diffusers 0.32.1__py3-none-any.whl → 0.33.0__py3-none-any.whl

diffusers/pipelines/consisid/pipeline_output.py

@@ -0,0 +1,20 @@
+from dataclasses import dataclass
+
+import torch
+
+from diffusers.utils import BaseOutput
+
+
+@dataclass
+class ConsisIDPipelineOutput(BaseOutput):
+    r"""
+    Output class for ConsisID pipelines.
+
+    Args:
+        frames (`torch.Tensor`, `np.ndarray`, or List[List[PIL.Image.Image]]):
+            List of video outputs - It can be a nested list of length `batch_size,` with each sub-list containing
+            denoised PIL image sequences of length `num_frames.` It can also be a NumPy array or Torch tensor of shape
+            `(batch_size, num_frames, channels, height, width)`.
+    """
+
+    frames: torch.Tensor

diffusers/pipelines/consistency_models/pipeline_consistency_models.py

@@ -19,6 +19,7 @@ import torch
 from ...models import UNet2DModel
 from ...schedulers import CMStochasticIterativeScheduler
 from ...utils import (
+    is_torch_xla_available,
     logging,
     replace_example_docstring,
 )
@@ -26,6 +27,13 @@ from ...utils.torch_utils import randn_tensor
 from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput
 
 
+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
 
 
@@ -263,6 +271,9 @@ class ConsistencyModelPipeline(DiffusionPipeline):
                 if callback is not None and i % callback_steps == 0:
                     callback(i, t, sample)
 
+                if XLA_AVAILABLE:
+                    xm.mark_step()
+
         # 6. Post-process image sample
         image = self.postprocess_image(sample, output_type=output_type)
 

diffusers/pipelines/controlnet/pipeline_controlnet.py

@@ -80,7 +80,7 @@ EXAMPLE_DOC_STRING = """
         >>> # load control net and stable diffusion v1-5
         >>> controlnet = ControlNetModel.from_pretrained("lllyasviel/sd-controlnet-canny", torch_dtype=torch.float16)
         >>> pipe = StableDiffusionControlNetPipeline.from_pretrained(
-        ...     "runwayml/stable-diffusion-v1-5", controlnet=controlnet, torch_dtype=torch.float16
+        ...     "stable-diffusion-v1-5/stable-diffusion-v1-5", controlnet=controlnet, torch_dtype=torch.float16
         ... )
 
         >>> # speed up diffusion process with faster scheduler and memory optimization
@@ -198,8 +198,8 @@ class StableDiffusionControlNetPipeline(
             [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`].
         safety_checker ([`StableDiffusionSafetyChecker`]):
             Classification module that estimates whether generated images could be considered offensive or harmful.
-            Please refer to the [model card](https://huggingface.co/runwayml/stable-diffusion-v1-5) for more details
-            about a model's potential harms.
+            Please refer to the [model card](https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5) for
+            more details about a model's potential harms.
         feature_extractor ([`~transformers.CLIPImageProcessor`]):
             A `CLIPImageProcessor` to extract features from generated images; used as inputs to the `safety_checker`.
     """
@@ -207,7 +207,7 @@ class StableDiffusionControlNetPipeline(
     model_cpu_offload_seq = "text_encoder->image_encoder->unet->vae"
     _optional_components = ["safety_checker", "feature_extractor", "image_encoder"]
     _exclude_from_cpu_offload = ["safety_checker"]
-    _callback_tensor_inputs = ["latents", "prompt_embeds", "negative_prompt_embeds"]
+    _callback_tensor_inputs = ["latents", "prompt_embeds", "negative_prompt_embeds", "image"]
 
     def __init__(
         self,
@@ -254,7 +254,7 @@ class StableDiffusionControlNetPipeline(
             feature_extractor=feature_extractor,
             image_encoder=image_encoder,
         )
-        self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)
+        self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) if getattr(self, "vae", None) else 8
         self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True)
         self.control_image_processor = VaeImageProcessor(
             vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True, do_normalize=False
@@ -1323,6 +1323,7 @@ class StableDiffusionControlNetPipeline(
                     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)
+                    image = callback_outputs.pop("image", image)
 
                 # 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/controlnet/pipeline_controlnet_blip_diffusion.py

@@ -21,6 +21,7 @@ from transformers import CLIPTokenizer
 from ...models import AutoencoderKL, ControlNetModel, UNet2DConditionModel
 from ...schedulers import PNDMScheduler
 from ...utils import (
+    is_torch_xla_available,
     logging,
     replace_example_docstring,
 )
@@ -31,8 +32,16 @@ from ..blip_diffusion.modeling_ctx_clip import ContextCLIPTextModel
 from ..pipeline_utils import DiffusionPipeline, ImagePipelineOutput
 
 
+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 = """
     Examples:
         ```py
@@ -401,6 +410,10 @@ class BlipDiffusionControlNetPipeline(DiffusionPipeline):
                 t,
                 latents,
             )["prev_sample"]
+
+            if XLA_AVAILABLE:
+                xm.mark_step()
+
         image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False)[0]
         image = self.image_processor.postprocess(image, output_type=output_type)
 

diffusers/pipelines/controlnet/pipeline_controlnet_img2img.py

@@ -30,6 +30,7 @@ from ...schedulers import KarrasDiffusionSchedulers
 from ...utils import (
     USE_PEFT_BACKEND,
     deprecate,
+    is_torch_xla_available,
     logging,
     replace_example_docstring,
     scale_lora_layers,
@@ -41,6 +42,13 @@ from ..stable_diffusion import StableDiffusionPipelineOutput
 from ..stable_diffusion.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
 
 
@@ -71,7 +79,7 @@ EXAMPLE_DOC_STRING = """
         >>> # load control net and stable diffusion v1-5
         >>> controlnet = ControlNetModel.from_pretrained("lllyasviel/sd-controlnet-canny", torch_dtype=torch.float16)
         >>> pipe = StableDiffusionControlNetImg2ImgPipeline.from_pretrained(
-        ...     "runwayml/stable-diffusion-v1-5", controlnet=controlnet, torch_dtype=torch.float16
+        ...     "stable-diffusion-v1-5/stable-diffusion-v1-5", controlnet=controlnet, torch_dtype=torch.float16
         ... )
 
         >>> # speed up diffusion process with faster scheduler and memory optimization
@@ -168,8 +176,8 @@ class StableDiffusionControlNetImg2ImgPipeline(
             [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`].
         safety_checker ([`StableDiffusionSafetyChecker`]):
             Classification module that estimates whether generated images could be considered offensive or harmful.
-            Please refer to the [model card](https://huggingface.co/runwayml/stable-diffusion-v1-5) for more details
-            about a model's potential harms.
+            Please refer to the [model card](https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5) for
+            more details about a model's potential harms.
         feature_extractor ([`~transformers.CLIPImageProcessor`]):
             A `CLIPImageProcessor` to extract features from generated images; used as inputs to the `safety_checker`.
     """
@@ -177,7 +185,7 @@ class StableDiffusionControlNetImg2ImgPipeline(
     model_cpu_offload_seq = "text_encoder->unet->vae"
     _optional_components = ["safety_checker", "feature_extractor", "image_encoder"]
     _exclude_from_cpu_offload = ["safety_checker"]
-    _callback_tensor_inputs = ["latents", "prompt_embeds", "negative_prompt_embeds"]
+    _callback_tensor_inputs = ["latents", "prompt_embeds", "negative_prompt_embeds", "control_image"]
 
     def __init__(
         self,
@@ -224,7 +232,7 @@ class StableDiffusionControlNetImg2ImgPipeline(
             feature_extractor=feature_extractor,
             image_encoder=image_encoder,
         )
-        self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)
+        self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) if getattr(self, "vae", None) else 8
         self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True)
         self.control_image_processor = VaeImageProcessor(
             vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True, do_normalize=False
@@ -1286,6 +1294,7 @@ class StableDiffusionControlNetImg2ImgPipeline(
                     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)
+                    control_image = callback_outputs.pop("control_image", control_image)
 
                 # call the callback, if provided
                 if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
@@ -1294,6 +1303,9 @@ class StableDiffusionControlNetImg2ImgPipeline(
                         step_idx = i // getattr(self.scheduler, "order", 1)
                         callback(step_idx, t, latents)
 
+                if XLA_AVAILABLE:
+                    xm.mark_step()
+
         # If we do sequential model offloading, let's offload unet and controlnet
         # manually for max memory savings
         if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:

diffusers/pipelines/controlnet/pipeline_controlnet_inpaint.py

@@ -32,6 +32,7 @@ from ...schedulers import KarrasDiffusionSchedulers
 from ...utils import (
     USE_PEFT_BACKEND,
     deprecate,
+    is_torch_xla_available,
     logging,
     replace_example_docstring,
     scale_lora_layers,
@@ -43,6 +44,13 @@ from ..stable_diffusion import StableDiffusionPipelineOutput
 from ..stable_diffusion.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
 
 
@@ -83,7 +91,7 @@ EXAMPLE_DOC_STRING = """
         ...     "lllyasviel/control_v11p_sd15_inpaint", torch_dtype=torch.float16
         ... )
         >>> pipe = StableDiffusionControlNetInpaintPipeline.from_pretrained(
-        ...     "runwayml/stable-diffusion-v1-5", controlnet=controlnet, torch_dtype=torch.float16
+        ...     "stable-diffusion-v1-5/stable-diffusion-v1-5", controlnet=controlnet, torch_dtype=torch.float16
         ... )
 
         >>> pipe.scheduler = DDIMScheduler.from_config(pipe.scheduler.config)
@@ -141,11 +149,11 @@ class StableDiffusionControlNetInpaintPipeline(
     <Tip>
 
     This pipeline can be used with checkpoints that have been specifically fine-tuned for inpainting
-    ([runwayml/stable-diffusion-inpainting](https://huggingface.co/runwayml/stable-diffusion-inpainting)) as well as
-    default text-to-image Stable Diffusion checkpoints
-    ([runwayml/stable-diffusion-v1-5](https://huggingface.co/runwayml/stable-diffusion-v1-5)). Default text-to-image
-    Stable Diffusion checkpoints might be preferable for ControlNets that have been fine-tuned on those, such as
-    [lllyasviel/control_v11p_sd15_inpaint](https://huggingface.co/lllyasviel/control_v11p_sd15_inpaint).
+    ([stable-diffusion-v1-5/stable-diffusion-inpainting](https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-inpainting))
+    as well as default text-to-image Stable Diffusion checkpoints
+    ([stable-diffusion-v1-5/stable-diffusion-v1-5](https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5)).
+    Default text-to-image Stable Diffusion checkpoints might be preferable for ControlNets that have been fine-tuned on
+    those, such as [lllyasviel/control_v11p_sd15_inpaint](https://huggingface.co/lllyasviel/control_v11p_sd15_inpaint).
 
     </Tip>
 
@@ -167,8 +175,8 @@ class StableDiffusionControlNetInpaintPipeline(
             [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`].
         safety_checker ([`StableDiffusionSafetyChecker`]):
             Classification module that estimates whether generated images could be considered offensive or harmful.
-            Please refer to the [model card](https://huggingface.co/runwayml/stable-diffusion-v1-5) for more details
-            about a model's potential harms.
+            Please refer to the [model card](https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5) for
+            more details about a model's potential harms.
         feature_extractor ([`~transformers.CLIPImageProcessor`]):
             A `CLIPImageProcessor` to extract features from generated images; used as inputs to the `safety_checker`.
     """
@@ -176,7 +184,14 @@ class StableDiffusionControlNetInpaintPipeline(
     model_cpu_offload_seq = "text_encoder->image_encoder->unet->vae"
     _optional_components = ["safety_checker", "feature_extractor", "image_encoder"]
     _exclude_from_cpu_offload = ["safety_checker"]
-    _callback_tensor_inputs = ["latents", "prompt_embeds", "negative_prompt_embeds"]
+    _callback_tensor_inputs = [
+        "latents",
+        "prompt_embeds",
+        "negative_prompt_embeds",
+        "control_image",
+        "mask",
+        "masked_image_latents",
+    ]
 
     def __init__(
         self,
@@ -223,7 +238,7 @@ class StableDiffusionControlNetInpaintPipeline(
             feature_extractor=feature_extractor,
             image_encoder=image_encoder,
         )
-        self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)
+        self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) if getattr(self, "vae", None) else 8
         self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
         self.mask_processor = VaeImageProcessor(
             vae_scale_factor=self.vae_scale_factor, do_normalize=False, do_binarize=True, do_convert_grayscale=True
@@ -642,7 +657,7 @@ class StableDiffusionControlNetInpaintPipeline(
         if padding_mask_crop is not None:
             if not isinstance(image, PIL.Image.Image):
                 raise ValueError(
-                    f"The image should be a PIL image when inpainting mask crop, but is of type" f" {type(image)}."
+                    f"The image should be a PIL image when inpainting mask crop, but is of type {type(image)}."
                 )
             if not isinstance(mask_image, PIL.Image.Image):
                 raise ValueError(
@@ -650,7 +665,7 @@ class StableDiffusionControlNetInpaintPipeline(
                     f" {type(mask_image)}."
                 )
             if output_type != "pil":
-                raise ValueError(f"The output type should be PIL when inpainting mask crop, but is" f" {output_type}.")
+                raise ValueError(f"The output type should be PIL when inpainting mask crop, but is {output_type}.")
 
         # `prompt` needs more sophisticated handling when there are multiple
         # conditionings.
@@ -1468,6 +1483,7 @@ class StableDiffusionControlNetInpaintPipeline(
                     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)
+                    control_image = callback_outputs.pop("control_image", control_image)
 
                 # call the callback, if provided
                 if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
@@ -1476,6 +1492,9 @@ class StableDiffusionControlNetInpaintPipeline(
                         step_idx = i // getattr(self.scheduler, "order", 1)
                         callback(step_idx, t, latents)
 
+                if XLA_AVAILABLE:
+                    xm.mark_step()
+
         # If we do sequential model offloading, let's offload unet and controlnet
         # manually for max memory savings
         if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:

diffusers/pipelines/controlnet/pipeline_controlnet_inpaint_sd_xl.py

@@ -60,6 +60,16 @@ if is_invisible_watermark_available():
     from diffusers.pipelines.stable_diffusion_xl.watermark import StableDiffusionXLWatermarker
 
 
+from ...utils import is_torch_xla_available
+
+
+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
 
 
@@ -227,6 +237,7 @@ class StableDiffusionXLControlNetInpaintPipeline(
         "add_neg_time_ids",
         "mask",
         "masked_image_latents",
+        "control_image",
     ]
 
     def __init__(
@@ -264,7 +275,7 @@ class StableDiffusionXLControlNetInpaintPipeline(
         )
         self.register_to_config(force_zeros_for_empty_prompt=force_zeros_for_empty_prompt)
         self.register_to_config(requires_aesthetics_score=requires_aesthetics_score)
-        self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)
+        self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) if getattr(self, "vae", None) else 8
         self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
         self.mask_processor = VaeImageProcessor(
             vae_scale_factor=self.vae_scale_factor, do_normalize=False, do_binarize=True, do_convert_grayscale=True
@@ -406,7 +417,9 @@ class StableDiffusionXLControlNetInpaintPipeline(
                 prompt_embeds = text_encoder(text_input_ids.to(device), output_hidden_states=True)
 
                 # We are only ALWAYS interested in the pooled output of the final text encoder
-                pooled_prompt_embeds = prompt_embeds[0]
+                if pooled_prompt_embeds is None and prompt_embeds[0].ndim == 2:
+                    pooled_prompt_embeds = prompt_embeds[0]
+
                 if clip_skip is None:
                     prompt_embeds = prompt_embeds.hidden_states[-2]
                 else:
@@ -465,8 +478,10 @@ class StableDiffusionXLControlNetInpaintPipeline(
                     uncond_input.input_ids.to(device),
                     output_hidden_states=True,
                 )
+
                 # We are only ALWAYS interested in the pooled output of the final text encoder
-                negative_pooled_prompt_embeds = negative_prompt_embeds[0]
+                if negative_pooled_prompt_embeds is None and negative_prompt_embeds[0].ndim == 2:
+                    negative_pooled_prompt_embeds = negative_prompt_embeds[0]
                 negative_prompt_embeds = negative_prompt_embeds.hidden_states[-2]
 
                 negative_prompt_embeds_list.append(negative_prompt_embeds)
@@ -729,7 +744,7 @@ class StableDiffusionXLControlNetInpaintPipeline(
         if padding_mask_crop is not None:
             if not isinstance(image, PIL.Image.Image):
                 raise ValueError(
-                    f"The image should be a PIL image when inpainting mask crop, but is of type" f" {type(image)}."
+                    f"The image should be a PIL image when inpainting mask crop, but is of type {type(image)}."
                 )
             if not isinstance(mask_image, PIL.Image.Image):
                 raise ValueError(
@@ -737,7 +752,7 @@ class StableDiffusionXLControlNetInpaintPipeline(
                     f" {type(mask_image)}."
                 )
             if output_type != "pil":
-                raise ValueError(f"The output type should be PIL when inpainting mask crop, but is" f" {output_type}.")
+                raise ValueError(f"The output type should be PIL when inpainting mask crop, but is {output_type}.")
 
         if prompt_embeds is not None and pooled_prompt_embeds is None:
             raise ValueError(
@@ -1622,7 +1637,7 @@ class StableDiffusionXLControlNetInpaintPipeline(
 
         # 8. Check that sizes of mask, masked image and latents match
         if num_channels_unet == 9:
-            # default case for runwayml/stable-diffusion-inpainting
+            # default case for stable-diffusion-v1-5/stable-diffusion-inpainting
             num_channels_mask = mask.shape[1]
             num_channels_masked_image = masked_image_latents.shape[1]
             if num_channels_latents + num_channels_mask + num_channels_masked_image != self.unet.config.in_channels:
@@ -1630,7 +1645,7 @@ class StableDiffusionXLControlNetInpaintPipeline(
                     f"Incorrect configuration settings! The config of `pipeline.unet`: {self.unet.config} expects"
                     f" {self.unet.config.in_channels} but received `num_channels_latents`: {num_channels_latents} +"
                     f" `num_channels_mask`: {num_channels_mask} + `num_channels_masked_image`: {num_channels_masked_image}"
-                    f" = {num_channels_latents+num_channels_masked_image+num_channels_mask}. Please verify the config of"
+                    f" = {num_channels_latents + num_channels_masked_image + num_channels_mask}. Please verify the config of"
                     " `pipeline.unet` or your `mask_image` or `image` input."
                 )
         elif num_channels_unet != 4:
@@ -1821,6 +1836,7 @@ class StableDiffusionXLControlNetInpaintPipeline(
                     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)
+                    control_image = callback_outputs.pop("control_image", control_image)
 
                 # call the callback, if provided
                 if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
@@ -1829,6 +1845,9 @@ class StableDiffusionXLControlNetInpaintPipeline(
                         step_idx = i // getattr(self.scheduler, "order", 1)
                         callback(step_idx, t, latents)
 
+                if XLA_AVAILABLE:
+                    xm.mark_step()
+
         # make sure the VAE is in float32 mode, as it overflows in float16
         if self.vae.dtype == torch.float16 and self.vae.config.force_upcast:
             self.upcast_vae()

diffusers/pipelines/controlnet/pipeline_controlnet_sd_xl.py

@@ -62,6 +62,16 @@ if is_invisible_watermark_available():
     from ..stable_diffusion_xl.watermark import StableDiffusionXLWatermarker
 
 
+from ...utils import is_torch_xla_available
+
+
+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
 
 
@@ -275,7 +285,7 @@ class StableDiffusionXLControlNetPipeline(
             feature_extractor=feature_extractor,
             image_encoder=image_encoder,
         )
-        self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)
+        self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) if getattr(self, "vae", None) else 8
         self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True)
         self.control_image_processor = VaeImageProcessor(
             vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True, do_normalize=False
@@ -415,7 +425,9 @@ class StableDiffusionXLControlNetPipeline(
                 prompt_embeds = text_encoder(text_input_ids.to(device), output_hidden_states=True)
 
                 # We are only ALWAYS interested in the pooled output of the final text encoder
-                pooled_prompt_embeds = prompt_embeds[0]
+                if pooled_prompt_embeds is None and prompt_embeds[0].ndim == 2:
+                    pooled_prompt_embeds = prompt_embeds[0]
+
                 if clip_skip is None:
                     prompt_embeds = prompt_embeds.hidden_states[-2]
                 else:
@@ -474,8 +486,10 @@ class StableDiffusionXLControlNetPipeline(
                     uncond_input.input_ids.to(device),
                     output_hidden_states=True,
                 )
+
                 # We are only ALWAYS interested in the pooled output of the final text encoder
-                negative_pooled_prompt_embeds = negative_prompt_embeds[0]
+                if negative_pooled_prompt_embeds is None and negative_prompt_embeds[0].ndim == 2:
+                    negative_pooled_prompt_embeds = negative_prompt_embeds[0]
                 negative_prompt_embeds = negative_prompt_embeds.hidden_states[-2]
 
                 negative_prompt_embeds_list.append(negative_prompt_embeds)
@@ -1548,6 +1562,9 @@ class StableDiffusionXLControlNetPipeline(
                         step_idx = i // getattr(self.scheduler, "order", 1)
                         callback(step_idx, t, latents)
 
+                if XLA_AVAILABLE:
+                    xm.mark_step()
+
         if not output_type == "latent":
             # make sure the VAE is in float32 mode, as it overflows in float16
             needs_upcasting = self.vae.dtype == torch.float16 and self.vae.config.force_upcast

diffusers/pipelines/controlnet/pipeline_controlnet_sd_xl_img2img.py

@@ -62,6 +62,16 @@ if is_invisible_watermark_available():
     from ..stable_diffusion_xl.watermark import StableDiffusionXLWatermarker
 
 
+from ...utils import is_torch_xla_available
+
+
+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
 
 
@@ -232,6 +242,7 @@ class StableDiffusionXLControlNetImg2ImgPipeline(
         "add_time_ids",
         "negative_pooled_prompt_embeds",
         "add_neg_time_ids",
+        "control_image",
     ]
 
     def __init__(
@@ -267,7 +278,7 @@ class StableDiffusionXLControlNetImg2ImgPipeline(
             feature_extractor=feature_extractor,
             image_encoder=image_encoder,
         )
-        self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)
+        self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) if getattr(self, "vae", None) else 8
         self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True)
         self.control_image_processor = VaeImageProcessor(
             vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True, do_normalize=False
@@ -408,7 +419,9 @@ class StableDiffusionXLControlNetImg2ImgPipeline(
                 prompt_embeds = text_encoder(text_input_ids.to(device), output_hidden_states=True)
 
                 # We are only ALWAYS interested in the pooled output of the final text encoder
-                pooled_prompt_embeds = prompt_embeds[0]
+                if pooled_prompt_embeds is None and prompt_embeds[0].ndim == 2:
+                    pooled_prompt_embeds = prompt_embeds[0]
+
                 if clip_skip is None:
                     prompt_embeds = prompt_embeds.hidden_states[-2]
                 else:
@@ -467,8 +480,10 @@ class StableDiffusionXLControlNetImg2ImgPipeline(
                     uncond_input.input_ids.to(device),
                     output_hidden_states=True,
                 )
+
                 # We are only ALWAYS interested in the pooled output of the final text encoder
-                negative_pooled_prompt_embeds = negative_prompt_embeds[0]
+                if negative_pooled_prompt_embeds is None and negative_prompt_embeds[0].ndim == 2:
+                    negative_pooled_prompt_embeds = negative_prompt_embeds[0]
                 negative_prompt_embeds = negative_prompt_embeds.hidden_states[-2]
 
                 negative_prompt_embeds_list.append(negative_prompt_embeds)
@@ -1600,6 +1615,7 @@ class StableDiffusionXLControlNetImg2ImgPipeline(
                     )
                     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)
+                    control_image = callback_outputs.pop("control_image", control_image)
 
                 # call the callback, if provided
                 if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
@@ -1608,6 +1624,9 @@ class StableDiffusionXLControlNetImg2ImgPipeline(
                         step_idx = i // getattr(self.scheduler, "order", 1)
                         callback(step_idx, t, latents)
 
+                if XLA_AVAILABLE:
+                    xm.mark_step()
+
         # If we do sequential model offloading, let's offload unet and controlnet
         # manually for max memory savings
         if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None: