diffusers 0.30.2__py3-none-any.whl → 0.31.0__py3-none-any.whl

diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py

@@ -28,6 +28,7 @@ from ...schedulers import KarrasDiffusionSchedulers
 from ...utils import (
     USE_PEFT_BACKEND,
     deprecate,
+    is_torch_xla_available,
     logging,
     replace_example_docstring,
     scale_lora_layers,
@@ -39,6 +40,13 @@ from .pipeline_output import StableDiffusionPipelineOutput
 from .safety_checker import StableDiffusionSafetyChecker
 
 
+if is_torch_xla_available():
+    import torch_xla.core.xla_model as xm
+
+    XLA_AVAILABLE = True
+else:
+    XLA_AVAILABLE = False
+
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
 
 EXAMPLE_DOC_STRING = """
@@ -57,9 +65,21 @@ EXAMPLE_DOC_STRING = """
 
 
 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)
@@ -78,7 +98,7 @@ 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`.
 
@@ -137,7 +157,7 @@ class StableDiffusionPipeline(
     IPAdapterMixin,
     FromSingleFileMixin,
 ):
-    r"""
+    """
     Pipeline for text-to-image generation using Stable Diffusion.
 
     This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods
@@ -1036,6 +1056,9 @@ class StableDiffusionPipeline(
                         step_idx = i // getattr(self.scheduler, "order", 1)
                         callback(step_idx, t, latents)
 
+                if XLA_AVAILABLE:
+                    xm.mark_step()
+
         if not output_type == "latent":
             image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False, generator=generator)[
                 0
@@ -1049,7 +1072,6 @@ class StableDiffusionPipeline(
             do_denormalize = [True] * image.shape[0]
         else:
             do_denormalize = [not has_nsfw for has_nsfw in has_nsfw_concept]
-
         image = self.image_processor.postprocess(image, output_type=output_type, do_denormalize=do_denormalize)
 
         # Offload all models

diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_img2img.py

@@ -119,7 +119,7 @@ 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/pipeline_stable_diffusion_inpaint.py

@@ -60,7 +60,7 @@ 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/pipeline_stable_diffusion_latent_upscale.py

@@ -33,6 +33,20 @@ from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput, StableDiffu
 logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
 
 
+# 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"
+):
+    if hasattr(encoder_output, "latent_dist") and sample_mode == "sample":
+        return encoder_output.latent_dist.sample(generator)
+    elif hasattr(encoder_output, "latent_dist") and sample_mode == "argmax":
+        return encoder_output.latent_dist.mode()
+    elif hasattr(encoder_output, "latents"):
+        return encoder_output.latents
+    else:
+        raise AttributeError("Could not access latents of provided encoder_output")
+
+
 # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.preprocess
 def preprocess(image):
     warnings.warn(
@@ -105,7 +119,54 @@ class StableDiffusionLatentUpscalePipeline(DiffusionPipeline, StableDiffusionMix
         self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)
         self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor, resample="bicubic")
 
-    def _encode_prompt(self, prompt, device, do_classifier_free_guidance, negative_prompt):
+    def _encode_prompt(
+        self,
+        prompt,
+        device,
+        do_classifier_free_guidance,
+        negative_prompt=None,
+        prompt_embeds: Optional[torch.Tensor] = None,
+        negative_prompt_embeds: Optional[torch.Tensor] = None,
+        pooled_prompt_embeds: Optional[torch.Tensor] = None,
+        negative_pooled_prompt_embeds: Optional[torch.Tensor] = None,
+        **kwargs,
+    ):
+        deprecation_message = "`_encode_prompt()` is deprecated and it will be removed in a future version. Use `encode_prompt()` instead. Also, be aware that the output format changed from a concatenated tensor to a tuple."
+        deprecate("_encode_prompt()", "1.0.0", deprecation_message, standard_warn=False)
+
+        (
+            prompt_embeds,
+            negative_prompt_embeds,
+            pooled_prompt_embeds,
+            negative_pooled_prompt_embeds,
+        ) = self.encode_prompt(
+            prompt=prompt,
+            device=device,
+            do_classifier_free_guidance=do_classifier_free_guidance,
+            negative_prompt=negative_prompt,
+            prompt_embeds=prompt_embeds,
+            negative_prompt_embeds=negative_prompt_embeds,
+            pooled_prompt_embeds=pooled_prompt_embeds,
+            negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
+            **kwargs,
+        )
+
+        prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds])
+        pooled_prompt_embeds = torch.cat([negative_pooled_prompt_embeds, pooled_prompt_embeds])
+
+        return prompt_embeds, pooled_prompt_embeds
+
+    def encode_prompt(
+        self,
+        prompt,
+        device,
+        do_classifier_free_guidance,
+        negative_prompt=None,
+        prompt_embeds: Optional[torch.Tensor] = None,
+        negative_prompt_embeds: Optional[torch.Tensor] = None,
+        pooled_prompt_embeds: Optional[torch.Tensor] = None,
+        negative_pooled_prompt_embeds: Optional[torch.Tensor] = None,
+    ):
         r"""
         Encodes the prompt into text encoder hidden states.
 
@@ -119,81 +180,100 @@ class StableDiffusionLatentUpscalePipeline(DiffusionPipeline, StableDiffusionMix
             negative_prompt (`str` or `List[str]`):
                 The prompt or prompts not to guide the image generation. Ignored when not using guidance (i.e., ignored
                 if `guidance_scale` is less than `1`).
+            prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+                provided, text embeddings will be generated from `prompt` input argument.
+            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+                weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+                argument.
+            pooled_prompt_embeds (`torch.Tensor`, *optional*):
+                Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
+                If not provided, pooled text embeddings will be generated from `prompt` input argument.
+            negative_pooled_prompt_embeds (`torch.Tensor`, *optional*):
+                Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+                weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
+                input argument.
         """
-        batch_size = len(prompt) if isinstance(prompt, list) else 1
-
-        text_inputs = self.tokenizer(
-            prompt,
-            padding="max_length",
-            max_length=self.tokenizer.model_max_length,
-            truncation=True,
-            return_length=True,
-            return_tensors="pt",
-        )
-        text_input_ids = text_inputs.input_ids
-
-        untruncated_ids = self.tokenizer(prompt, padding="longest", return_tensors="pt").input_ids
-
-        if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(text_input_ids, untruncated_ids):
-            removed_text = self.tokenizer.batch_decode(untruncated_ids[:, self.tokenizer.model_max_length - 1 : -1])
-            logger.warning(
-                "The following part of your input was truncated because CLIP can only handle sequences up to"
-                f" {self.tokenizer.model_max_length} tokens: {removed_text}"
-            )
-
-        text_encoder_out = self.text_encoder(
-            text_input_ids.to(device),
-            output_hidden_states=True,
-        )
-        text_embeddings = text_encoder_out.hidden_states[-1]
-        text_pooler_out = text_encoder_out.pooler_output
-
-        # get unconditional embeddings for classifier free guidance
-        if do_classifier_free_guidance:
-            uncond_tokens: List[str]
-            if negative_prompt is None:
-                uncond_tokens = [""] * batch_size
-            elif type(prompt) is not type(negative_prompt):
-                raise TypeError(
-                    f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
-                    f" {type(prompt)}."
-                )
-            elif isinstance(negative_prompt, str):
-                uncond_tokens = [negative_prompt]
-            elif batch_size != len(negative_prompt):
-                raise ValueError(
-                    f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
-                    f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
-                    " the batch size of `prompt`."
-                )
-            else:
-                uncond_tokens = negative_prompt
+        if prompt is not None and isinstance(prompt, str):
+            batch_size = 1
+        elif prompt is not None and isinstance(prompt, list):
+            batch_size = len(prompt)
+        else:
+            batch_size = prompt_embeds.shape[0]
 
-            max_length = text_input_ids.shape[-1]
-            uncond_input = self.tokenizer(
-                uncond_tokens,
+        if prompt_embeds is None or pooled_prompt_embeds is None:
+            text_inputs = self.tokenizer(
+                prompt,
                 padding="max_length",
-                max_length=max_length,
+                max_length=self.tokenizer.model_max_length,
                 truncation=True,
                 return_length=True,
                 return_tensors="pt",
             )
+            text_input_ids = text_inputs.input_ids
 
-            uncond_encoder_out = self.text_encoder(
-                uncond_input.input_ids.to(device),
+            untruncated_ids = self.tokenizer(prompt, padding="longest", return_tensors="pt").input_ids
+
+            if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(
+                text_input_ids, untruncated_ids
+            ):
+                removed_text = self.tokenizer.batch_decode(
+                    untruncated_ids[:, self.tokenizer.model_max_length - 1 : -1]
+                )
+                logger.warning(
+                    "The following part of your input was truncated because CLIP can only handle sequences up to"
+                    f" {self.tokenizer.model_max_length} tokens: {removed_text}"
+                )
+
+            text_encoder_out = self.text_encoder(
+                text_input_ids.to(device),
                 output_hidden_states=True,
             )
+            prompt_embeds = text_encoder_out.hidden_states[-1]
+            pooled_prompt_embeds = text_encoder_out.pooler_output
 
-            uncond_embeddings = uncond_encoder_out.hidden_states[-1]
-            uncond_pooler_out = uncond_encoder_out.pooler_output
+        # get unconditional embeddings for classifier free guidance
+        if do_classifier_free_guidance:
+            if negative_prompt_embeds is None or negative_pooled_prompt_embeds is None:
+                uncond_tokens: List[str]
+                if negative_prompt is None:
+                    uncond_tokens = [""] * batch_size
+                elif type(prompt) is not type(negative_prompt):
+                    raise TypeError(
+                        f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
+                        f" {type(prompt)}."
+                    )
+                elif isinstance(negative_prompt, str):
+                    uncond_tokens = [negative_prompt]
+                elif batch_size != len(negative_prompt):
+                    raise ValueError(
+                        f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
+                        f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
+                        " the batch size of `prompt`."
+                    )
+                else:
+                    uncond_tokens = negative_prompt
+
+                max_length = text_input_ids.shape[-1]
+                uncond_input = self.tokenizer(
+                    uncond_tokens,
+                    padding="max_length",
+                    max_length=max_length,
+                    truncation=True,
+                    return_length=True,
+                    return_tensors="pt",
+                )
+
+                uncond_encoder_out = self.text_encoder(
+                    uncond_input.input_ids.to(device),
+                    output_hidden_states=True,
+                )
 
-            # For classifier free guidance, we need to do two forward passes.
-            # Here we concatenate the unconditional and text embeddings into a single batch
-            # to avoid doing two forward passes
-            text_embeddings = torch.cat([uncond_embeddings, text_embeddings])
-            text_pooler_out = torch.cat([uncond_pooler_out, text_pooler_out])
+                negative_prompt_embeds = uncond_encoder_out.hidden_states[-1]
+                negative_pooled_prompt_embeds = uncond_encoder_out.pooler_output
 
-        return text_embeddings, text_pooler_out
+        return prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds
 
     # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.decode_latents
     def decode_latents(self, latents):
@@ -207,12 +287,56 @@ class StableDiffusionLatentUpscalePipeline(DiffusionPipeline, StableDiffusionMix
         image = image.cpu().permute(0, 2, 3, 1).float().numpy()
         return image
 
-    def check_inputs(self, prompt, image, callback_steps):
-        if not isinstance(prompt, str) and not isinstance(prompt, list):
+    def check_inputs(
+        self,
+        prompt,
+        image,
+        callback_steps,
+        negative_prompt=None,
+        prompt_embeds=None,
+        negative_prompt_embeds=None,
+        pooled_prompt_embeds=None,
+        negative_pooled_prompt_embeds=None,
+    ):
+        if prompt is not None and prompt_embeds is not None:
+            raise ValueError(
+                f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+                " only forward one of the two."
+            )
+        elif prompt is None and prompt_embeds is None:
+            raise ValueError(
+                "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
+            )
+        elif prompt is not None and not isinstance(prompt, str) and not isinstance(prompt, list):
             raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
 
+        if negative_prompt is not None and negative_prompt_embeds is not None:
+            raise ValueError(
+                f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
+                f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+            )
+
+        if prompt_embeds is not None and negative_prompt_embeds is not None:
+            if prompt_embeds.shape != negative_prompt_embeds.shape:
+                raise ValueError(
+                    "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but"
+                    f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`"
+                    f" {negative_prompt_embeds.shape}."
+                )
+
+        if prompt_embeds is not None and pooled_prompt_embeds is None:
+            raise ValueError(
+                "If `prompt_embeds` are provided, `pooled_prompt_embeds` also have to be passed. Make sure to generate `pooled_prompt_embeds` from the same text encoder that was used to generate `prompt_embeds`."
+            )
+
+        if negative_prompt_embeds is not None and negative_pooled_prompt_embeds is None:
+            raise ValueError(
+                "If `negative_prompt_embeds` are provided, `negative_pooled_prompt_embeds` also have to be passed. Make sure to generate `negative_pooled_prompt_embeds` from the same text encoder that was used to generate `negative_prompt_embeds`."
+            )
+
         if (
             not isinstance(image, torch.Tensor)
+            and not isinstance(image, np.ndarray)
             and not isinstance(image, PIL.Image.Image)
             and not isinstance(image, list)
         ):
@@ -222,10 +346,14 @@ class StableDiffusionLatentUpscalePipeline(DiffusionPipeline, StableDiffusionMix
 
         # verify batch size of prompt and image are same if image is a list or tensor
         if isinstance(image, (list, torch.Tensor)):
-            if isinstance(prompt, str):
-                batch_size = 1
+            if prompt is not None:
+                if isinstance(prompt, str):
+                    batch_size = 1
+                else:
+                    batch_size = len(prompt)
             else:
-                batch_size = len(prompt)
+                batch_size = prompt_embeds.shape[0]
+
             if isinstance(image, list):
                 image_batch_size = len(image)
             else:
@@ -261,13 +389,17 @@ class StableDiffusionLatentUpscalePipeline(DiffusionPipeline, StableDiffusionMix
     @torch.no_grad()
     def __call__(
         self,
-        prompt: Union[str, List[str]],
+        prompt: Union[str, List[str]] = None,
         image: PipelineImageInput = None,
         num_inference_steps: int = 75,
         guidance_scale: float = 9.0,
         negative_prompt: Optional[Union[str, List[str]]] = None,
         generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
         latents: Optional[torch.Tensor] = None,
+        prompt_embeds: Optional[torch.Tensor] = None,
+        negative_prompt_embeds: Optional[torch.Tensor] = None,
+        pooled_prompt_embeds: Optional[torch.Tensor] = None,
+        negative_pooled_prompt_embeds: Optional[torch.Tensor] = None,
         output_type: Optional[str] = "pil",
         return_dict: bool = True,
         callback: Optional[Callable[[int, int, torch.Tensor], None]] = None,
@@ -359,10 +491,22 @@ class StableDiffusionLatentUpscalePipeline(DiffusionPipeline, StableDiffusionMix
         """
 
         # 1. Check inputs
-        self.check_inputs(prompt, image, callback_steps)
+        self.check_inputs(
+            prompt,
+            image,
+            callback_steps,
+            negative_prompt,
+            prompt_embeds,
+            negative_prompt_embeds,
+            pooled_prompt_embeds,
+            negative_pooled_prompt_embeds,
+        )
 
         # 2. Define call parameters
-        batch_size = 1 if isinstance(prompt, str) else len(prompt)
+        if prompt is not None:
+            batch_size = 1 if isinstance(prompt, str) else len(prompt)
+        else:
+            batch_size = prompt_embeds.shape[0]
         device = self._execution_device
         # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
         # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
@@ -373,16 +517,32 @@ class StableDiffusionLatentUpscalePipeline(DiffusionPipeline, StableDiffusionMix
             prompt = [""] * batch_size
 
         # 3. Encode input prompt
-        text_embeddings, text_pooler_out = self._encode_prompt(
-            prompt, device, do_classifier_free_guidance, negative_prompt
+        (
+            prompt_embeds,
+            negative_prompt_embeds,
+            pooled_prompt_embeds,
+            negative_pooled_prompt_embeds,
+        ) = self.encode_prompt(
+            prompt,
+            device,
+            do_classifier_free_guidance,
+            negative_prompt,
+            prompt_embeds,
+            negative_prompt_embeds,
+            pooled_prompt_embeds,
+            negative_pooled_prompt_embeds,
         )
 
+        if do_classifier_free_guidance:
+            prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds])
+            pooled_prompt_embeds = torch.cat([negative_pooled_prompt_embeds, pooled_prompt_embeds])
+
         # 4. Preprocess image
         image = self.image_processor.preprocess(image)
-        image = image.to(dtype=text_embeddings.dtype, device=device)
+        image = image.to(dtype=prompt_embeds.dtype, device=device)
         if image.shape[1] == 3:
             # encode image if not in latent-space yet
-            image = self.vae.encode(image).latent_dist.sample() * self.vae.config.scaling_factor
+            image = retrieve_latents(self.vae.encode(image), generator=generator) * self.vae.config.scaling_factor
 
         # 5. set timesteps
         self.scheduler.set_timesteps(num_inference_steps, device=device)
@@ -400,17 +560,17 @@ class StableDiffusionLatentUpscalePipeline(DiffusionPipeline, StableDiffusionMix
         inv_noise_level = (noise_level**2 + 1) ** (-0.5)
 
         image_cond = F.interpolate(image, scale_factor=2, mode="nearest") * inv_noise_level[:, None, None, None]
-        image_cond = image_cond.to(text_embeddings.dtype)
+        image_cond = image_cond.to(prompt_embeds.dtype)
 
         noise_level_embed = torch.cat(
             [
-                torch.ones(text_pooler_out.shape[0], 64, dtype=text_pooler_out.dtype, device=device),
-                torch.zeros(text_pooler_out.shape[0], 64, dtype=text_pooler_out.dtype, device=device),
+                torch.ones(pooled_prompt_embeds.shape[0], 64, dtype=pooled_prompt_embeds.dtype, device=device),
+                torch.zeros(pooled_prompt_embeds.shape[0], 64, dtype=pooled_prompt_embeds.dtype, device=device),
             ],
             dim=1,
         )
 
-        timestep_condition = torch.cat([noise_level_embed, text_pooler_out], dim=1)
+        timestep_condition = torch.cat([noise_level_embed, pooled_prompt_embeds], dim=1)
 
         # 6. Prepare latent variables
         height, width = image.shape[2:]
@@ -420,7 +580,7 @@ class StableDiffusionLatentUpscalePipeline(DiffusionPipeline, StableDiffusionMix
             num_channels_latents,
             height * 2,  # 2x upscale
             width * 2,
-            text_embeddings.dtype,
+            prompt_embeds.dtype,
             device,
             generator,
             latents,
@@ -454,7 +614,7 @@ class StableDiffusionLatentUpscalePipeline(DiffusionPipeline, StableDiffusionMix
                 noise_pred = self.unet(
                     scaled_model_input,
                     timestep,
-                    encoder_hidden_states=text_embeddings,
+                    encoder_hidden_states=prompt_embeds,
                     timestep_cond=timestep_condition,
                 ).sample
 

diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3.py

@@ -77,7 +77,7 @@ 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`.
 
@@ -203,6 +203,9 @@ class StableDiffusion3Pipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingle
             if hasattr(self, "transformer") and self.transformer is not None
             else 128
         )
+        self.patch_size = (
+            self.transformer.config.patch_size if hasattr(self, "transformer") and self.transformer is not None else 2
+        )
 
     def _get_t5_prompt_embeds(
         self,
@@ -525,8 +528,14 @@ class StableDiffusion3Pipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingle
         callback_on_step_end_tensor_inputs=None,
         max_sequence_length=None,
     ):
-        if height % 8 != 0 or width % 8 != 0:
-            raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.")
+        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 callback_on_step_end_tensor_inputs is not None and not all(
             k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs

diffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3_img2img.py

@@ -13,7 +13,7 @@
 # limitations under the License.
 
 import inspect
-from typing import Callable, Dict, List, Optional, Union
+from typing import Any, Callable, Dict, List, Optional, Union
 
 import PIL.Image
 import torch
@@ -25,7 +25,7 @@ from transformers import (
 )
 
 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
@@ -98,7 +98,7 @@ 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`.
 
@@ -149,7 +149,7 @@ def retrieve_timesteps(
     return timesteps, num_inference_steps
 
 
-class StableDiffusion3Img2ImgPipeline(DiffusionPipeline):
+class StableDiffusion3Img2ImgPipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSingleFileMixin):
     r"""
     Args:
         transformer ([`SD3Transformer2DModel`]):
@@ -680,6 +680,10 @@ class StableDiffusion3Img2ImgPipeline(DiffusionPipeline):
     def guidance_scale(self):
         return self._guidance_scale
 
+    @property
+    def joint_attention_kwargs(self):
+        return self._joint_attention_kwargs
+
     @property
     def clip_skip(self):
         return self._clip_skip
@@ -723,6 +727,7 @@ class StableDiffusion3Img2ImgPipeline(DiffusionPipeline):
         negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
         output_type: Optional[str] = "pil",
         return_dict: bool = True,
+        joint_attention_kwargs: Optional[Dict[str, Any]] = None,
         clip_skip: Optional[int] = None,
         callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
         callback_on_step_end_tensor_inputs: List[str] = ["latents"],
@@ -797,6 +802,10 @@ class StableDiffusion3Img2ImgPipeline(DiffusionPipeline):
             return_dict (`bool`, *optional*, defaults to `True`):
                 Whether or not to return a [`~pipelines.stable_diffusion_xl.StableDiffusionXLPipelineOutput`] instead
                 of a plain tuple.
+            joint_attention_kwargs (`dict`, *optional*):
+               A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
+                `self.processor` in
+                [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
             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,
@@ -835,6 +844,7 @@ class StableDiffusion3Img2ImgPipeline(DiffusionPipeline):
 
         self._guidance_scale = guidance_scale
         self._clip_skip = clip_skip
+        self._joint_attention_kwargs = joint_attention_kwargs
         self._interrupt = False
 
         # 2. Define call parameters
@@ -847,6 +857,10 @@ class StableDiffusion3Img2ImgPipeline(DiffusionPipeline):
 
         device = self._execution_device
 
+        lora_scale = (
+            self.joint_attention_kwargs.get("scale", None) if self.joint_attention_kwargs is not None else None
+        )
+
         (
             prompt_embeds,
             negative_prompt_embeds,
@@ -868,6 +882,7 @@ class StableDiffusion3Img2ImgPipeline(DiffusionPipeline):
             clip_skip=self.clip_skip,
             num_images_per_prompt=num_images_per_prompt,
             max_sequence_length=max_sequence_length,
+            lora_scale=lora_scale,
         )
 
         if self.do_classifier_free_guidance:
@@ -912,6 +927,7 @@ class StableDiffusion3Img2ImgPipeline(DiffusionPipeline):
                     timestep=timestep,
                     encoder_hidden_states=prompt_embeds,
                     pooled_projections=pooled_prompt_embeds,
+                    joint_attention_kwargs=self.joint_attention_kwargs,
                     return_dict=False,
                 )[0]