diffusers 0.30.3__py3-none-any.whl → 0.32.0__py3-none-any.whl

diffusers/models/unets/unet_motion_model.py

@@ -116,7 +116,7 @@ class AnimateDiffTransformer3D(nn.Module):
 
         self.in_channels = in_channels
 
-        self.norm = torch.nn.GroupNorm(num_groups=norm_num_groups, num_channels=in_channels, eps=1e-6, affine=True)
+        self.norm = nn.GroupNorm(num_groups=norm_num_groups, num_channels=in_channels, eps=1e-6, affine=True)
         self.proj_in = nn.Linear(in_channels, inner_dim)
 
         # 3. Define transformers blocks
@@ -187,12 +187,12 @@ class AnimateDiffTransformer3D(nn.Module):
         hidden_states = self.norm(hidden_states)
         hidden_states = hidden_states.permute(0, 3, 4, 2, 1).reshape(batch_size * height * width, num_frames, channel)
 
-        hidden_states = self.proj_in(hidden_states)
+        hidden_states = self.proj_in(input=hidden_states)
 
         # 2. Blocks
         for block in self.transformer_blocks:
             hidden_states = block(
-                hidden_states,
+                hidden_states=hidden_states,
                 encoder_hidden_states=encoder_hidden_states,
                 timestep=timestep,
                 cross_attention_kwargs=cross_attention_kwargs,
@@ -200,7 +200,7 @@ class AnimateDiffTransformer3D(nn.Module):
             )
 
         # 3. Output
-        hidden_states = self.proj_out(hidden_states)
+        hidden_states = self.proj_out(input=hidden_states)
         hidden_states = (
             hidden_states[None, None, :]
             .reshape(batch_size, height, width, num_frames, channel)
@@ -323,7 +323,7 @@ class DownBlockMotion(nn.Module):
 
         blocks = zip(self.resnets, self.motion_modules)
         for resnet, motion_module in blocks:
-            if self.training and self.gradient_checkpointing:
+            if torch.is_grad_enabled() and self.gradient_checkpointing:
 
                 def create_custom_forward(module):
                     def custom_forward(*inputs):
@@ -344,7 +344,7 @@ class DownBlockMotion(nn.Module):
                     )
 
             else:
-                hidden_states = resnet(hidden_states, temb)
+                hidden_states = resnet(input_tensor=hidden_states, temb=temb)
 
             hidden_states = motion_module(hidden_states, num_frames=num_frames)
 
@@ -352,7 +352,7 @@ class DownBlockMotion(nn.Module):
 
         if self.downsamplers is not None:
             for downsampler in self.downsamplers:
-                hidden_states = downsampler(hidden_states)
+                hidden_states = downsampler(hidden_states=hidden_states)
 
             output_states = output_states + (hidden_states,)
 
@@ -513,7 +513,7 @@ class CrossAttnDownBlockMotion(nn.Module):
 
         blocks = list(zip(self.resnets, self.attentions, self.motion_modules))
         for i, (resnet, attn, motion_module) in enumerate(blocks):
-            if self.training and self.gradient_checkpointing:
+            if torch.is_grad_enabled() and self.gradient_checkpointing:
 
                 def create_custom_forward(module, return_dict=None):
                     def custom_forward(*inputs):
@@ -531,25 +531,18 @@ class CrossAttnDownBlockMotion(nn.Module):
                     temb,
                     **ckpt_kwargs,
                 )
-                hidden_states = attn(
-                    hidden_states,
-                    encoder_hidden_states=encoder_hidden_states,
-                    cross_attention_kwargs=cross_attention_kwargs,
-                    attention_mask=attention_mask,
-                    encoder_attention_mask=encoder_attention_mask,
-                    return_dict=False,
-                )[0]
             else:
-                hidden_states = resnet(hidden_states, temb)
+                hidden_states = resnet(input_tensor=hidden_states, temb=temb)
+
+            hidden_states = attn(
+                hidden_states=hidden_states,
+                encoder_hidden_states=encoder_hidden_states,
+                cross_attention_kwargs=cross_attention_kwargs,
+                attention_mask=attention_mask,
+                encoder_attention_mask=encoder_attention_mask,
+                return_dict=False,
+            )[0]
 
-                hidden_states = attn(
-                    hidden_states,
-                    encoder_hidden_states=encoder_hidden_states,
-                    cross_attention_kwargs=cross_attention_kwargs,
-                    attention_mask=attention_mask,
-                    encoder_attention_mask=encoder_attention_mask,
-                    return_dict=False,
-                )[0]
             hidden_states = motion_module(
                 hidden_states,
                 num_frames=num_frames,
@@ -563,7 +556,7 @@ class CrossAttnDownBlockMotion(nn.Module):
 
         if self.downsamplers is not None:
             for downsampler in self.downsamplers:
-                hidden_states = downsampler(hidden_states)
+                hidden_states = downsampler(hidden_states=hidden_states)
 
             output_states = output_states + (hidden_states,)
 
@@ -739,7 +732,7 @@ class CrossAttnUpBlockMotion(nn.Module):
 
             hidden_states = torch.cat([hidden_states, res_hidden_states], dim=1)
 
-            if self.training and self.gradient_checkpointing:
+            if torch.is_grad_enabled() and self.gradient_checkpointing:
 
                 def create_custom_forward(module, return_dict=None):
                     def custom_forward(*inputs):
@@ -757,25 +750,18 @@ class CrossAttnUpBlockMotion(nn.Module):
                     temb,
                     **ckpt_kwargs,
                 )
-                hidden_states = attn(
-                    hidden_states,
-                    encoder_hidden_states=encoder_hidden_states,
-                    cross_attention_kwargs=cross_attention_kwargs,
-                    attention_mask=attention_mask,
-                    encoder_attention_mask=encoder_attention_mask,
-                    return_dict=False,
-                )[0]
             else:
-                hidden_states = resnet(hidden_states, temb)
+                hidden_states = resnet(input_tensor=hidden_states, temb=temb)
+
+            hidden_states = attn(
+                hidden_states=hidden_states,
+                encoder_hidden_states=encoder_hidden_states,
+                cross_attention_kwargs=cross_attention_kwargs,
+                attention_mask=attention_mask,
+                encoder_attention_mask=encoder_attention_mask,
+                return_dict=False,
+            )[0]
 
-                hidden_states = attn(
-                    hidden_states,
-                    encoder_hidden_states=encoder_hidden_states,
-                    cross_attention_kwargs=cross_attention_kwargs,
-                    attention_mask=attention_mask,
-                    encoder_attention_mask=encoder_attention_mask,
-                    return_dict=False,
-                )[0]
             hidden_states = motion_module(
                 hidden_states,
                 num_frames=num_frames,
@@ -783,7 +769,7 @@ class CrossAttnUpBlockMotion(nn.Module):
 
         if self.upsamplers is not None:
             for upsampler in self.upsamplers:
-                hidden_states = upsampler(hidden_states, upsample_size)
+                hidden_states = upsampler(hidden_states=hidden_states, output_size=upsample_size)
 
         return hidden_states
 
@@ -909,7 +895,7 @@ class UpBlockMotion(nn.Module):
 
             hidden_states = torch.cat([hidden_states, res_hidden_states], dim=1)
 
-            if self.training and self.gradient_checkpointing:
+            if torch.is_grad_enabled() and self.gradient_checkpointing:
 
                 def create_custom_forward(module):
                     def custom_forward(*inputs):
@@ -929,13 +915,13 @@ class UpBlockMotion(nn.Module):
                         create_custom_forward(resnet), hidden_states, temb
                     )
             else:
-                hidden_states = resnet(hidden_states, temb)
+                hidden_states = resnet(input_tensor=hidden_states, temb=temb)
 
             hidden_states = motion_module(hidden_states, num_frames=num_frames)
 
         if self.upsamplers is not None:
             for upsampler in self.upsamplers:
-                hidden_states = upsampler(hidden_states, upsample_size)
+                hidden_states = upsampler(hidden_states=hidden_states, output_size=upsample_size)
 
         return hidden_states
 
@@ -1080,11 +1066,20 @@ class UNetMidBlockCrossAttnMotion(nn.Module):
             if cross_attention_kwargs.get("scale", None) is not None:
                 logger.warning("Passing `scale` to `cross_attention_kwargs` is deprecated. `scale` will be ignored.")
 
-        hidden_states = self.resnets[0](hidden_states, temb)
+        hidden_states = self.resnets[0](input_tensor=hidden_states, temb=temb)
 
         blocks = zip(self.attentions, self.resnets[1:], self.motion_modules)
         for attn, resnet, motion_module in blocks:
-            if self.training and self.gradient_checkpointing:
+            hidden_states = attn(
+                hidden_states=hidden_states,
+                encoder_hidden_states=encoder_hidden_states,
+                cross_attention_kwargs=cross_attention_kwargs,
+                attention_mask=attention_mask,
+                encoder_attention_mask=encoder_attention_mask,
+                return_dict=False,
+            )[0]
+
+            if torch.is_grad_enabled() and self.gradient_checkpointing:
 
                 def create_custom_forward(module, return_dict=None):
                     def custom_forward(*inputs):
@@ -1096,14 +1091,6 @@ class UNetMidBlockCrossAttnMotion(nn.Module):
                     return custom_forward
 
                 ckpt_kwargs: Dict[str, Any] = {"use_reentrant": False} if is_torch_version(">=", "1.11.0") else {}
-                hidden_states = attn(
-                    hidden_states,
-                    encoder_hidden_states=encoder_hidden_states,
-                    cross_attention_kwargs=cross_attention_kwargs,
-                    attention_mask=attention_mask,
-                    encoder_attention_mask=encoder_attention_mask,
-                    return_dict=False,
-                )[0]
                 hidden_states = torch.utils.checkpoint.checkpoint(
                     create_custom_forward(motion_module),
                     hidden_states,
@@ -1117,19 +1104,11 @@ class UNetMidBlockCrossAttnMotion(nn.Module):
                     **ckpt_kwargs,
                 )
             else:
-                hidden_states = attn(
-                    hidden_states,
-                    encoder_hidden_states=encoder_hidden_states,
-                    cross_attention_kwargs=cross_attention_kwargs,
-                    attention_mask=attention_mask,
-                    encoder_attention_mask=encoder_attention_mask,
-                    return_dict=False,
-                )[0]
                 hidden_states = motion_module(
                     hidden_states,
                     num_frames=num_frames,
                 )
-                hidden_states = resnet(hidden_states, temb)
+                hidden_states = resnet(input_tensor=hidden_states, temb=temb)
 
         return hidden_states
 
@@ -2178,7 +2157,6 @@ class UNetMotionModel(ModelMixin, ConfigMixin, UNet2DConditionLoadersMixin, Peft
 
         emb = emb if aug_emb is None else emb + aug_emb
         emb = emb.repeat_interleave(repeats=num_frames, dim=0)
-        encoder_hidden_states = encoder_hidden_states.repeat_interleave(repeats=num_frames, dim=0)
 
         if self.encoder_hid_proj is not None and self.config.encoder_hid_dim_type == "ip_image_proj":
             if "image_embeds" not in added_cond_kwargs:

diffusers/models/unets/unet_spatio_temporal_condition.py

@@ -382,6 +382,20 @@ class UNetSpatioTemporalConditionModel(ModelMixin, ConfigMixin, UNet2DConditionL
                 If `return_dict` is True, an [`~models.unet_slatio_temporal.UNetSpatioTemporalConditionOutput`] is
                 returned, otherwise a `tuple` is returned where the first element is the sample tensor.
         """
+        # By default samples have to be AT least a multiple of the overall upsampling factor.
+        # The overall upsampling factor is equal to 2 ** (# num of upsampling layears).
+        # However, the upsampling interpolation output size can be forced to fit any upsampling size
+        # on the fly if necessary.
+        default_overall_up_factor = 2**self.num_upsamplers
+
+        # upsample size should be forwarded when sample is not a multiple of `default_overall_up_factor`
+        forward_upsample_size = False
+        upsample_size = None
+
+        if any(s % default_overall_up_factor != 0 for s in sample.shape[-2:]):
+            logger.info("Forward upsample size to force interpolation output size.")
+            forward_upsample_size = True
+
         # 1. time
         timesteps = timestep
         if not torch.is_tensor(timesteps):
@@ -457,15 +471,23 @@ class UNetSpatioTemporalConditionModel(ModelMixin, ConfigMixin, UNet2DConditionL
 
         # 5. up
         for i, upsample_block in enumerate(self.up_blocks):
+            is_final_block = i == len(self.up_blocks) - 1
+
             res_samples = down_block_res_samples[-len(upsample_block.resnets) :]
             down_block_res_samples = down_block_res_samples[: -len(upsample_block.resnets)]
 
+            # if we have not reached the final block and need to forward the
+            # upsample size, we do it here
+            if not is_final_block and forward_upsample_size:
+                upsample_size = down_block_res_samples[-1].shape[2:]
+
             if hasattr(upsample_block, "has_cross_attention") and upsample_block.has_cross_attention:
                 sample = upsample_block(
                     hidden_states=sample,
                     temb=emb,
                     res_hidden_states_tuple=res_samples,
                     encoder_hidden_states=encoder_hidden_states,
+                    upsample_size=upsample_size,
                     image_only_indicator=image_only_indicator,
                 )
             else:
@@ -473,6 +495,7 @@ class UNetSpatioTemporalConditionModel(ModelMixin, ConfigMixin, UNet2DConditionL
                     hidden_states=sample,
                     temb=emb,
                     res_hidden_states_tuple=res_samples,
+                    upsample_size=upsample_size,
                     image_only_indicator=image_only_indicator,
                 )
 

diffusers/models/unets/unet_stable_cascade.py

@@ -455,7 +455,7 @@ class StableCascadeUNet(ModelMixin, ConfigMixin, FromOriginalModelMixin):
         level_outputs = []
         block_group = zip(self.down_blocks, self.down_downscalers, self.down_repeat_mappers)
 
-        if self.training and self.gradient_checkpointing:
+        if torch.is_grad_enabled() and self.gradient_checkpointing:
 
             def create_custom_forward(module):
                 def custom_forward(*inputs):
@@ -504,7 +504,7 @@ class StableCascadeUNet(ModelMixin, ConfigMixin, FromOriginalModelMixin):
         x = level_outputs[0]
         block_group = zip(self.up_blocks, self.up_upscalers, self.up_repeat_mappers)
 
-        if self.training and self.gradient_checkpointing:
+        if torch.is_grad_enabled() and self.gradient_checkpointing:
 
             def create_custom_forward(module):
                 def custom_forward(*inputs):

diffusers/models/unets/uvit_2d.py

@@ -181,7 +181,7 @@ class UVit2DModel(ModelMixin, ConfigMixin, PeftAdapterMixin):
         hidden_states = self.project_to_hidden(hidden_states)
 
         for layer in self.transformer_layers:
-            if self.training and self.gradient_checkpointing:
+            if torch.is_grad_enabled() and self.gradient_checkpointing:
 
                 def layer_(*args):
                     return checkpoint(layer, *args)

diffusers/models/upsampling.py

@@ -19,6 +19,7 @@ import torch.nn as nn
 import torch.nn.functional as F
 
 from ..utils import deprecate
+from ..utils.import_utils import is_torch_version
 from .normalization import RMSNorm
 
 
@@ -151,11 +152,10 @@ class Upsample2D(nn.Module):
         if self.use_conv_transpose:
             return self.conv(hidden_states)
 
-        # Cast to float32 to as 'upsample_nearest2d_out_frame' op does not support bfloat16
-        # TODO(Suraj): Remove this cast once the issue is fixed in PyTorch
-        # https://github.com/pytorch/pytorch/issues/86679
+        # Cast to float32 to as 'upsample_nearest2d_out_frame' op does not support bfloat16 until PyTorch 2.1
+        # https://github.com/pytorch/pytorch/issues/86679#issuecomment-1783978767
         dtype = hidden_states.dtype
-        if dtype == torch.bfloat16:
+        if dtype == torch.bfloat16 and is_torch_version("<", "2.1"):
             hidden_states = hidden_states.to(torch.float32)
 
         # upsample_nearest_nhwc fails with large batch sizes. see https://github.com/huggingface/diffusers/issues/984
@@ -165,13 +165,21 @@ class Upsample2D(nn.Module):
         # if `output_size` is passed we force the interpolation output
         # size and do not make use of `scale_factor=2`
         if self.interpolate:
+            # upsample_nearest_nhwc also fails when the number of output elements is large
+            # https://github.com/pytorch/pytorch/issues/141831
+            scale_factor = (
+                2 if output_size is None else max([f / s for f, s in zip(output_size, hidden_states.shape[-2:])])
+            )
+            if hidden_states.numel() * scale_factor > pow(2, 31):
+                hidden_states = hidden_states.contiguous()
+
             if output_size is None:
                 hidden_states = F.interpolate(hidden_states, scale_factor=2.0, mode="nearest")
             else:
                 hidden_states = F.interpolate(hidden_states, size=output_size, mode="nearest")
 
-        # If the input is bfloat16, we cast back to bfloat16
-        if dtype == torch.bfloat16:
+        # Cast back to original dtype
+        if dtype == torch.bfloat16 and is_torch_version("<", "2.1"):
             hidden_states = hidden_states.to(dtype)
 
         # TODO(Suraj, Patrick) - clean up after weight dicts are correctly renamed

diffusers/pipelines/__init__.py

@@ -116,6 +116,7 @@ else:
             "VersatileDiffusionTextToImagePipeline",
         ]
     )
+    _import_structure["allegro"] = ["AllegroPipeline"]
     _import_structure["amused"] = ["AmusedImg2ImgPipeline", "AmusedInpaintPipeline", "AmusedPipeline"]
     _import_structure["animatediff"] = [
         "AnimateDiffPipeline",
@@ -123,8 +124,22 @@ else:
         "AnimateDiffSDXLPipeline",
         "AnimateDiffSparseControlNetPipeline",
         "AnimateDiffVideoToVideoPipeline",
+        "AnimateDiffVideoToVideoControlNetPipeline",
+    ]
+    _import_structure["flux"] = [
+        "FluxControlPipeline",
+        "FluxControlInpaintPipeline",
+        "FluxControlImg2ImgPipeline",
+        "FluxControlNetPipeline",
+        "FluxControlNetImg2ImgPipeline",
+        "FluxControlNetInpaintPipeline",
+        "FluxImg2ImgPipeline",
+        "FluxInpaintPipeline",
+        "FluxPipeline",
+        "FluxFillPipeline",
+        "FluxPriorReduxPipeline",
+        "ReduxImageEncoder",
     ]
-    _import_structure["flux"] = ["FluxPipeline"]
     _import_structure["audioldm"] = ["AudioLDMPipeline"]
     _import_structure["audioldm2"] = [
         "AudioLDM2Pipeline",
@@ -136,7 +151,9 @@ else:
         "CogVideoXPipeline",
         "CogVideoXImageToVideoPipeline",
         "CogVideoXVideoToVideoPipeline",
+        "CogVideoXFunControlPipeline",
     ]
+    _import_structure["cogview3"] = ["CogView3PlusPipeline"]
     _import_structure["controlnet"].extend(
         [
             "BlipDiffusionControlNetPipeline",
@@ -146,21 +163,30 @@ else:
             "StableDiffusionXLControlNetImg2ImgPipeline",
             "StableDiffusionXLControlNetInpaintPipeline",
             "StableDiffusionXLControlNetPipeline",
+            "StableDiffusionXLControlNetUnionPipeline",
+            "StableDiffusionXLControlNetUnionInpaintPipeline",
+            "StableDiffusionXLControlNetUnionImg2ImgPipeline",
         ]
     )
     _import_structure["pag"].extend(
         [
+            "StableDiffusionControlNetPAGInpaintPipeline",
             "AnimateDiffPAGPipeline",
             "KolorsPAGPipeline",
             "HunyuanDiTPAGPipeline",
             "StableDiffusion3PAGPipeline",
+            "StableDiffusion3PAGImg2ImgPipeline",
             "StableDiffusionPAGPipeline",
+            "StableDiffusionPAGImg2ImgPipeline",
+            "StableDiffusionPAGInpaintPipeline",
             "StableDiffusionControlNetPAGPipeline",
             "StableDiffusionXLPAGPipeline",
             "StableDiffusionXLPAGInpaintPipeline",
+            "StableDiffusionXLControlNetPAGImg2ImgPipeline",
             "StableDiffusionXLControlNetPAGPipeline",
             "StableDiffusionXLPAGImg2ImgPipeline",
             "PixArtSigmaPAGPipeline",
+            "SanaPAGPipeline",
         ]
     )
     _import_structure["controlnet_xs"].extend(
@@ -177,6 +203,7 @@ else:
     _import_structure["controlnet_sd3"].extend(
         [
             "StableDiffusion3ControlNetPipeline",
+            "StableDiffusion3ControlNetInpaintingPipeline",
         ]
     )
     _import_structure["deepfloyd_if"] = [
@@ -188,6 +215,7 @@ else:
         "IFSuperResolutionPipeline",
     ]
     _import_structure["hunyuandit"] = ["HunyuanDiTPipeline"]
+    _import_structure["hunyuan_video"] = ["HunyuanVideoPipeline"]
     _import_structure["kandinsky"] = [
         "KandinskyCombinedPipeline",
         "KandinskyImg2ImgCombinedPipeline",
@@ -225,6 +253,7 @@ else:
         ]
     )
     _import_structure["latte"] = ["LattePipeline"]
+    _import_structure["ltx"] = ["LTXPipeline", "LTXImageToVideoPipeline"]
     _import_structure["lumina"] = ["LuminaText2ImgPipeline"]
     _import_structure["marigold"].extend(
         [
@@ -232,10 +261,12 @@ else:
             "MarigoldNormalsPipeline",
         ]
     )
+    _import_structure["mochi"] = ["MochiPipeline"]
     _import_structure["musicldm"] = ["MusicLDMPipeline"]
     _import_structure["paint_by_example"] = ["PaintByExamplePipeline"]
     _import_structure["pia"] = ["PIAPipeline"]
     _import_structure["pixart_alpha"] = ["PixArtAlphaPipeline", "PixArtSigmaPipeline"]
+    _import_structure["sana"] = ["SanaPipeline"]
     _import_structure["semantic_stable_diffusion"] = ["SemanticStableDiffusionPipeline"]
     _import_structure["shap_e"] = ["ShapEImg2ImgPipeline", "ShapEPipeline"]
     _import_structure["stable_audio"] = [
@@ -440,12 +471,14 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
     except OptionalDependencyNotAvailable:
         from ..utils.dummy_torch_and_transformers_objects import *
     else:
+        from .allegro import AllegroPipeline
         from .amused import AmusedImg2ImgPipeline, AmusedInpaintPipeline, AmusedPipeline
         from .animatediff import (
             AnimateDiffControlNetPipeline,
             AnimateDiffPipeline,
             AnimateDiffSDXLPipeline,
             AnimateDiffSparseControlNetPipeline,
+            AnimateDiffVideoToVideoControlNetPipeline,
             AnimateDiffVideoToVideoPipeline,
         )
         from .audioldm import AudioLDMPipeline
@@ -456,7 +489,13 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
         )
         from .aura_flow import AuraFlowPipeline
         from .blip_diffusion import BlipDiffusionPipeline
-        from .cogvideo import CogVideoXImageToVideoPipeline, CogVideoXPipeline, CogVideoXVideoToVideoPipeline
+        from .cogvideo import (
+            CogVideoXFunControlPipeline,
+            CogVideoXImageToVideoPipeline,
+            CogVideoXPipeline,
+            CogVideoXVideoToVideoPipeline,
+        )
+        from .cogview3 import CogView3PlusPipeline
         from .controlnet import (
             BlipDiffusionControlNetPipeline,
             StableDiffusionControlNetImg2ImgPipeline,
@@ -465,13 +504,14 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
             StableDiffusionXLControlNetImg2ImgPipeline,
             StableDiffusionXLControlNetInpaintPipeline,
             StableDiffusionXLControlNetPipeline,
+            StableDiffusionXLControlNetUnionImg2ImgPipeline,
+            StableDiffusionXLControlNetUnionInpaintPipeline,
+            StableDiffusionXLControlNetUnionPipeline,
         )
         from .controlnet_hunyuandit import (
             HunyuanDiTControlNetPipeline,
         )
-        from .controlnet_sd3 import (
-            StableDiffusion3ControlNetPipeline,
-        )
+        from .controlnet_sd3 import StableDiffusion3ControlNetInpaintingPipeline, StableDiffusion3ControlNetPipeline
         from .controlnet_xs import (
             StableDiffusionControlNetXSPipeline,
             StableDiffusionXLControlNetXSPipeline,
@@ -498,7 +538,21 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
             VersatileDiffusionTextToImagePipeline,
             VQDiffusionPipeline,
         )
-        from .flux import FluxPipeline
+        from .flux import (
+            FluxControlImg2ImgPipeline,
+            FluxControlInpaintPipeline,
+            FluxControlNetImg2ImgPipeline,
+            FluxControlNetInpaintPipeline,
+            FluxControlNetPipeline,
+            FluxControlPipeline,
+            FluxFillPipeline,
+            FluxImg2ImgPipeline,
+            FluxInpaintPipeline,
+            FluxPipeline,
+            FluxPriorReduxPipeline,
+            ReduxImageEncoder,
+        )
+        from .hunyuan_video import HunyuanVideoPipeline
         from .hunyuandit import HunyuanDiTPipeline
         from .i2vgen_xl import I2VGenXLPipeline
         from .kandinsky import (
@@ -538,20 +592,28 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
             LEditsPPPipelineStableDiffusion,
             LEditsPPPipelineStableDiffusionXL,
         )
+        from .ltx import LTXImageToVideoPipeline, LTXPipeline
         from .lumina import LuminaText2ImgPipeline
         from .marigold import (
             MarigoldDepthPipeline,
             MarigoldNormalsPipeline,
         )
+        from .mochi import MochiPipeline
         from .musicldm import MusicLDMPipeline
         from .pag import (
             AnimateDiffPAGPipeline,
             HunyuanDiTPAGPipeline,
             KolorsPAGPipeline,
             PixArtSigmaPAGPipeline,
+            SanaPAGPipeline,
+            StableDiffusion3PAGImg2ImgPipeline,
             StableDiffusion3PAGPipeline,
+            StableDiffusionControlNetPAGInpaintPipeline,
             StableDiffusionControlNetPAGPipeline,
+            StableDiffusionPAGImg2ImgPipeline,
+            StableDiffusionPAGInpaintPipeline,
             StableDiffusionPAGPipeline,
+            StableDiffusionXLControlNetPAGImg2ImgPipeline,
             StableDiffusionXLControlNetPAGPipeline,
             StableDiffusionXLPAGImg2ImgPipeline,
             StableDiffusionXLPAGInpaintPipeline,
@@ -560,6 +622,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
         from .paint_by_example import PaintByExamplePipeline
         from .pia import PIAPipeline
         from .pixart_alpha import PixArtAlphaPipeline, PixArtSigmaPipeline
+        from .sana import SanaPipeline
         from .semantic_stable_diffusion import SemanticStableDiffusionPipeline
         from .shap_e import ShapEImg2ImgPipeline, ShapEPipeline
         from .stable_audio import StableAudioPipeline, StableAudioProjectionModel

diffusers/pipelines/allegro/__init__.py

@@ -0,0 +1,48 @@
+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  # noqa F403
+
+    _dummy_objects.update(get_objects_from_module(dummy_torch_and_transformers_objects))
+else:
+    _import_structure["pipeline_allegro"] = ["AllegroPipeline"]
+
+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_allegro import AllegroPipeline
+
+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)