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

diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3_inpaint.py

@@ -25,7 +25,7 @@ from transformers import (
 
 from ...callbacks import MultiPipelineCallbacks, PipelineCallback
 from ...image_processor import PipelineImageInput, VaeImageProcessor
-from ...loaders import SD3LoraLoaderMixin
+from ...loaders import FromSingleFileMixin, SD3LoraLoaderMixin
 from ...models.autoencoders import AutoencoderKL
 from ...models.transformers import SD3Transformer2DModel
 from ...schedulers import FlowMatchEulerDiscreteScheduler
@@ -74,6 +74,20 @@ EXAMPLE_DOC_STRING = """
 """
 
 
+# Copied from diffusers.pipelines.flux.pipeline_flux.calculate_shift
+def calculate_shift(
+    image_seq_len,
+    base_seq_len: int = 256,
+    max_seq_len: int = 4096,
+    base_shift: float = 0.5,
+    max_shift: float = 1.16,
+):
+    m = (max_shift - base_shift) / (max_seq_len - base_seq_len)
+    b = base_shift - m * base_seq_len
+    mu = image_seq_len * m + b
+    return mu
+
+
 # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.retrieve_latents
 def retrieve_latents(
     encoder_output: torch.Tensor, generator: Optional[torch.Generator] = None, sample_mode: str = "sample"
@@ -97,7 +111,7 @@ def retrieve_timesteps(
     sigmas: Optional[List[float]] = None,
     **kwargs,
 ):
-    """
+    r"""
     Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
     custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
 
@@ -148,7 +162,7 @@ def retrieve_timesteps(
     return timesteps, num_inference_steps
 
 
-class StableDiffusion3InpaintPipeline(DiffusionPipeline):
+class StableDiffusion3InpaintPipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingleFileMixin):
     r"""
     Args:
         transformer ([`SD3Transformer2DModel`]):
@@ -224,6 +238,9 @@ class StableDiffusion3InpaintPipeline(DiffusionPipeline):
         )
         self.tokenizer_max_length = self.tokenizer.model_max_length
         self.default_sample_size = self.transformer.config.sample_size
+        self.patch_size = (
+            self.transformer.config.patch_size if hasattr(self, "transformer") and self.transformer is not None else 2
+        )
 
     # Copied from diffusers.pipelines.stable_diffusion_3.pipeline_stable_diffusion_3.StableDiffusion3Pipeline._get_t5_prompt_embeds
     def _get_t5_prompt_embeds(
@@ -538,6 +555,8 @@ class StableDiffusion3InpaintPipeline(DiffusionPipeline):
         prompt,
         prompt_2,
         prompt_3,
+        height,
+        width,
         strength,
         negative_prompt=None,
         negative_prompt_2=None,
@@ -549,6 +568,15 @@ class StableDiffusion3InpaintPipeline(DiffusionPipeline):
         callback_on_step_end_tensor_inputs=None,
         max_sequence_length=None,
     ):
+        if (
+            height % (self.vae_scale_factor * self.patch_size) != 0
+            or width % (self.vae_scale_factor * self.patch_size) != 0
+        ):
+            raise ValueError(
+                f"`height` and `width` have to be divisible by {self.vae_scale_factor * self.patch_size} but are {height} and {width}."
+                f"You can use height {height - height % (self.vae_scale_factor * self.patch_size)} and width {width - width % (self.vae_scale_factor * self.patch_size)}."
+            )
+
         if strength < 0 or strength > 1:
             raise ValueError(f"The value of strength should in [0.0, 1.0] but is {strength}")
 
@@ -806,7 +834,7 @@ class StableDiffusion3InpaintPipeline(DiffusionPipeline):
         padding_mask_crop: Optional[int] = None,
         strength: float = 0.6,
         num_inference_steps: int = 50,
-        timesteps: List[int] = None,
+        sigmas: Optional[List[float]] = None,
         guidance_scale: float = 7.0,
         negative_prompt: Optional[Union[str, List[str]]] = None,
         negative_prompt_2: Optional[Union[str, List[str]]] = None,
@@ -824,6 +852,7 @@ class StableDiffusion3InpaintPipeline(DiffusionPipeline):
         callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
         callback_on_step_end_tensor_inputs: List[str] = ["latents"],
         max_sequence_length: int = 256,
+        mu: Optional[float] = None,
     ):
         r"""
         Function invoked when calling the pipeline for generation.
@@ -874,10 +903,10 @@ class StableDiffusion3InpaintPipeline(DiffusionPipeline):
             num_inference_steps (`int`, *optional*, defaults to 50):
                 The number of denoising steps. More denoising steps usually lead to a higher quality image at the
                 expense of slower inference.
-            timesteps (`List[int]`, *optional*):
-                Custom timesteps to use for the denoising process with schedulers which support a `timesteps` argument
-                in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is
-                passed will be used. Must be in descending order.
+            sigmas (`List[float]`, *optional*):
+                Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in
+                their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
+                will be used.
             guidance_scale (`float`, *optional*, defaults to 7.0):
                 Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
                 `guidance_scale` is defined as `w` of equation 2. of [Imagen
@@ -921,8 +950,8 @@ class StableDiffusion3InpaintPipeline(DiffusionPipeline):
                 The output format of the generate image. Choose between
                 [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
             return_dict (`bool`, *optional*, defaults to `True`):
-                Whether or not to return a [`~pipelines.stable_diffusion_xl.StableDiffusionXLPipelineOutput`] instead
-                of a plain tuple.
+                Whether or not to return a [`~pipelines.stable_diffusion_3.StableDiffusion3PipelineOutput`] instead of
+                a plain tuple.
             callback_on_step_end (`Callable`, *optional*):
                 A function that calls at the end of each denoising steps during the inference. The function is called
                 with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int,
@@ -933,6 +962,7 @@ class StableDiffusion3InpaintPipeline(DiffusionPipeline):
                 will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
                 `._callback_tensor_inputs` attribute of your pipeline class.
             max_sequence_length (`int` defaults to 256): Maximum sequence length to use with the `prompt`.
+            mu (`float`, *optional*): `mu` value used for `dynamic_shifting`.
 
         Examples:
 
@@ -953,6 +983,8 @@ class StableDiffusion3InpaintPipeline(DiffusionPipeline):
             prompt,
             prompt_2,
             prompt_3,
+            height,
+            width,
             strength,
             negative_prompt=negative_prompt,
             negative_prompt_2=negative_prompt_2,
@@ -1007,7 +1039,24 @@ class StableDiffusion3InpaintPipeline(DiffusionPipeline):
             pooled_prompt_embeds = torch.cat([negative_pooled_prompt_embeds, pooled_prompt_embeds], dim=0)
 
         # 3. Prepare timesteps
-        timesteps, num_inference_steps = retrieve_timesteps(self.scheduler, num_inference_steps, device, timesteps)
+        scheduler_kwargs = {}
+        if self.scheduler.config.get("use_dynamic_shifting", None) and mu is None:
+            image_seq_len = (int(height) // self.vae_scale_factor // self.transformer.config.patch_size) * (
+                int(width) // self.vae_scale_factor // self.transformer.config.patch_size
+            )
+            mu = calculate_shift(
+                image_seq_len,
+                self.scheduler.config.base_image_seq_len,
+                self.scheduler.config.max_image_seq_len,
+                self.scheduler.config.base_shift,
+                self.scheduler.config.max_shift,
+            )
+            scheduler_kwargs["mu"] = mu
+        elif mu is not None:
+            scheduler_kwargs["mu"] = mu
+        timesteps, num_inference_steps = retrieve_timesteps(
+            self.scheduler, num_inference_steps, device, sigmas=sigmas, **scheduler_kwargs
+        )
         timesteps, num_inference_steps = self.get_timesteps(num_inference_steps, strength, device)
         # check that number of inference steps is not < 1 - as this doesn't make sense
         if num_inference_steps < 1:

diffusers/pipelines/stable_diffusion_gligen/pipeline_stable_diffusion_gligen_text_image.py

@@ -446,13 +446,14 @@ class StableDiffusionGLIGENTextImagePipeline(DiffusionPipeline, StableDiffusionM
             extra_step_kwargs["generator"] = generator
         return extra_step_kwargs
 
-    # Copied from diffusers.pipelines.stable_diffusion_k_diffusion.pipeline_stable_diffusion_k_diffusion.StableDiffusionKDiffusionPipeline.check_inputs
     def check_inputs(
         self,
         prompt,
         height,
         width,
         callback_steps,
+        gligen_images,
+        gligen_phrases,
         negative_prompt=None,
         prompt_embeds=None,
         negative_prompt_embeds=None,
@@ -499,6 +500,13 @@ class StableDiffusionGLIGENTextImagePipeline(DiffusionPipeline, StableDiffusionM
                     f" {negative_prompt_embeds.shape}."
                 )
 
+        if gligen_images is not None and gligen_phrases is not None:
+            if len(gligen_images) != len(gligen_phrases):
+                raise ValueError(
+                    "`gligen_images` and `gligen_phrases` must have the same length when both are provided, but"
+                    f" got: `gligen_images` with length {len(gligen_images)} != `gligen_phrases` with length {len(gligen_phrases)}."
+                )
+
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_latents
     def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None):
         shape = (
@@ -814,6 +822,8 @@ class StableDiffusionGLIGENTextImagePipeline(DiffusionPipeline, StableDiffusionM
             height,
             width,
             callback_steps,
+            gligen_images,
+            gligen_phrases,
             negative_prompt,
             prompt_embeds,
             negative_prompt_embeds,

diffusers/pipelines/stable_diffusion_k_diffusion/pipeline_stable_diffusion_k_diffusion.py

@@ -602,9 +602,9 @@ class StableDiffusionKDiffusionPipeline(
             sigma_min: float = self.k_diffusion_model.sigmas[0].item()
             sigma_max: float = self.k_diffusion_model.sigmas[-1].item()
             sigmas = get_sigmas_karras(n=num_inference_steps, sigma_min=sigma_min, sigma_max=sigma_max)
-            sigmas = sigmas.to(device)
         else:
             sigmas = self.scheduler.sigmas
+        sigmas = sigmas.to(device)
         sigmas = sigmas.to(prompt_embeds.dtype)
 
         # 6. Prepare latent variables

diffusers/pipelines/stable_diffusion_ldm3d/pipeline_stable_diffusion_ldm3d.py

@@ -61,9 +61,21 @@ EXAMPLE_DOC_STRING = """
 
 # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.rescale_noise_cfg
 def rescale_noise_cfg(noise_cfg, noise_pred_text, guidance_rescale=0.0):
-    """
-    Rescale `noise_cfg` according to `guidance_rescale`. Based on findings of [Common Diffusion Noise Schedules and
-    Sample Steps are Flawed](https://arxiv.org/pdf/2305.08891.pdf). See Section 3.4
+    r"""
+    Rescales `noise_cfg` tensor based on `guidance_rescale` to improve image quality and fix overexposure. Based on
+    Section 3.4 from [Common Diffusion Noise Schedules and Sample Steps are
+    Flawed](https://arxiv.org/pdf/2305.08891.pdf).
+
+    Args:
+        noise_cfg (`torch.Tensor`):
+            The predicted noise tensor for the guided diffusion process.
+        noise_pred_text (`torch.Tensor`):
+            The predicted noise tensor for the text-guided diffusion process.
+        guidance_rescale (`float`, *optional*, defaults to 0.0):
+            A rescale factor applied to the noise predictions.
+
+    Returns:
+        noise_cfg (`torch.Tensor`): The rescaled noise prediction tensor.
     """
     std_text = noise_pred_text.std(dim=list(range(1, noise_pred_text.ndim)), keepdim=True)
     std_cfg = noise_cfg.std(dim=list(range(1, noise_cfg.ndim)), keepdim=True)
@@ -83,7 +95,7 @@ def retrieve_timesteps(
     sigmas: Optional[List[float]] = None,
     **kwargs,
 ):
-    """
+    r"""
     Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
     custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
 

diffusers/pipelines/stable_diffusion_panorama/pipeline_stable_diffusion_panorama.py

@@ -61,9 +61,21 @@ EXAMPLE_DOC_STRING = """
 
 # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.rescale_noise_cfg
 def rescale_noise_cfg(noise_cfg, noise_pred_text, guidance_rescale=0.0):
-    """
-    Rescale `noise_cfg` according to `guidance_rescale`. Based on findings of [Common Diffusion Noise Schedules and
-    Sample Steps are Flawed](https://arxiv.org/pdf/2305.08891.pdf). See Section 3.4
+    r"""
+    Rescales `noise_cfg` tensor based on `guidance_rescale` to improve image quality and fix overexposure. Based on
+    Section 3.4 from [Common Diffusion Noise Schedules and Sample Steps are
+    Flawed](https://arxiv.org/pdf/2305.08891.pdf).
+
+    Args:
+        noise_cfg (`torch.Tensor`):
+            The predicted noise tensor for the guided diffusion process.
+        noise_pred_text (`torch.Tensor`):
+            The predicted noise tensor for the text-guided diffusion process.
+        guidance_rescale (`float`, *optional*, defaults to 0.0):
+            A rescale factor applied to the noise predictions.
+
+    Returns:
+        noise_cfg (`torch.Tensor`): The rescaled noise prediction tensor.
     """
     std_text = noise_pred_text.std(dim=list(range(1, noise_pred_text.ndim)), keepdim=True)
     std_cfg = noise_cfg.std(dim=list(range(1, noise_cfg.ndim)), keepdim=True)
@@ -83,7 +95,7 @@ def retrieve_timesteps(
     sigmas: Optional[List[float]] = None,
     **kwargs,
 ):
-    """
+    r"""
     Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
     custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
 

diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py

@@ -87,9 +87,21 @@ EXAMPLE_DOC_STRING = """
 
 # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.rescale_noise_cfg
 def rescale_noise_cfg(noise_cfg, noise_pred_text, guidance_rescale=0.0):
-    """
-    Rescale `noise_cfg` according to `guidance_rescale`. Based on findings of [Common Diffusion Noise Schedules and
-    Sample Steps are Flawed](https://arxiv.org/pdf/2305.08891.pdf). See Section 3.4
+    r"""
+    Rescales `noise_cfg` tensor based on `guidance_rescale` to improve image quality and fix overexposure. Based on
+    Section 3.4 from [Common Diffusion Noise Schedules and Sample Steps are
+    Flawed](https://arxiv.org/pdf/2305.08891.pdf).
+
+    Args:
+        noise_cfg (`torch.Tensor`):
+            The predicted noise tensor for the guided diffusion process.
+        noise_pred_text (`torch.Tensor`):
+            The predicted noise tensor for the text-guided diffusion process.
+        guidance_rescale (`float`, *optional*, defaults to 0.0):
+            A rescale factor applied to the noise predictions.
+
+    Returns:
+        noise_cfg (`torch.Tensor`): The rescaled noise prediction tensor.
     """
     std_text = noise_pred_text.std(dim=list(range(1, noise_pred_text.ndim)), keepdim=True)
     std_cfg = noise_cfg.std(dim=list(range(1, noise_cfg.ndim)), keepdim=True)
@@ -109,7 +121,7 @@ def retrieve_timesteps(
     sigmas: Optional[List[float]] = None,
     **kwargs,
 ):
-    """
+    r"""
     Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
     custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
 
@@ -225,11 +237,8 @@ class StableDiffusionXLPipeline(
     _callback_tensor_inputs = [
         "latents",
         "prompt_embeds",
-        "negative_prompt_embeds",
         "add_text_embeds",
         "add_time_ids",
-        "negative_pooled_prompt_embeds",
-        "negative_add_time_ids",
     ]
 
     def __init__(
@@ -1231,13 +1240,8 @@ class StableDiffusionXLPipeline(
 
                     latents = callback_outputs.pop("latents", latents)
                     prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
-                    negative_prompt_embeds = callback_outputs.pop("negative_prompt_embeds", negative_prompt_embeds)
                     add_text_embeds = callback_outputs.pop("add_text_embeds", add_text_embeds)
-                    negative_pooled_prompt_embeds = callback_outputs.pop(
-                        "negative_pooled_prompt_embeds", negative_pooled_prompt_embeds
-                    )
                     add_time_ids = callback_outputs.pop("add_time_ids", add_time_ids)
-                    negative_add_time_ids = callback_outputs.pop("negative_add_time_ids", negative_add_time_ids)
 
                 # call the callback, if provided
                 if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):

diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_img2img.py

@@ -90,9 +90,21 @@ EXAMPLE_DOC_STRING = """
 
 # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.rescale_noise_cfg
 def rescale_noise_cfg(noise_cfg, noise_pred_text, guidance_rescale=0.0):
-    """
-    Rescale `noise_cfg` according to `guidance_rescale`. Based on findings of [Common Diffusion Noise Schedules and
-    Sample Steps are Flawed](https://arxiv.org/pdf/2305.08891.pdf). See Section 3.4
+    r"""
+    Rescales `noise_cfg` tensor based on `guidance_rescale` to improve image quality and fix overexposure. Based on
+    Section 3.4 from [Common Diffusion Noise Schedules and Sample Steps are
+    Flawed](https://arxiv.org/pdf/2305.08891.pdf).
+
+    Args:
+        noise_cfg (`torch.Tensor`):
+            The predicted noise tensor for the guided diffusion process.
+        noise_pred_text (`torch.Tensor`):
+            The predicted noise tensor for the text-guided diffusion process.
+        guidance_rescale (`float`, *optional*, defaults to 0.0):
+            A rescale factor applied to the noise predictions.
+
+    Returns:
+        noise_cfg (`torch.Tensor`): The rescaled noise prediction tensor.
     """
     std_text = noise_pred_text.std(dim=list(range(1, noise_pred_text.ndim)), keepdim=True)
     std_cfg = noise_cfg.std(dim=list(range(1, noise_cfg.ndim)), keepdim=True)
@@ -126,7 +138,7 @@ def retrieve_timesteps(
     sigmas: Optional[List[float]] = None,
     **kwargs,
 ):
-    """
+    r"""
     Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
     custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
 
@@ -245,11 +257,8 @@ class StableDiffusionXLImg2ImgPipeline(
     _callback_tensor_inputs = [
         "latents",
         "prompt_embeds",
-        "negative_prompt_embeds",
         "add_text_embeds",
         "add_time_ids",
-        "negative_pooled_prompt_embeds",
-        "add_neg_time_ids",
     ]
 
     def __init__(
@@ -640,14 +649,16 @@ class StableDiffusionXLImg2ImgPipeline(
         if denoising_start is None:
             init_timestep = min(int(num_inference_steps * strength), num_inference_steps)
             t_start = max(num_inference_steps - init_timestep, 0)
-        else:
-            t_start = 0
 
-        timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :]
+            timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :]
+            if hasattr(self.scheduler, "set_begin_index"):
+                self.scheduler.set_begin_index(t_start * self.scheduler.order)
+
+            return timesteps, num_inference_steps - t_start
 
-        # Strength is irrelevant if we directly request a timestep to start at;
-        # that is, strength is determined by the denoising_start instead.
-        if denoising_start is not None:
+        else:
+            # Strength is irrelevant if we directly request a timestep to start at;
+            # that is, strength is determined by the denoising_start instead.
             discrete_timestep_cutoff = int(
                 round(
                     self.scheduler.config.num_train_timesteps
@@ -655,7 +666,7 @@ class StableDiffusionXLImg2ImgPipeline(
                 )
             )
 
-            num_inference_steps = (timesteps < discrete_timestep_cutoff).sum().item()
+            num_inference_steps = (self.scheduler.timesteps < discrete_timestep_cutoff).sum().item()
             if self.scheduler.order == 2 and num_inference_steps % 2 == 0:
                 # if the scheduler is a 2nd order scheduler we might have to do +1
                 # because `num_inference_steps` might be even given that every timestep
@@ -666,11 +677,12 @@ class StableDiffusionXLImg2ImgPipeline(
                 num_inference_steps = num_inference_steps + 1
 
             # because t_n+1 >= t_n, we slice the timesteps starting from the end
-            timesteps = timesteps[-num_inference_steps:]
+            t_start = len(self.scheduler.timesteps) - num_inference_steps
+            timesteps = self.scheduler.timesteps[t_start:]
+            if hasattr(self.scheduler, "set_begin_index"):
+                self.scheduler.set_begin_index(t_start)
             return timesteps, num_inference_steps
 
-        return timesteps, num_inference_steps - t_start
-
     def prepare_latents(
         self, image, timestep, batch_size, num_images_per_prompt, dtype, device, generator=None, add_noise=True
     ):
@@ -1423,13 +1435,8 @@ class StableDiffusionXLImg2ImgPipeline(
 
                     latents = callback_outputs.pop("latents", latents)
                     prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
-                    negative_prompt_embeds = callback_outputs.pop("negative_prompt_embeds", negative_prompt_embeds)
                     add_text_embeds = callback_outputs.pop("add_text_embeds", add_text_embeds)
-                    negative_pooled_prompt_embeds = callback_outputs.pop(
-                        "negative_pooled_prompt_embeds", negative_pooled_prompt_embeds
-                    )
                     add_time_ids = callback_outputs.pop("add_time_ids", add_time_ids)
-                    add_neg_time_ids = callback_outputs.pop("add_neg_time_ids", add_neg_time_ids)
 
                 # call the callback, if provided
                 if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):

diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_inpaint.py

@@ -101,9 +101,21 @@ EXAMPLE_DOC_STRING = """
 
 # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.rescale_noise_cfg
 def rescale_noise_cfg(noise_cfg, noise_pred_text, guidance_rescale=0.0):
-    """
-    Rescale `noise_cfg` according to `guidance_rescale`. Based on findings of [Common Diffusion Noise Schedules and
-    Sample Steps are Flawed](https://arxiv.org/pdf/2305.08891.pdf). See Section 3.4
+    r"""
+    Rescales `noise_cfg` tensor based on `guidance_rescale` to improve image quality and fix overexposure. Based on
+    Section 3.4 from [Common Diffusion Noise Schedules and Sample Steps are
+    Flawed](https://arxiv.org/pdf/2305.08891.pdf).
+
+    Args:
+        noise_cfg (`torch.Tensor`):
+            The predicted noise tensor for the guided diffusion process.
+        noise_pred_text (`torch.Tensor`):
+            The predicted noise tensor for the text-guided diffusion process.
+        guidance_rescale (`float`, *optional*, defaults to 0.0):
+            A rescale factor applied to the noise predictions.
+
+    Returns:
+        noise_cfg (`torch.Tensor`): The rescaled noise prediction tensor.
     """
     std_text = noise_pred_text.std(dim=list(range(1, noise_pred_text.ndim)), keepdim=True)
     std_cfg = noise_cfg.std(dim=list(range(1, noise_cfg.ndim)), keepdim=True)
@@ -153,7 +165,7 @@ def retrieve_timesteps(
     sigmas: Optional[List[float]] = None,
     **kwargs,
 ):
-    """
+    r"""
     Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
     custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
 
@@ -273,11 +285,8 @@ class StableDiffusionXLInpaintPipeline(
     _callback_tensor_inputs = [
         "latents",
         "prompt_embeds",
-        "negative_prompt_embeds",
         "add_text_embeds",
         "add_time_ids",
-        "negative_pooled_prompt_embeds",
-        "add_neg_time_ids",
         "mask",
         "masked_image_latents",
     ]
@@ -901,14 +910,16 @@ class StableDiffusionXLInpaintPipeline(
         if denoising_start is None:
             init_timestep = min(int(num_inference_steps * strength), num_inference_steps)
             t_start = max(num_inference_steps - init_timestep, 0)
-        else:
-            t_start = 0
 
-        timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :]
+            timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :]
+            if hasattr(self.scheduler, "set_begin_index"):
+                self.scheduler.set_begin_index(t_start * self.scheduler.order)
+
+            return timesteps, num_inference_steps - t_start
 
-        # Strength is irrelevant if we directly request a timestep to start at;
-        # that is, strength is determined by the denoising_start instead.
-        if denoising_start is not None:
+        else:
+            # Strength is irrelevant if we directly request a timestep to start at;
+            # that is, strength is determined by the denoising_start instead.
             discrete_timestep_cutoff = int(
                 round(
                     self.scheduler.config.num_train_timesteps
@@ -916,7 +927,7 @@ class StableDiffusionXLInpaintPipeline(
                 )
             )
 
-            num_inference_steps = (timesteps < discrete_timestep_cutoff).sum().item()
+            num_inference_steps = (self.scheduler.timesteps < discrete_timestep_cutoff).sum().item()
             if self.scheduler.order == 2 and num_inference_steps % 2 == 0:
                 # if the scheduler is a 2nd order scheduler we might have to do +1
                 # because `num_inference_steps` might be even given that every timestep
@@ -927,11 +938,12 @@ class StableDiffusionXLInpaintPipeline(
                 num_inference_steps = num_inference_steps + 1
 
             # because t_n+1 >= t_n, we slice the timesteps starting from the end
-            timesteps = timesteps[-num_inference_steps:]
+            t_start = len(self.scheduler.timesteps) - num_inference_steps
+            timesteps = self.scheduler.timesteps[t_start:]
+            if hasattr(self.scheduler, "set_begin_index"):
+                self.scheduler.set_begin_index(t_start)
             return timesteps, num_inference_steps
 
-        return timesteps, num_inference_steps - t_start
-
     # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl_img2img.StableDiffusionXLImg2ImgPipeline._get_add_time_ids
     def _get_add_time_ids(
         self,
@@ -1656,13 +1668,8 @@ class StableDiffusionXLInpaintPipeline(
 
                     latents = callback_outputs.pop("latents", latents)
                     prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
-                    negative_prompt_embeds = callback_outputs.pop("negative_prompt_embeds", negative_prompt_embeds)
                     add_text_embeds = callback_outputs.pop("add_text_embeds", add_text_embeds)
-                    negative_pooled_prompt_embeds = callback_outputs.pop(
-                        "negative_pooled_prompt_embeds", negative_pooled_prompt_embeds
-                    )
                     add_time_ids = callback_outputs.pop("add_time_ids", add_time_ids)
-                    add_neg_time_ids = callback_outputs.pop("add_neg_time_ids", add_neg_time_ids)
                     mask = callback_outputs.pop("mask", mask)
                     masked_image_latents = callback_outputs.pop("masked_image_latents", masked_image_latents)
 

diffusers/pipelines/stable_video_diffusion/pipeline_stable_video_diffusion.py

@@ -71,7 +71,7 @@ def retrieve_timesteps(
     sigmas: Optional[List[float]] = None,
     **kwargs,
 ):
-    """
+    r"""
     Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
     custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
 

diffusers/pipelines/t2i_adapter/pipeline_stable_diffusion_adapter.py

@@ -127,7 +127,7 @@ def retrieve_timesteps(
     sigmas: Optional[List[float]] = None,
     **kwargs,
 ):
-    """
+    r"""
     Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
     custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
 

diffusers/pipelines/t2i_adapter/pipeline_stable_diffusion_xl_adapter.py

@@ -119,9 +119,21 @@ def _preprocess_adapter_image(image, height, width):
 
 # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.rescale_noise_cfg
 def rescale_noise_cfg(noise_cfg, noise_pred_text, guidance_rescale=0.0):
-    """
-    Rescale `noise_cfg` according to `guidance_rescale`. Based on findings of [Common Diffusion Noise Schedules and
-    Sample Steps are Flawed](https://arxiv.org/pdf/2305.08891.pdf). See Section 3.4
+    r"""
+    Rescales `noise_cfg` tensor based on `guidance_rescale` to improve image quality and fix overexposure. Based on
+    Section 3.4 from [Common Diffusion Noise Schedules and Sample Steps are
+    Flawed](https://arxiv.org/pdf/2305.08891.pdf).
+
+    Args:
+        noise_cfg (`torch.Tensor`):
+            The predicted noise tensor for the guided diffusion process.
+        noise_pred_text (`torch.Tensor`):
+            The predicted noise tensor for the text-guided diffusion process.
+        guidance_rescale (`float`, *optional*, defaults to 0.0):
+            A rescale factor applied to the noise predictions.
+
+    Returns:
+        noise_cfg (`torch.Tensor`): The rescaled noise prediction tensor.
     """
     std_text = noise_pred_text.std(dim=list(range(1, noise_pred_text.ndim)), keepdim=True)
     std_cfg = noise_cfg.std(dim=list(range(1, noise_cfg.ndim)), keepdim=True)
@@ -141,7 +153,7 @@ def retrieve_timesteps(
     sigmas: Optional[List[float]] = None,
     **kwargs,
 ):
-    """
+    r"""
     Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
     custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
 

diffusers/pipelines/text_to_video_synthesis/pipeline_text_to_video_zero_sdxl.py

@@ -310,9 +310,21 @@ def create_motion_field_and_warp_latents(motion_field_strength_x, motion_field_s
 
 # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.rescale_noise_cfg
 def rescale_noise_cfg(noise_cfg, noise_pred_text, guidance_rescale=0.0):
-    """
-    Rescale `noise_cfg` according to `guidance_rescale`. Based on findings of [Common Diffusion Noise Schedules and
-    Sample Steps are Flawed](https://arxiv.org/pdf/2305.08891.pdf). See Section 3.4
+    r"""
+    Rescales `noise_cfg` tensor based on `guidance_rescale` to improve image quality and fix overexposure. Based on
+    Section 3.4 from [Common Diffusion Noise Schedules and Sample Steps are
+    Flawed](https://arxiv.org/pdf/2305.08891.pdf).
+
+    Args:
+        noise_cfg (`torch.Tensor`):
+            The predicted noise tensor for the guided diffusion process.
+        noise_pred_text (`torch.Tensor`):
+            The predicted noise tensor for the text-guided diffusion process.
+        guidance_rescale (`float`, *optional*, defaults to 0.0):
+            A rescale factor applied to the noise predictions.
+
+    Returns:
+        noise_cfg (`torch.Tensor`): The rescaled noise prediction tensor.
     """
     std_text = noise_pred_text.std(dim=list(range(1, noise_pred_text.ndim)), keepdim=True)
     std_cfg = noise_cfg.std(dim=list(range(1, noise_cfg.ndim)), keepdim=True)

diffusers/pipelines/unidiffuser/modeling_uvit.py

@@ -104,8 +104,8 @@ class PatchEmbed(nn.Module):
 
         self.use_pos_embed = use_pos_embed
         if self.use_pos_embed:
-            pos_embed = get_2d_sincos_pos_embed(embed_dim, int(num_patches**0.5))
-            self.register_buffer("pos_embed", torch.from_numpy(pos_embed).float().unsqueeze(0), persistent=False)
+            pos_embed = get_2d_sincos_pos_embed(embed_dim, int(num_patches**0.5), output_type="pt")
+            self.register_buffer("pos_embed", pos_embed.float().unsqueeze(0), persistent=False)
 
     def forward(self, latent):
         latent = self.proj(latent)