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

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

@@ -19,13 +19,14 @@ import torch
 from packaging import version
 from transformers import CLIPImageProcessor, CLIPVisionModelWithProjection, XLMRobertaTokenizer
 
-from ...configuration_utils import FrozenDict
-from ...image_processor import PipelineImageInput, VaeImageProcessor
-from ...loaders import FromSingleFileMixin, IPAdapterMixin, LoraLoaderMixin, TextualInversionLoaderMixin
-from ...models import AutoencoderKL, UNet2DConditionModel
-from ...models.lora import adjust_lora_scale_text_encoder
-from ...schedulers import KarrasDiffusionSchedulers
-from ...utils import (
+from ....configuration_utils import FrozenDict
+from ....image_processor import PipelineImageInput, VaeImageProcessor
+from ....loaders import FromSingleFileMixin, IPAdapterMixin, LoraLoaderMixin, TextualInversionLoaderMixin
+from ....models import AutoencoderKL, ImageProjection, UNet2DConditionModel
+from ....models.attention_processor import FusedAttnProcessor2_0
+from ....models.lora import adjust_lora_scale_text_encoder
+from ....schedulers import KarrasDiffusionSchedulers
+from ....utils import (
     USE_PEFT_BACKEND,
     deprecate,
     logging,
@@ -33,9 +34,9 @@ from ...utils import (
     scale_lora_layers,
     unscale_lora_layers,
 )
-from ...utils.torch_utils import randn_tensor
-from ..pipeline_utils import DiffusionPipeline
-from ..stable_diffusion.safety_checker import StableDiffusionSafetyChecker
+from ....utils.torch_utils import randn_tensor
+from ...pipeline_utils import DiffusionPipeline
+from ...stable_diffusion.safety_checker import StableDiffusionSafetyChecker
 from .modeling_roberta_series import RobertaSeriesModelWithTransformation
 from .pipeline_output import AltDiffusionPipelineOutput
 
@@ -118,7 +119,6 @@ def retrieve_timesteps(
     return timesteps, num_inference_steps
 
 
-# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline with Stable->Alt, CLIPTextModel->RobertaSeriesModelWithTransformation, CLIPTokenizer->XLMRobertaTokenizer, AltDiffusionSafetyChecker->StableDiffusionSafetyChecker
 class AltDiffusionPipeline(
     DiffusionPipeline, TextualInversionLoaderMixin, LoraLoaderMixin, IPAdapterMixin, FromSingleFileMixin
 ):
@@ -155,7 +155,7 @@ class AltDiffusionPipeline(
             A `CLIPImageProcessor` to extract features from generated images; used as inputs to the `safety_checker`.
     """
 
-    model_cpu_offload_seq = "text_encoder->unet->vae"
+    model_cpu_offload_seq = "text_encoder->image_encoder->unet->vae"
     _optional_components = ["safety_checker", "feature_extractor", "image_encoder"]
     _exclude_from_cpu_offload = ["safety_checker"]
     _callback_tensor_inputs = ["latents", "prompt_embeds", "negative_prompt_embeds"]
@@ -494,18 +494,29 @@ class AltDiffusionPipeline(
 
         return prompt_embeds, negative_prompt_embeds
 
-    def encode_image(self, image, device, num_images_per_prompt):
+    def encode_image(self, image, device, num_images_per_prompt, output_hidden_states=None):
         dtype = next(self.image_encoder.parameters()).dtype
 
         if not isinstance(image, torch.Tensor):
             image = self.feature_extractor(image, return_tensors="pt").pixel_values
 
         image = image.to(device=device, dtype=dtype)
-        image_embeds = self.image_encoder(image).image_embeds
-        image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0)
+        if output_hidden_states:
+            image_enc_hidden_states = self.image_encoder(image, output_hidden_states=True).hidden_states[-2]
+            image_enc_hidden_states = image_enc_hidden_states.repeat_interleave(num_images_per_prompt, dim=0)
+            uncond_image_enc_hidden_states = self.image_encoder(
+                torch.zeros_like(image), output_hidden_states=True
+            ).hidden_states[-2]
+            uncond_image_enc_hidden_states = uncond_image_enc_hidden_states.repeat_interleave(
+                num_images_per_prompt, dim=0
+            )
+            return image_enc_hidden_states, uncond_image_enc_hidden_states
+        else:
+            image_embeds = self.image_encoder(image).image_embeds
+            image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0)
+            uncond_image_embeds = torch.zeros_like(image_embeds)
 
-        uncond_image_embeds = torch.zeros_like(image_embeds)
-        return image_embeds, uncond_image_embeds
+            return image_embeds, uncond_image_embeds
 
     def run_safety_checker(self, image, device, dtype):
         if self.safety_checker is None:
@@ -644,6 +655,65 @@ class AltDiffusionPipeline(
         """Disables the FreeU mechanism if enabled."""
         self.unet.disable_freeu()
 
+    def fuse_qkv_projections(self, unet: bool = True, vae: bool = True):
+        """
+        Enables fused QKV projections. For self-attention modules, all projection matrices (i.e., query,
+        key, value) are fused. For cross-attention modules, key and value projection matrices are fused.
+
+        <Tip warning={true}>
+
+        This API is 🧪 experimental.
+
+        </Tip>
+
+        Args:
+            unet (`bool`, defaults to `True`): To apply fusion on the UNet.
+            vae (`bool`, defaults to `True`): To apply fusion on the VAE.
+        """
+        self.fusing_unet = False
+        self.fusing_vae = False
+
+        if unet:
+            self.fusing_unet = True
+            self.unet.fuse_qkv_projections()
+            self.unet.set_attn_processor(FusedAttnProcessor2_0())
+
+        if vae:
+            if not isinstance(self.vae, AutoencoderKL):
+                raise ValueError("`fuse_qkv_projections()` is only supported for the VAE of type `AutoencoderKL`.")
+
+            self.fusing_vae = True
+            self.vae.fuse_qkv_projections()
+            self.vae.set_attn_processor(FusedAttnProcessor2_0())
+
+    def unfuse_qkv_projections(self, unet: bool = True, vae: bool = True):
+        """Disable QKV projection fusion if enabled.
+
+        <Tip warning={true}>
+
+        This API is 🧪 experimental.
+
+        </Tip>
+
+        Args:
+            unet (`bool`, defaults to `True`): To apply fusion on the UNet.
+            vae (`bool`, defaults to `True`): To apply fusion on the VAE.
+
+        """
+        if unet:
+            if not self.fusing_unet:
+                logger.warning("The UNet was not initially fused for QKV projections. Doing nothing.")
+            else:
+                self.unet.unfuse_qkv_projections()
+                self.fusing_unet = False
+
+        if vae:
+            if not self.fusing_vae:
+                logger.warning("The VAE was not initially fused for QKV projections. Doing nothing.")
+            else:
+                self.vae.unfuse_qkv_projections()
+                self.fusing_vae = False
+
     def get_guidance_scale_embedding(self, w, embedding_dim=512, dtype=torch.float32):
         """
         See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298
@@ -875,7 +945,10 @@ class AltDiffusionPipeline(
             prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds])
 
         if ip_adapter_image is not None:
-            image_embeds, negative_image_embeds = self.encode_image(ip_adapter_image, device, num_images_per_prompt)
+            output_hidden_state = False if isinstance(self.unet.encoder_hid_proj, ImageProjection) else True
+            image_embeds, negative_image_embeds = self.encode_image(
+                ip_adapter_image, device, num_images_per_prompt, output_hidden_state
+            )
             if self.do_classifier_free_guidance:
                 image_embeds = torch.cat([negative_image_embeds, image_embeds])
 

diffusers/pipelines/{alt_diffusion → deprecated/alt_diffusion}/pipeline_alt_diffusion_img2img.py

@@ -21,13 +21,14 @@ import torch
 from packaging import version
 from transformers import CLIPImageProcessor, CLIPVisionModelWithProjection, XLMRobertaTokenizer
 
-from ...configuration_utils import FrozenDict
-from ...image_processor import PipelineImageInput, VaeImageProcessor
-from ...loaders import FromSingleFileMixin, IPAdapterMixin, LoraLoaderMixin, TextualInversionLoaderMixin
-from ...models import AutoencoderKL, UNet2DConditionModel
-from ...models.lora import adjust_lora_scale_text_encoder
-from ...schedulers import KarrasDiffusionSchedulers
-from ...utils import (
+from ....configuration_utils import FrozenDict
+from ....image_processor import PipelineImageInput, VaeImageProcessor
+from ....loaders import FromSingleFileMixin, IPAdapterMixin, LoraLoaderMixin, TextualInversionLoaderMixin
+from ....models import AutoencoderKL, ImageProjection, UNet2DConditionModel
+from ....models.attention_processor import FusedAttnProcessor2_0
+from ....models.lora import adjust_lora_scale_text_encoder
+from ....schedulers import KarrasDiffusionSchedulers
+from ....utils import (
     PIL_INTERPOLATION,
     USE_PEFT_BACKEND,
     deprecate,
@@ -36,9 +37,9 @@ from ...utils import (
     scale_lora_layers,
     unscale_lora_layers,
 )
-from ...utils.torch_utils import randn_tensor
-from ..pipeline_utils import DiffusionPipeline
-from ..stable_diffusion.safety_checker import StableDiffusionSafetyChecker
+from ....utils.torch_utils import randn_tensor
+from ...pipeline_utils import DiffusionPipeline
+from ...stable_diffusion.safety_checker import StableDiffusionSafetyChecker
 from .modeling_roberta_series import RobertaSeriesModelWithTransformation
 from .pipeline_output import AltDiffusionPipelineOutput
 
@@ -158,7 +159,6 @@ def retrieve_timesteps(
     return timesteps, num_inference_steps
 
 
-# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.StableDiffusionImg2ImgPipeline with Stable->Alt, CLIPTextModel->RobertaSeriesModelWithTransformation, CLIPTokenizer->XLMRobertaTokenizer, AltDiffusionSafetyChecker->StableDiffusionSafetyChecker
 class AltDiffusionImg2ImgPipeline(
     DiffusionPipeline, TextualInversionLoaderMixin, IPAdapterMixin, LoraLoaderMixin, FromSingleFileMixin
 ):
@@ -195,7 +195,7 @@ class AltDiffusionImg2ImgPipeline(
             A `CLIPImageProcessor` to extract features from generated images; used as inputs to the `safety_checker`.
     """
 
-    model_cpu_offload_seq = "text_encoder->unet->vae"
+    model_cpu_offload_seq = "text_encoder->image_encoder->unet->vae"
     _optional_components = ["safety_checker", "feature_extractor", "image_encoder"]
     _exclude_from_cpu_offload = ["safety_checker"]
     _callback_tensor_inputs = ["latents", "prompt_embeds", "negative_prompt_embeds"]
@@ -505,18 +505,29 @@ class AltDiffusionImg2ImgPipeline(
 
         return prompt_embeds, negative_prompt_embeds
 
-    def encode_image(self, image, device, num_images_per_prompt):
+    def encode_image(self, image, device, num_images_per_prompt, output_hidden_states=None):
         dtype = next(self.image_encoder.parameters()).dtype
 
         if not isinstance(image, torch.Tensor):
             image = self.feature_extractor(image, return_tensors="pt").pixel_values
 
         image = image.to(device=device, dtype=dtype)
-        image_embeds = self.image_encoder(image).image_embeds
-        image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0)
+        if output_hidden_states:
+            image_enc_hidden_states = self.image_encoder(image, output_hidden_states=True).hidden_states[-2]
+            image_enc_hidden_states = image_enc_hidden_states.repeat_interleave(num_images_per_prompt, dim=0)
+            uncond_image_enc_hidden_states = self.image_encoder(
+                torch.zeros_like(image), output_hidden_states=True
+            ).hidden_states[-2]
+            uncond_image_enc_hidden_states = uncond_image_enc_hidden_states.repeat_interleave(
+                num_images_per_prompt, dim=0
+            )
+            return image_enc_hidden_states, uncond_image_enc_hidden_states
+        else:
+            image_embeds = self.image_encoder(image).image_embeds
+            image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0)
+            uncond_image_embeds = torch.zeros_like(image_embeds)
 
-        uncond_image_embeds = torch.zeros_like(image_embeds)
-        return image_embeds, uncond_image_embeds
+            return image_embeds, uncond_image_embeds
 
     def run_safety_checker(self, image, device, dtype):
         if self.safety_checker is None:
@@ -704,6 +715,65 @@ class AltDiffusionImg2ImgPipeline(
         """Disables the FreeU mechanism if enabled."""
         self.unet.disable_freeu()
 
+    def fuse_qkv_projections(self, unet: bool = True, vae: bool = True):
+        """
+        Enables fused QKV projections. For self-attention modules, all projection matrices (i.e., query,
+        key, value) are fused. For cross-attention modules, key and value projection matrices are fused.
+
+        <Tip warning={true}>
+
+        This API is 🧪 experimental.
+
+        </Tip>
+
+        Args:
+            unet (`bool`, defaults to `True`): To apply fusion on the UNet.
+            vae (`bool`, defaults to `True`): To apply fusion on the VAE.
+        """
+        self.fusing_unet = False
+        self.fusing_vae = False
+
+        if unet:
+            self.fusing_unet = True
+            self.unet.fuse_qkv_projections()
+            self.unet.set_attn_processor(FusedAttnProcessor2_0())
+
+        if vae:
+            if not isinstance(self.vae, AutoencoderKL):
+                raise ValueError("`fuse_qkv_projections()` is only supported for the VAE of type `AutoencoderKL`.")
+
+            self.fusing_vae = True
+            self.vae.fuse_qkv_projections()
+            self.vae.set_attn_processor(FusedAttnProcessor2_0())
+
+    def unfuse_qkv_projections(self, unet: bool = True, vae: bool = True):
+        """Disable QKV projection fusion if enabled.
+
+        <Tip warning={true}>
+
+        This API is 🧪 experimental.
+
+        </Tip>
+
+        Args:
+            unet (`bool`, defaults to `True`): To apply fusion on the UNet.
+            vae (`bool`, defaults to `True`): To apply fusion on the VAE.
+
+        """
+        if unet:
+            if not self.fusing_unet:
+                logger.warning("The UNet was not initially fused for QKV projections. Doing nothing.")
+            else:
+                self.unet.unfuse_qkv_projections()
+                self.fusing_unet = False
+
+        if vae:
+            if not self.fusing_vae:
+                logger.warning("The VAE was not initially fused for QKV projections. Doing nothing.")
+            else:
+                self.vae.unfuse_qkv_projections()
+                self.fusing_vae = False
+
     def get_guidance_scale_embedding(self, w, embedding_dim=512, dtype=torch.float32):
         """
         See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298
@@ -919,7 +989,10 @@ class AltDiffusionImg2ImgPipeline(
             prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds])
 
         if ip_adapter_image is not None:
-            image_embeds, negative_image_embeds = self.encode_image(ip_adapter_image, device, num_images_per_prompt)
+            output_hidden_state = False if isinstance(self.unet.encoder_hid_proj, ImageProjection) else True
+            image_embeds, negative_image_embeds = self.encode_image(
+                ip_adapter_image, device, num_images_per_prompt, output_hidden_state
+            )
             if self.do_classifier_free_guidance:
                 image_embeds = torch.cat([negative_image_embeds, image_embeds])
 

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

@@ -4,7 +4,7 @@ from typing import List, Optional, Union
 import numpy as np
 import PIL.Image
 
-from ...utils import (
+from ....utils import (
     BaseOutput,
 )
 

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

@@ -1,6 +1,6 @@
 from typing import TYPE_CHECKING
 
-from ...utils import DIFFUSERS_SLOW_IMPORT, _LazyModule
+from ....utils import DIFFUSERS_SLOW_IMPORT, _LazyModule
 
 
 _import_structure = {

diffusers/pipelines/{audio_diffusion → deprecated/audio_diffusion}/mel.py

@@ -15,8 +15,8 @@
 
 import numpy as np  # noqa: E402
 
-from ...configuration_utils import ConfigMixin, register_to_config
-from ...schedulers.scheduling_utils import SchedulerMixin
+from ....configuration_utils import ConfigMixin, register_to_config
+from ....schedulers.scheduling_utils import SchedulerMixin
 
 
 try:

diffusers/pipelines/{audio_diffusion → deprecated/audio_diffusion}/pipeline_audio_diffusion.py

@@ -20,10 +20,10 @@ import numpy as np
 import torch
 from PIL import Image
 
-from ...models import AutoencoderKL, UNet2DConditionModel
-from ...schedulers import DDIMScheduler, DDPMScheduler
-from ...utils.torch_utils import randn_tensor
-from ..pipeline_utils import AudioPipelineOutput, BaseOutput, DiffusionPipeline, ImagePipelineOutput
+from ....models import AutoencoderKL, UNet2DConditionModel
+from ....schedulers import DDIMScheduler, DDPMScheduler
+from ....utils.torch_utils import randn_tensor
+from ...pipeline_utils import AudioPipelineOutput, BaseOutput, DiffusionPipeline, ImagePipelineOutput
 from .mel import Mel
 
 

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

@@ -1,6 +1,6 @@
 from typing import TYPE_CHECKING
 
-from ...utils import DIFFUSERS_SLOW_IMPORT, _LazyModule
+from ....utils import DIFFUSERS_SLOW_IMPORT, _LazyModule
 
 
 _import_structure = {"pipeline_latent_diffusion_uncond": ["LDMPipeline"]}

diffusers/pipelines/{latent_diffusion_uncond → deprecated/latent_diffusion_uncond}/pipeline_latent_diffusion_uncond.py

@@ -17,10 +17,10 @@ from typing import List, Optional, Tuple, Union
 
 import torch
 
-from ...models import UNet2DModel, VQModel
-from ...schedulers import DDIMScheduler
-from ...utils.torch_utils import randn_tensor
-from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput
+from ....models import UNet2DModel, VQModel
+from ....schedulers import DDIMScheduler
+from ....utils.torch_utils import randn_tensor
+from ...pipeline_utils import DiffusionPipeline, ImagePipelineOutput
 
 
 class LDMPipeline(DiffusionPipeline):

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

@@ -1,6 +1,6 @@
 from typing import TYPE_CHECKING
 
-from ...utils import DIFFUSERS_SLOW_IMPORT, _LazyModule
+from ....utils import DIFFUSERS_SLOW_IMPORT, _LazyModule
 
 
 _import_structure = {"pipeline_pndm": ["PNDMPipeline"]}

diffusers/pipelines/{pndm → deprecated/pndm}/pipeline_pndm.py

@@ -17,10 +17,10 @@ from typing import List, Optional, Tuple, Union
 
 import torch
 
-from ...models import UNet2DModel
-from ...schedulers import PNDMScheduler
-from ...utils.torch_utils import randn_tensor
-from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput
+from ....models import UNet2DModel
+from ....schedulers import PNDMScheduler
+from ....utils.torch_utils import randn_tensor
+from ...pipeline_utils import DiffusionPipeline, ImagePipelineOutput
 
 
 class PNDMPipeline(DiffusionPipeline):

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

@@ -1,6 +1,6 @@
 from typing import TYPE_CHECKING
 
-from ...utils import DIFFUSERS_SLOW_IMPORT, _LazyModule
+from ....utils import DIFFUSERS_SLOW_IMPORT, _LazyModule
 
 
 _import_structure = {"pipeline_repaint": ["RePaintPipeline"]}

diffusers/pipelines/{repaint → deprecated/repaint}/pipeline_repaint.py

@@ -19,11 +19,11 @@ import numpy as np
 import PIL.Image
 import torch
 
-from ...models import UNet2DModel
-from ...schedulers import RePaintScheduler
-from ...utils import PIL_INTERPOLATION, deprecate, logging
-from ...utils.torch_utils import randn_tensor
-from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput
+from ....models import UNet2DModel
+from ....schedulers import RePaintScheduler
+from ....utils import PIL_INTERPOLATION, deprecate, logging
+from ....utils.torch_utils import randn_tensor
+from ...pipeline_utils import DiffusionPipeline, ImagePipelineOutput
 
 
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name

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

@@ -1,6 +1,6 @@
 from typing import TYPE_CHECKING
 
-from ...utils import DIFFUSERS_SLOW_IMPORT, _LazyModule
+from ....utils import DIFFUSERS_SLOW_IMPORT, _LazyModule
 
 
 _import_structure = {"pipeline_score_sde_ve": ["ScoreSdeVePipeline"]}

diffusers/pipelines/{score_sde_ve → deprecated/score_sde_ve}/pipeline_score_sde_ve.py

@@ -16,10 +16,10 @@ from typing import List, Optional, Tuple, Union
 
 import torch
 
-from ...models import UNet2DModel
-from ...schedulers import ScoreSdeVeScheduler
-from ...utils.torch_utils import randn_tensor
-from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput
+from ....models import UNet2DModel
+from ....schedulers import ScoreSdeVeScheduler
+from ....utils.torch_utils import randn_tensor
+from ...pipeline_utils import DiffusionPipeline, ImagePipelineOutput
 
 
 class ScoreSdeVePipeline(DiffusionPipeline):

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

@@ -1,7 +1,7 @@
 # flake8: noqa
 from typing import TYPE_CHECKING
-from ...utils import DIFFUSERS_SLOW_IMPORT
-from ...utils import (
+from ....utils import (
+    DIFFUSERS_SLOW_IMPORT,
     _LazyModule,
     is_note_seq_available,
     OptionalDependencyNotAvailable,
@@ -17,7 +17,7 @@ try:
     if not (is_transformers_available() and is_torch_available()):
         raise OptionalDependencyNotAvailable()
 except OptionalDependencyNotAvailable:
-    from ...utils import dummy_torch_and_transformers_objects  # noqa F403
+    from ....utils import dummy_torch_and_transformers_objects  # noqa F403
 
     _dummy_objects.update(get_objects_from_module(dummy_torch_and_transformers_objects))
 else:
@@ -32,7 +32,7 @@ try:
     if not (is_transformers_available() and is_torch_available() and is_note_seq_available()):
         raise OptionalDependencyNotAvailable()
 except OptionalDependencyNotAvailable:
-    from ...utils import dummy_transformers_and_torch_and_note_seq_objects
+    from ....utils import dummy_transformers_and_torch_and_note_seq_objects
 
     _dummy_objects.update(get_objects_from_module(dummy_transformers_and_torch_and_note_seq_objects))
 else:
@@ -45,7 +45,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
             raise OptionalDependencyNotAvailable()
 
     except OptionalDependencyNotAvailable:
-        from ...utils.dummy_torch_and_transformers_objects import *
+        from ....utils.dummy_torch_and_transformers_objects import *
     else:
         from .pipeline_spectrogram_diffusion import SpectrogramDiffusionPipeline
         from .pipeline_spectrogram_diffusion import SpectrogramContEncoder
@@ -56,7 +56,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
         if not (is_transformers_available() and is_torch_available() and is_note_seq_available()):
             raise OptionalDependencyNotAvailable()
     except OptionalDependencyNotAvailable:
-        from ...utils.dummy_transformers_and_torch_and_note_seq_objects import *
+        from ....utils.dummy_transformers_and_torch_and_note_seq_objects import *
 
     else:
         from .midi_utils import MidiProcessor

diffusers/pipelines/{spectrogram_diffusion/continous_encoder.py → deprecated/spectrogram_diffusion/continuous_encoder.py}

@@ -22,8 +22,8 @@ from transformers.models.t5.modeling_t5 import (
     T5LayerNorm,
 )
 
-from ...configuration_utils import ConfigMixin, register_to_config
-from ...models import ModelMixin
+from ....configuration_utils import ConfigMixin, register_to_config
+from ....models import ModelMixin
 
 
 class SpectrogramContEncoder(ModelMixin, ConfigMixin, ModuleUtilsMixin):

diffusers/pipelines/{spectrogram_diffusion → deprecated/spectrogram_diffusion}/midi_utils.py

@@ -22,7 +22,7 @@ import numpy as np
 import torch
 import torch.nn.functional as F
 
-from ...utils import is_note_seq_available
+from ....utils import is_note_seq_available
 from .pipeline_spectrogram_diffusion import TARGET_FEATURE_LENGTH
 
 

diffusers/pipelines/{spectrogram_diffusion → deprecated/spectrogram_diffusion}/notes_encoder.py

@@ -18,8 +18,8 @@ import torch.nn as nn
 from transformers.modeling_utils import ModuleUtilsMixin
 from transformers.models.t5.modeling_t5 import T5Block, T5Config, T5LayerNorm
 
-from ...configuration_utils import ConfigMixin, register_to_config
-from ...models import ModelMixin
+from ....configuration_utils import ConfigMixin, register_to_config
+from ....models import ModelMixin
 
 
 class SpectrogramNotesEncoder(ModelMixin, ConfigMixin, ModuleUtilsMixin):

diffusers/pipelines/{spectrogram_diffusion → deprecated/spectrogram_diffusion}/pipeline_spectrogram_diffusion.py

@@ -19,17 +19,17 @@ from typing import Any, Callable, List, Optional, Tuple, Union
 import numpy as np
 import torch
 
-from ...models import T5FilmDecoder
-from ...schedulers import DDPMScheduler
-from ...utils import is_onnx_available, logging
-from ...utils.torch_utils import randn_tensor
+from ....models import T5FilmDecoder
+from ....schedulers import DDPMScheduler
+from ....utils import is_onnx_available, logging
+from ....utils.torch_utils import randn_tensor
 
 
 if is_onnx_available():
-    from ..onnx_utils import OnnxRuntimeModel
+    from ...onnx_utils import OnnxRuntimeModel
 
-from ..pipeline_utils import AudioPipelineOutput, DiffusionPipeline
-from .continous_encoder import SpectrogramContEncoder
+from ...pipeline_utils import AudioPipelineOutput, DiffusionPipeline
+from .continuous_encoder import SpectrogramContEncoder
 from .notes_encoder import SpectrogramNotesEncoder
 
 

diffusers/pipelines/deprecated/stable_diffusion_variants/__init__.py

@@ -0,0 +1,55 @@
+from typing import TYPE_CHECKING
+
+from ....utils import (
+    DIFFUSERS_SLOW_IMPORT,
+    OptionalDependencyNotAvailable,
+    _LazyModule,
+    get_objects_from_module,
+    is_torch_available,
+    is_transformers_available,
+)
+
+
+_dummy_objects = {}
+_import_structure = {}
+
+try:
+    if not (is_transformers_available() and is_torch_available()):
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    from ....utils import dummy_torch_and_transformers_objects
+
+    _dummy_objects.update(get_objects_from_module(dummy_torch_and_transformers_objects))
+else:
+    _import_structure["pipeline_cycle_diffusion"] = ["CycleDiffusionPipeline"]
+    _import_structure["pipeline_stable_diffusion_inpaint_legacy"] = ["StableDiffusionInpaintPipelineLegacy"]
+    _import_structure["pipeline_stable_diffusion_model_editing"] = ["StableDiffusionModelEditingPipeline"]
+
+    _import_structure["pipeline_stable_diffusion_paradigms"] = ["StableDiffusionParadigmsPipeline"]
+    _import_structure["pipeline_stable_diffusion_pix2pix_zero"] = ["StableDiffusionPix2PixZeroPipeline"]
+
+if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
+    try:
+        if not (is_transformers_available() and is_torch_available()):
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        from ....utils.dummy_torch_and_transformers_objects import *
+
+    else:
+        from .pipeline_cycle_diffusion import CycleDiffusionPipeline
+        from .pipeline_stable_diffusion_inpaint_legacy import StableDiffusionInpaintPipelineLegacy
+        from .pipeline_stable_diffusion_model_editing import StableDiffusionModelEditingPipeline
+        from .pipeline_stable_diffusion_paradigms import StableDiffusionParadigmsPipeline
+        from .pipeline_stable_diffusion_pix2pix_zero import StableDiffusionPix2PixZeroPipeline
+
+else:
+    import sys
+
+    sys.modules[__name__] = _LazyModule(
+        __name__,
+        globals()["__file__"],
+        _import_structure,
+        module_spec=__spec__,
+    )
+    for name, value in _dummy_objects.items():
+        setattr(sys.modules[__name__], name, value)