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

diffusers/pipelines/pag/pipeline_pag_sana.py

@@ -16,10 +16,11 @@ import html
 import inspect
 import re
 import urllib.parse as ul
+import warnings
 from typing import Callable, Dict, List, Optional, Tuple, Union
 
 import torch
-from transformers import AutoModelForCausalLM, AutoTokenizer
+from transformers import Gemma2PreTrainedModel, GemmaTokenizer, GemmaTokenizerFast
 
 from ...callbacks import MultiPipelineCallbacks, PipelineCallback
 from ...image_processor import PixArtImageProcessor
@@ -30,6 +31,7 @@ from ...utils import (
     BACKENDS_MAPPING,
     is_bs4_available,
     is_ftfy_available,
+    is_torch_xla_available,
     logging,
     replace_example_docstring,
 )
@@ -40,11 +42,20 @@ from ..pixart_alpha.pipeline_pixart_alpha import (
     ASPECT_RATIO_1024_BIN,
 )
 from ..pixart_alpha.pipeline_pixart_sigma import ASPECT_RATIO_2048_BIN
+from ..sana.pipeline_sana import ASPECT_RATIO_4096_BIN
 from .pag_utils import PAGMixin
 
 
+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
 
+
 if is_bs4_available():
     from bs4 import BeautifulSoup
 
@@ -149,8 +160,8 @@ class SanaPAGPipeline(DiffusionPipeline, PAGMixin):
 
     def __init__(
         self,
-        tokenizer: AutoTokenizer,
-        text_encoder: AutoModelForCausalLM,
+        tokenizer: Union[GemmaTokenizer, GemmaTokenizerFast],
+        text_encoder: Gemma2PreTrainedModel,
         vae: AutoencoderDC,
         transformer: SanaTransformer2DModel,
         scheduler: FlowMatchEulerDiscreteScheduler,
@@ -162,7 +173,11 @@ class SanaPAGPipeline(DiffusionPipeline, PAGMixin):
             tokenizer=tokenizer, text_encoder=text_encoder, vae=vae, transformer=transformer, scheduler=scheduler
         )
 
-        self.vae_scale_factor = 2 ** (len(self.vae.config.encoder_block_out_channels) - 1)
+        self.vae_scale_factor = (
+            2 ** (len(self.vae.config.encoder_block_out_channels) - 1)
+            if hasattr(self, "vae") and self.vae is not None
+            else 8
+        )
         self.image_processor = PixArtImageProcessor(vae_scale_factor=self.vae_scale_factor)
 
         self.set_pag_applied_layers(
@@ -170,6 +185,35 @@ class SanaPAGPipeline(DiffusionPipeline, PAGMixin):
             pag_attn_processors=(PAGCFGSanaLinearAttnProcessor2_0(), PAGIdentitySanaLinearAttnProcessor2_0()),
         )
 
+    def enable_vae_slicing(self):
+        r"""
+        Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to
+        compute decoding in several steps. This is useful to save some memory and allow larger batch sizes.
+        """
+        self.vae.enable_slicing()
+
+    def disable_vae_slicing(self):
+        r"""
+        Disable sliced VAE decoding. If `enable_vae_slicing` was previously enabled, this method will go back to
+        computing decoding in one step.
+        """
+        self.vae.disable_slicing()
+
+    def enable_vae_tiling(self):
+        r"""
+        Enable tiled VAE decoding. When this option is enabled, the VAE will split the input tensor into tiles to
+        compute decoding and encoding in several steps. This is useful for saving a large amount of memory and to allow
+        processing larger images.
+        """
+        self.vae.enable_tiling()
+
+    def disable_vae_tiling(self):
+        r"""
+        Disable tiled VAE decoding. If `enable_vae_tiling` was previously enabled, this method will go back to
+        computing decoding in one step.
+        """
+        self.vae.disable_tiling()
+
     def encode_prompt(
         self,
         prompt: Union[str, List[str]],
@@ -224,7 +268,8 @@ class SanaPAGPipeline(DiffusionPipeline, PAGMixin):
         else:
             batch_size = prompt_embeds.shape[0]
 
-        self.tokenizer.padding_side = "right"
+        if getattr(self, "tokenizer", None) is not None:
+            self.tokenizer.padding_side = "right"
 
         # See Section 3.1. of the paper.
         max_length = max_sequence_length
@@ -597,7 +642,7 @@ class SanaPAGPipeline(DiffusionPipeline, PAGMixin):
         negative_prompt_attention_mask: Optional[torch.Tensor] = None,
         output_type: Optional[str] = "pil",
         return_dict: bool = True,
-        clean_caption: bool = True,
+        clean_caption: bool = False,
         use_resolution_binning: bool = True,
         callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
         callback_on_step_end_tensor_inputs: List[str] = ["latents"],
@@ -713,7 +758,9 @@ class SanaPAGPipeline(DiffusionPipeline, PAGMixin):
             callback_on_step_end_tensor_inputs = callback_on_step_end.tensor_inputs
 
         if use_resolution_binning:
-            if self.transformer.config.sample_size == 64:
+            if self.transformer.config.sample_size == 128:
+                aspect_ratio_bin = ASPECT_RATIO_4096_BIN
+            elif self.transformer.config.sample_size == 64:
                 aspect_ratio_bin = ASPECT_RATIO_2048_BIN
             elif self.transformer.config.sample_size == 32:
                 aspect_ratio_bin = ASPECT_RATIO_1024_BIN
@@ -863,11 +910,21 @@ class SanaPAGPipeline(DiffusionPipeline, PAGMixin):
                 if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                     progress_bar.update()
 
+                if XLA_AVAILABLE:
+                    xm.mark_step()
+
         if output_type == "latent":
             image = latents
         else:
             latents = latents.to(self.vae.dtype)
-            image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False)[0]
+            try:
+                image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False)[0]
+            except torch.cuda.OutOfMemoryError as e:
+                warnings.warn(
+                    f"{e}. \n"
+                    f"Try to use VAE tiling for large images. For example: \n"
+                    f"pipe.vae.enable_tiling(tile_sample_min_width=512, tile_sample_min_height=512)"
+                )
             if use_resolution_binning:
                 image = self.image_processor.resize_and_crop_tensor(image, orig_width, orig_height)
 

diffusers/pipelines/pag/pipeline_pag_sd.py

@@ -27,6 +27,7 @@ from ...schedulers import KarrasDiffusionSchedulers
 from ...utils import (
     USE_PEFT_BACKEND,
     deprecate,
+    is_torch_xla_available,
     logging,
     replace_example_docstring,
     scale_lora_layers,
@@ -39,8 +40,16 @@ from ..stable_diffusion.safety_checker import StableDiffusionSafetyChecker
 from .pag_utils import PAGMixin
 
 
+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
@@ -207,7 +216,7 @@ class StableDiffusionPAGPipeline(
     ):
         super().__init__()
 
-        if hasattr(scheduler.config, "steps_offset") and scheduler.config.steps_offset != 1:
+        if scheduler is not None and getattr(scheduler.config, "steps_offset", 1) != 1:
             deprecation_message = (
                 f"The configuration file of this scheduler: {scheduler} is outdated. `steps_offset`"
                 f" should be set to 1 instead of {scheduler.config.steps_offset}. Please make sure "
@@ -221,7 +230,7 @@ class StableDiffusionPAGPipeline(
             new_config["steps_offset"] = 1
             scheduler._internal_dict = FrozenDict(new_config)
 
-        if hasattr(scheduler.config, "clip_sample") and scheduler.config.clip_sample is True:
+        if scheduler is not None and getattr(scheduler.config, "clip_sample", False) is True:
             deprecation_message = (
                 f"The configuration file of this scheduler: {scheduler} has not set the configuration `clip_sample`."
                 " `clip_sample` should be set to False in the configuration file. Please make sure to update the"
@@ -250,10 +259,14 @@ class StableDiffusionPAGPipeline(
                 " checker. If you do not want to use the safety checker, you can pass `'safety_checker=None'` instead."
             )
 
-        is_unet_version_less_0_9_0 = hasattr(unet.config, "_diffusers_version") and version.parse(
-            version.parse(unet.config._diffusers_version).base_version
-        ) < version.parse("0.9.0.dev0")
-        is_unet_sample_size_less_64 = hasattr(unet.config, "sample_size") and unet.config.sample_size < 64
+        is_unet_version_less_0_9_0 = (
+            unet is not None
+            and hasattr(unet.config, "_diffusers_version")
+            and version.parse(version.parse(unet.config._diffusers_version).base_version) < version.parse("0.9.0.dev0")
+        )
+        is_unet_sample_size_less_64 = (
+            unet is not None and hasattr(unet.config, "sample_size") and unet.config.sample_size < 64
+        )
         if is_unet_version_less_0_9_0 and is_unet_sample_size_less_64:
             deprecation_message = (
                 "The configuration file of the unet has set the default `sample_size` to smaller than"
@@ -281,7 +294,7 @@ class StableDiffusionPAGPipeline(
             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.register_to_config(requires_safety_checker=requires_safety_checker)
 
@@ -1034,6 +1047,9 @@ class StableDiffusionPAGPipeline(
                 if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                     progress_bar.update()
 
+                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

diffusers/pipelines/pag/pipeline_pag_sd_3.py

@@ -200,9 +200,7 @@ class StableDiffusion3PAGPipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSin
             transformer=transformer,
             scheduler=scheduler,
         )
-        self.vae_scale_factor = (
-            2 ** (len(self.vae.config.block_out_channels) - 1) if hasattr(self, "vae") and self.vae is not None else 8
-        )
+        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.tokenizer_max_length = (
             self.tokenizer.model_max_length if hasattr(self, "tokenizer") and self.tokenizer is not None else 77
@@ -377,9 +375,9 @@ class StableDiffusion3PAGPipeline(DiffusionPipeline, SD3LoraLoaderMixin, FromSin
             negative_prompt_2 (`str` or `List[str]`, *optional*):
                 The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
                 `text_encoder_2`. If not defined, `negative_prompt` is used in all the text-encoders.
-            negative_prompt_2 (`str` or `List[str]`, *optional*):
+            negative_prompt_3 (`str` or `List[str]`, *optional*):
                 The prompt or prompts not to guide the image generation to be sent to `tokenizer_3` and
-                `text_encoder_3`. If not defined, `negative_prompt` is used in both text-encoders
+                `text_encoder_3`. If not defined, `negative_prompt` is used in all the text-encoders.
             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.

diffusers/pipelines/pag/pipeline_pag_sd_3_img2img.py

@@ -216,9 +216,7 @@ class StableDiffusion3PAGImg2ImgPipeline(DiffusionPipeline, SD3LoraLoaderMixin,
             transformer=transformer,
             scheduler=scheduler,
         )
-        self.vae_scale_factor = (
-            2 ** (len(self.vae.config.block_out_channels) - 1) if hasattr(self, "vae") and self.vae is not None else 8
-        )
+        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.tokenizer_max_length = (
             self.tokenizer.model_max_length if hasattr(self, "tokenizer") and self.tokenizer is not None else 77
@@ -393,9 +391,9 @@ class StableDiffusion3PAGImg2ImgPipeline(DiffusionPipeline, SD3LoraLoaderMixin,
             negative_prompt_2 (`str` or `List[str]`, *optional*):
                 The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
                 `text_encoder_2`. If not defined, `negative_prompt` is used in all the text-encoders.
-            negative_prompt_2 (`str` or `List[str]`, *optional*):
+            negative_prompt_3 (`str` or `List[str]`, *optional*):
                 The prompt or prompts not to guide the image generation to be sent to `tokenizer_3` and
-                `text_encoder_3`. If not defined, `negative_prompt` is used in both text-encoders
+                `text_encoder_3`. If not defined, `negative_prompt` is used in all the text-encoders.
             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.

diffusers/pipelines/pag/pipeline_pag_sd_animatediff.py

@@ -26,6 +26,7 @@ from ...models.unets.unet_motion_model import MotionAdapter
 from ...schedulers import KarrasDiffusionSchedulers
 from ...utils import (
     USE_PEFT_BACKEND,
+    is_torch_xla_available,
     logging,
     replace_example_docstring,
     scale_lora_layers,
@@ -40,8 +41,16 @@ from ..pipeline_utils import DiffusionPipeline, StableDiffusionMixin
 from .pag_utils import PAGMixin
 
 
+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
@@ -147,7 +156,7 @@ class AnimateDiffPAGPipeline(
             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.video_processor = VideoProcessor(do_resize=False, vae_scale_factor=self.vae_scale_factor)
 
         self.set_pag_applied_layers(pag_applied_layers)
@@ -847,6 +856,9 @@ class AnimateDiffPAGPipeline(
                     if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                         progress_bar.update()
 
+                    if XLA_AVAILABLE:
+                        xm.mark_step()
+
         # 9. Post processing
         if output_type == "latent":
             video = latents

diffusers/pipelines/pag/pipeline_pag_sd_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,
@@ -42,8 +43,16 @@ from ..stable_diffusion.safety_checker import StableDiffusionSafetyChecker
 from .pag_utils import PAGMixin
 
 
+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
@@ -202,7 +211,7 @@ class StableDiffusionPAGImg2ImgPipeline(
     ):
         super().__init__()
 
-        if hasattr(scheduler.config, "steps_offset") and scheduler.config.steps_offset != 1:
+        if scheduler is not None and getattr(scheduler.config, "steps_offset", 1) != 1:
             deprecation_message = (
                 f"The configuration file of this scheduler: {scheduler} is outdated. `steps_offset`"
                 f" should be set to 1 instead of {scheduler.config.steps_offset}. Please make sure "
@@ -216,7 +225,7 @@ class StableDiffusionPAGImg2ImgPipeline(
             new_config["steps_offset"] = 1
             scheduler._internal_dict = FrozenDict(new_config)
 
-        if hasattr(scheduler.config, "clip_sample") and scheduler.config.clip_sample is True:
+        if scheduler is not None and getattr(scheduler.config, "clip_sample", False) is True:
             deprecation_message = (
                 f"The configuration file of this scheduler: {scheduler} has not set the configuration `clip_sample`."
                 " `clip_sample` should be set to False in the configuration file. Please make sure to update the"
@@ -245,10 +254,14 @@ class StableDiffusionPAGImg2ImgPipeline(
                 " checker. If you do not want to use the safety checker, you can pass `'safety_checker=None'` instead."
             )
 
-        is_unet_version_less_0_9_0 = hasattr(unet.config, "_diffusers_version") and version.parse(
-            version.parse(unet.config._diffusers_version).base_version
-        ) < version.parse("0.9.0.dev0")
-        is_unet_sample_size_less_64 = hasattr(unet.config, "sample_size") and unet.config.sample_size < 64
+        is_unet_version_less_0_9_0 = (
+            unet is not None
+            and hasattr(unet.config, "_diffusers_version")
+            and version.parse(version.parse(unet.config._diffusers_version).base_version) < version.parse("0.9.0.dev0")
+        )
+        is_unet_sample_size_less_64 = (
+            unet is not None and hasattr(unet.config, "sample_size") and unet.config.sample_size < 64
+        )
         if is_unet_version_less_0_9_0 and is_unet_sample_size_less_64:
             deprecation_message = (
                 "The configuration file of the unet has set the default `sample_size` to smaller than"
@@ -276,7 +289,7 @@ class StableDiffusionPAGImg2ImgPipeline(
             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.register_to_config(requires_safety_checker=requires_safety_checker)
 
@@ -1066,6 +1079,9 @@ class StableDiffusionPAGImg2ImgPipeline(
                 if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                     progress_bar.update()
 
+                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

diffusers/pipelines/pag/pipeline_pag_sd_inpaint.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,
@@ -40,8 +41,16 @@ from ..stable_diffusion.safety_checker import StableDiffusionSafetyChecker
 from .pag_utils import PAGMixin
 
 
+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
@@ -234,7 +243,7 @@ class StableDiffusionPAGInpaintPipeline(
     ):
         super().__init__()
 
-        if hasattr(scheduler.config, "steps_offset") and scheduler.config.steps_offset != 1:
+        if scheduler is not None and getattr(scheduler.config, "steps_offset", 1) != 1:
             deprecation_message = (
                 f"The configuration file of this scheduler: {scheduler} is outdated. `steps_offset`"
                 f" should be set to 1 instead of {scheduler.config.steps_offset}. Please make sure "
@@ -248,7 +257,7 @@ class StableDiffusionPAGInpaintPipeline(
             new_config["steps_offset"] = 1
             scheduler._internal_dict = FrozenDict(new_config)
 
-        if hasattr(scheduler.config, "clip_sample") and scheduler.config.clip_sample is True:
+        if scheduler is not None and getattr(scheduler.config, "clip_sample", False) is True:
             deprecation_message = (
                 f"The configuration file of this scheduler: {scheduler} has not set the configuration `clip_sample`."
                 " `clip_sample` should be set to False in the configuration file. Please make sure to update the"
@@ -277,10 +286,14 @@ class StableDiffusionPAGInpaintPipeline(
                 " checker. If you do not want to use the safety checker, you can pass `'safety_checker=None'` instead."
             )
 
-        is_unet_version_less_0_9_0 = hasattr(unet.config, "_diffusers_version") and version.parse(
-            version.parse(unet.config._diffusers_version).base_version
-        ) < version.parse("0.9.0.dev0")
-        is_unet_sample_size_less_64 = hasattr(unet.config, "sample_size") and unet.config.sample_size < 64
+        is_unet_version_less_0_9_0 = (
+            unet is not None
+            and hasattr(unet.config, "_diffusers_version")
+            and version.parse(version.parse(unet.config._diffusers_version).base_version) < version.parse("0.9.0.dev0")
+        )
+        is_unet_sample_size_less_64 = (
+            unet is not None and hasattr(unet.config, "sample_size") and unet.config.sample_size < 64
+        )
         if is_unet_version_less_0_9_0 and is_unet_sample_size_less_64:
             deprecation_message = (
                 "The configuration file of the unet has set the default `sample_size` to smaller than"
@@ -308,7 +321,7 @@ class StableDiffusionPAGInpaintPipeline(
             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
@@ -670,7 +683,7 @@ class StableDiffusionPAGInpaintPipeline(
         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(
@@ -678,7 +691,7 @@ class StableDiffusionPAGInpaintPipeline(
                     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 ip_adapter_image is not None and ip_adapter_image_embeds is not None:
             raise ValueError(
@@ -1178,7 +1191,7 @@ class StableDiffusionPAGInpaintPipeline(
                     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:
@@ -1318,6 +1331,9 @@ class StableDiffusionPAGInpaintPipeline(
                 if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
                     progress_bar.update()
 
+                if XLA_AVAILABLE:
+                    xm.mark_step()
+
         if not output_type == "latent":
             condition_kwargs = {}
             if isinstance(self.vae, AsymmetricAutoencoderKL):

diffusers/pipelines/pag/pipeline_pag_sd_xl.py

@@ -275,10 +275,14 @@ class StableDiffusionXLPAGPipeline(
             feature_extractor=feature_extractor,
         )
         self.register_to_config(force_zeros_for_empty_prompt=force_zeros_for_empty_prompt)
-        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.default_sample_size = self.unet.config.sample_size
+        self.default_sample_size = (
+            self.unet.config.sample_size
+            if hasattr(self, "unet") and self.unet is not None and hasattr(self.unet.config, "sample_size")
+            else 128
+        )
 
         add_watermarker = add_watermarker if add_watermarker is not None else is_invisible_watermark_available()
 
@@ -415,7 +419,9 @@ class StableDiffusionXLPAGPipeline(
                 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 +480,10 @@ class StableDiffusionXLPAGPipeline(
                     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)

diffusers/pipelines/pag/pipeline_pag_sd_xl_img2img.py

@@ -298,7 +298,7 @@ class StableDiffusionXLPAGImg2ImgPipeline(
         )
         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)
 
         add_watermarker = add_watermarker if add_watermarker is not None else is_invisible_watermark_available()
@@ -436,7 +436,9 @@ class StableDiffusionXLPAGImg2ImgPipeline(
                 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:
@@ -495,8 +497,10 @@ class StableDiffusionXLPAGImg2ImgPipeline(
                     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)

diffusers/pipelines/pag/pipeline_pag_sd_xl_inpaint.py

@@ -314,7 +314,7 @@ class StableDiffusionXLPAGInpaintPipeline(
         )
         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
@@ -526,7 +526,9 @@ class StableDiffusionXLPAGInpaintPipeline(
                 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:
@@ -585,8 +587,10 @@ class StableDiffusionXLPAGInpaintPipeline(
                     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)
@@ -733,7 +737,7 @@ class StableDiffusionXLPAGInpaintPipeline(
         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(
@@ -741,7 +745,7 @@ class StableDiffusionXLPAGInpaintPipeline(
                     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 ip_adapter_image is not None and ip_adapter_image_embeds is not None:
             raise ValueError(
@@ -1505,7 +1509,7 @@ class StableDiffusionXLPAGInpaintPipeline(
                     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: