ....configuration_utils import FrozenDict
 from ....image_processor import VaeImageProcessor
-from ....loaders import FromSingleFileMixin, LoraLoaderMixin, TextualInversionLoaderMixin
+from ....loaders import FromSingleFileMixin, StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
 from ....models import AutoencoderKL, UNet2DConditionModel
 from ....models.lora import adjust_lora_scale_text_encoder
 from ....schedulers import KarrasDiffusionSchedulers
@@ -79,7 +79,7 @@ def preprocess_mask(mask, batch_size, scale_factor=8):
 
 
 class StableDiffusionInpaintPipelineLegacy(
-    DiffusionPipeline, TextualInversionLoaderMixin, LoraLoaderMixin, FromSingleFileMixin
+    DiffusionPipeline, TextualInversionLoaderMixin, StableDiffusionLoraLoaderMixin, FromSingleFileMixin
 ):
     r"""
     Pipeline for text-guided image inpainting using Stable Diffusion. *This is an experimental feature*.
@@ -89,11 +89,11 @@ class StableDiffusionInpaintPipelineLegacy(
 
     In addition the pipeline inherits the following loading methods:
         - *Textual-Inversion*: [`loaders.TextualInversionLoaderMixin.load_textual_inversion`]
-        - *LoRA*: [`loaders.LoraLoaderMixin.load_lora_weights`]
+        - *LoRA*: [`loaders.StableDiffusionLoraLoaderMixin.load_lora_weights`]
         - *Ckpt*: [`loaders.FromSingleFileMixin.from_single_file`]
 
     as well as the following saving methods:
-        - *LoRA*: [`loaders.LoraLoaderMixin.save_lora_weights`]
+        - *LoRA*: [`loaders.StableDiffusionLoraLoaderMixin.save_lora_weights`]
 
     Args:
         vae ([`AutoencoderKL`]):
@@ -294,7 +294,7 @@ class StableDiffusionInpaintPipelineLegacy(
         """
         # 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
@@ -427,7 +427,7 @@ class StableDiffusionInpaintPipelineLegacy(
             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/deprecated/stable_diffusion_variants/pipeline_stable_diffusion_model_editing.py

@@ -19,7 +19,7 @@ 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.lora import adjust_lora_scale_text_encoder
 from ....schedulers import PNDMScheduler
@@ -37,7 +37,7 @@ AUGS_CONST = ["A photo of ", "An image of ", "A picture of "]
 
 
 class StableDiffusionModelEditingPipeline(
-    DiffusionPipeline, StableDiffusionMixin, TextualInversionLoaderMixin, LoraLoaderMixin
+    DiffusionPipeline, StableDiffusionMixin, TextualInversionLoaderMixin, StableDiffusionLoraLoaderMixin
 ):
     r"""
     Pipeline for text-to-image model editing.
@@ -47,8 +47,8 @@ class StableDiffusionModelEditingPipeline(
 
     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`]):
@@ -232,7 +232,7 @@ class StableDiffusionModelEditingPipeline(
         """
         # 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
@@ -365,7 +365,7 @@ class StableDiffusionModelEditingPipeline(
             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/deprecated/stable_diffusion_variants/pipeline_stable_diffusion_paradigms.py

@@ -19,7 +19,7 @@ import torch
 from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer
 
 from ....image_processor import VaeImageProcessor
-from ....loaders import FromSingleFileMixin, LoraLoaderMixin, TextualInversionLoaderMixin
+from ....loaders import FromSingleFileMixin, StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
 from ....models import AutoencoderKL, UNet2DConditionModel
 from ....models.lora import adjust_lora_scale_text_encoder
 from ....schedulers import KarrasDiffusionSchedulers
@@ -63,7 +63,11 @@ EXAMPLE_DOC_STRING = """
 
 
 class StableDiffusionParadigmsPipeline(
-    DiffusionPipeline, StableDiffusionMixin, TextualInversionLoaderMixin, LoraLoaderMixin, FromSingleFileMixin
+    DiffusionPipeline,
+    StableDiffusionMixin,
+    TextualInversionLoaderMixin,
+    StableDiffusionLoraLoaderMixin,
+    FromSingleFileMixin,
 ):
     r"""
     Pipeline for text-to-image generation using a parallelized version of Stable Diffusion.
@@ -73,8 +77,8 @@ class StableDiffusionParadigmsPipeline(
 
     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:
@@ -223,7 +227,7 @@ class StableDiffusionParadigmsPipeline(
         """
         # 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 +360,7 @@ class StableDiffusionParadigmsPipeline(
             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/deprecated/stable_diffusion_variants/pipeline_stable_diffusion_pix2pix_zero.py

@@ -29,7 +29,7 @@ from transformers import (
 )
 
 from ....image_processor import PipelineImageInput, 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
@@ -446,7 +446,7 @@ class StableDiffusionPix2PixZeroPipeline(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
@@ -579,7 +579,7 @@ class StableDiffusionPix2PixZeroPipeline(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/kandinsky3/pipeline_kandinsky3.py

@@ -3,7 +3,7 @@ from typing import Callable, Dict, List, Optional, Union
 import torch
 from transformers import T5EncoderModel, T5Tokenizer
 
-from ...loaders import LoraLoaderMixin
+from ...loaders import StableDiffusionLoraLoaderMixin
 from ...models import Kandinsky3UNet, VQModel
 from ...schedulers import DDPMScheduler
 from ...utils import (
@@ -47,7 +47,7 @@ def downscale_height_and_width(height, width, scale_factor=8):
     return new_height * scale_factor, new_width * scale_factor
 
 
-class Kandinsky3Pipeline(DiffusionPipeline, LoraLoaderMixin):
+class Kandinsky3Pipeline(DiffusionPipeline, StableDiffusionLoraLoaderMixin):
     model_cpu_offload_seq = "text_encoder->unet->movq"
     _callback_tensor_inputs = [
         "latents",

src/diffusers/pipelines/kandinsky3/pipeline_kandinsky3_img2img.py

@@ -7,7 +7,7 @@ import PIL.Image
 import torch
 from transformers import T5EncoderModel, T5Tokenizer
 
-from ...loaders import LoraLoaderMixin
+from ...loaders import StableDiffusionLoraLoaderMixin
 from ...models import Kandinsky3UNet, VQModel
 from ...schedulers import DDPMScheduler
 from ...utils import (
@@ -62,7 +62,7 @@ def prepare_image(pil_image):
     return image
 
 
-class Kandinsky3Img2ImgPipeline(DiffusionPipeline, LoraLoaderMixin):
+class Kandinsky3Img2ImgPipeline(DiffusionPipeline, StableDiffusionLoraLoaderMixin):
     model_cpu_offload_seq = "text_encoder->movq->unet->movq"
     _callback_tensor_inputs = [
         "latents",

src/diffusers/pipelines/latent_consistency_models/pipeline_latent_consistency_img2img.py

@@ -23,7 +23,7 @@ import torch
 from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer, CLIPVisionModelWithProjection
 
 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 LCMScheduler
@@ -148,7 +148,7 @@ class LatentConsistencyModelImg2ImgPipeline(
     StableDiffusionMixin,
     TextualInversionLoaderMixin,
     IPAdapterMixin,
-    LoraLoaderMixin,
+    StableDiffusionLoraLoaderMixin,
     FromSingleFileMixin,
 ):
     r"""
@@ -159,8 +159,8 @@ class LatentConsistencyModelImg2ImgPipeline(
 
     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
 
@@ -273,7 +273,7 @@ class LatentConsistencyModelImg2ImgPipeline(
         """
         # 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
@@ -406,7 +406,7 @@ class LatentConsistencyModelImg2ImgPipeline(
             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/latent_consistency_models/pipeline_latent_consistency_text2img.py

@@ -22,7 +22,7 @@ import torch
 from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer, CLIPVisionModelWithProjection
 
 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 LCMScheduler
@@ -126,7 +126,7 @@ class LatentConsistencyModelPipeline(
     StableDiffusionMixin,
     TextualInversionLoaderMixin,
     IPAdapterMixin,
-    LoraLoaderMixin,
+    StableDiffusionLoraLoaderMixin,
     FromSingleFileMixin,
 ):
     r"""
@@ -137,8 +137,8 @@ class LatentConsistencyModelPipeline(
 
     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
 
@@ -257,7 +257,7 @@ class LatentConsistencyModelPipeline(
         """
         # 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
@@ -390,7 +390,7 @@ class LatentConsistencyModelPipeline(
             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/ledits_pp/pipeline_leditspp_stable_diffusion.py

@@ -10,7 +10,7 @@ from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer
 
 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, UNet2DConditionModel
 from ...models.attention_processor import Attention, AttnProcessor
 from ...models.lora import adjust_lora_scale_text_encoder
@@ -248,7 +248,7 @@ def rescale_noise_cfg(noise_cfg, noise_pred_text, guidance_rescale=0.0):
 
 
 class LEditsPPPipelineStableDiffusion(
-    DiffusionPipeline, TextualInversionLoaderMixin, LoraLoaderMixin, IPAdapterMixin, FromSingleFileMixin
+    DiffusionPipeline, TextualInversionLoaderMixin, StableDiffusionLoraLoaderMixin, IPAdapterMixin, FromSingleFileMixin
 ):
     """
     Pipeline for textual image editing using LEDits++ with Stable Diffusion.
@@ -538,7 +538,7 @@ class LEditsPPPipelineStableDiffusion(
         """
         # 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
@@ -676,7 +676,7 @@ class LEditsPPPipelineStableDiffusion(
         negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1)
         negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
 
-        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/pag/pipeline_pag_controlnet_sd.py

@@ -24,7 +24,7 @@ from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer, CLIPV
 
 from ...callbacks import MultiPipelineCallbacks, PipelineCallback
 from ...image_processor import PipelineImageInput, VaeImageProcessor
-from ...loaders import FromSingleFileMixin, IPAdapterMixin, LoraLoaderMixin, TextualInversionLoaderMixin
+from ...loaders import FromSingleFileMixin, IPAdapterMixin, StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
 from ...models import AutoencoderKL, ControlNetModel, ImageProjection, UNet2DConditionModel
 from ...models.lora import adjust_lora_scale_text_encoder
 from ...schedulers import KarrasDiffusionSchedulers
@@ -159,7 +159,7 @@ class StableDiffusionControlNetPAGPipeline(
     DiffusionPipeline,
     StableDiffusionMixin,
     TextualInversionLoaderMixin,
-    LoraLoaderMixin,
+    StableDiffusionLoraLoaderMixin,
     IPAdapterMixin,
     FromSingleFileMixin,
     PAGMixin,
@@ -172,8 +172,8 @@ class StableDiffusionControlNetPAGPipeline(
 
     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
 
@@ -305,7 +305,7 @@ class StableDiffusionControlNetPAGPipeline(
         """
         # 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
@@ -438,7 +438,7 @@ class StableDiffusionControlNetPAGPipeline(
             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/pag/pipeline_pag_sd.py

@@ -20,7 +20,7 @@ from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer, CLIPV
 
 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
@@ -137,7 +137,7 @@ class StableDiffusionPAGPipeline(
     DiffusionPipeline,
     StableDiffusionMixin,
     TextualInversionLoaderMixin,
-    LoraLoaderMixin,
+    StableDiffusionLoraLoaderMixin,
     IPAdapterMixin,
     FromSingleFileMixin,
     PAGMixin,
@@ -150,8 +150,8 @@ class StableDiffusionPAGPipeline(
 
     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
 
@@ -319,7 +319,7 @@ class StableDiffusionPAGPipeline(
         """
         # 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
@@ -452,7 +452,7 @@ class StableDiffusionPAGPipeline(
             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/pia/pipeline_pia.py

@@ -22,7 +22,7 @@ import torch
 from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer, CLIPVisionModelWithProjection
 
 from ...image_processor import PipelineImageInput
-from ...loaders import FromSingleFileMixin, IPAdapterMixin, LoraLoaderMixin, TextualInversionLoaderMixin
+from ...loaders import FromSingleFileMixin, IPAdapterMixin, StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
 from ...models import AutoencoderKL, ImageProjection, UNet2DConditionModel, UNetMotionModel
 from ...models.lora import adjust_lora_scale_text_encoder
 from ...models.unets.unet_motion_model import MotionAdapter
@@ -128,7 +128,7 @@ class PIAPipeline(
     StableDiffusionMixin,
     TextualInversionLoaderMixin,
     IPAdapterMixin,
-    LoraLoaderMixin,
+    StableDiffusionLoraLoaderMixin,
     FromSingleFileMixin,
     FreeInitMixin,
 ):
@@ -140,8 +140,8 @@ class PIAPipeline(
 
     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:
@@ -243,7 +243,7 @@ class PIAPipeline(
         """
         # 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 +376,7 @@ class PIAPipeline(
             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.py

@@ -21,7 +21,7 @@ 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
@@ -133,7 +133,7 @@ class StableDiffusionPipeline(
     DiffusionPipeline,
     StableDiffusionMixin,
     TextualInversionLoaderMixin,
-    LoraLoaderMixin,
+    StableDiffusionLoraLoaderMixin,
     IPAdapterMixin,
     FromSingleFileMixin,
 ):
@@ -145,8 +145,8 @@ class StableDiffusionPipeline(
 
     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
 
@@ -342,7 +342,7 @@ class StableDiffusionPipeline(
         """
         # 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
@@ -475,7 +475,7 @@ class StableDiffusionPipeline(
             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_depth2img.py

@@ -24,7 +24,7 @@ from transformers import CLIPTextModel, CLIPTokenizer, DPTFeatureExtractor, DPTF
 
 from ...configuration_utils import FrozenDict
 from ...image_processor import PipelineImageInput, 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 KarrasDiffusionSchedulers
@@ -74,7 +74,7 @@ def preprocess(image):
     return image
 
 
-class StableDiffusionDepth2ImgPipeline(DiffusionPipeline, TextualInversionLoaderMixin, LoraLoaderMixin):
+class StableDiffusionDepth2ImgPipeline(DiffusionPipeline, TextualInversionLoaderMixin, StableDiffusionLoraLoaderMixin):
     r"""
     Pipeline for text-guided depth-based image-to-image generation using Stable Diffusion.
 
@@ -83,8 +83,8 @@ class StableDiffusionDepth2ImgPipeline(DiffusionPipeline, TextualInversionLoader
 
     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`]):
@@ -225,7 +225,7 @@ class StableDiffusionDepth2ImgPipeline(DiffusionPipeline, TextualInversionLoader
         """
         # 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
@@ -358,7 +358,7 @@ class StableDiffusionDepth2ImgPipeline(DiffusionPipeline, TextualInversionLoader
             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)