diffusers 0.23.0__py3-none-any.whl → 0.24.0__py3-none-any.whl

diffusers/pipelines/t2i_adapter/pipeline_stable_diffusion_adapter.py

@@ -118,6 +118,51 @@ def _preprocess_adapter_image(image, height, width):
     return image
 
 
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.retrieve_timesteps
+def retrieve_timesteps(
+    scheduler,
+    num_inference_steps: Optional[int] = None,
+    device: Optional[Union[str, torch.device]] = None,
+    timesteps: Optional[List[int]] = None,
+    **kwargs,
+):
+    """
+    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`.
+
+    Args:
+        scheduler (`SchedulerMixin`):
+            The scheduler to get timesteps from.
+        num_inference_steps (`int`):
+            The number of diffusion steps used when generating samples with a pre-trained model. If used,
+            `timesteps` must be `None`.
+        device (`str` or `torch.device`, *optional*):
+            The device to which the timesteps should be moved to. If `None`, the timesteps are not moved.
+        timesteps (`List[int]`, *optional*):
+                Custom timesteps used to support arbitrary spacing between timesteps. If `None`, then the default
+                timestep spacing strategy of the scheduler is used. If `timesteps` is passed, `num_inference_steps`
+                must be `None`.
+
+    Returns:
+        `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the
+        second element is the number of inference steps.
+    """
+    if timesteps is not None:
+        accepts_timesteps = "timesteps" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+        if not accepts_timesteps:
+            raise ValueError(
+                f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+                f" timestep schedules. Please check whether you are using the correct scheduler."
+            )
+        scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+        num_inference_steps = len(timesteps)
+    else:
+        scheduler.set_timesteps(num_inference_steps, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+    return timesteps, num_inference_steps
+
+
 class StableDiffusionAdapterPipeline(DiffusionPipeline):
     r"""
     Pipeline for text-to-image generation using Stable Diffusion augmented with T2I-Adapter
@@ -152,6 +197,7 @@ class StableDiffusionAdapterPipeline(DiffusionPipeline):
         feature_extractor ([`CLIPFeatureExtractor`]):
             Model that extracts features from generated images to be used as inputs for the `safety_checker`.
     """
+
     model_cpu_offload_seq = "text_encoder->adapter->unet->vae"
     _optional_components = ["safety_checker", "feature_extractor"]
 
@@ -610,6 +656,46 @@ class StableDiffusionAdapterPipeline(DiffusionPipeline):
         """Disables the FreeU mechanism if enabled."""
         self.unet.disable_freeu()
 
+    # Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
+    def get_guidance_scale_embedding(self, w, embedding_dim=512, dtype=torch.float32):
+        """
+        See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298
+
+        Args:
+            timesteps (`torch.Tensor`):
+                generate embedding vectors at these timesteps
+            embedding_dim (`int`, *optional*, defaults to 512):
+                dimension of the embeddings to generate
+            dtype:
+                data type of the generated embeddings
+
+        Returns:
+            `torch.FloatTensor`: Embedding vectors with shape `(len(timesteps), embedding_dim)`
+        """
+        assert len(w.shape) == 1
+        w = w * 1000.0
+
+        half_dim = embedding_dim // 2
+        emb = torch.log(torch.tensor(10000.0)) / (half_dim - 1)
+        emb = torch.exp(torch.arange(half_dim, dtype=dtype) * -emb)
+        emb = w.to(dtype)[:, None] * emb[None, :]
+        emb = torch.cat([torch.sin(emb), torch.cos(emb)], dim=1)
+        if embedding_dim % 2 == 1:  # zero pad
+            emb = torch.nn.functional.pad(emb, (0, 1))
+        assert emb.shape == (w.shape[0], embedding_dim)
+        return emb
+
+    @property
+    def guidance_scale(self):
+        return self._guidance_scale
+
+    # 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`
+    # corresponds to doing no classifier free guidance.
+    @property
+    def do_classifier_free_guidance(self):
+        return self._guidance_scale > 1 and self.unet.config.time_cond_proj_dim is None
+
     @torch.no_grad()
     @replace_example_docstring(EXAMPLE_DOC_STRING)
     def __call__(
@@ -619,6 +705,7 @@ class StableDiffusionAdapterPipeline(DiffusionPipeline):
         height: Optional[int] = None,
         width: Optional[int] = None,
         num_inference_steps: int = 50,
+        timesteps: List[int] = None,
         guidance_scale: float = 7.5,
         negative_prompt: Optional[Union[str, List[str]]] = None,
         num_images_per_prompt: Optional[int] = 1,
@@ -653,6 +740,10 @@ class StableDiffusionAdapterPipeline(DiffusionPipeline):
             num_inference_steps (`int`, *optional*, defaults to 50):
                 The number of denoising steps. More denoising steps usually lead to a higher quality image at the
                 expense of slower inference.
+            timesteps (`List[int]`, *optional*):
+                Custom timesteps to use for the denoising process with schedulers which support a `timesteps` argument
+                in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is
+                passed will be used. Must be in descending order.
             guidance_scale (`float`, *optional*, defaults to 7.5):
                 Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
                 `guidance_scale` is defined as `w` of equation 2. of [Imagen
@@ -723,6 +814,8 @@ class StableDiffusionAdapterPipeline(DiffusionPipeline):
             prompt, height, width, callback_steps, image, negative_prompt, prompt_embeds, negative_prompt_embeds
         )
 
+        self._guidance_scale = guidance_scale
+
         if isinstance(self.adapter, MultiAdapter):
             adapter_input = []
 
@@ -742,17 +835,12 @@ class StableDiffusionAdapterPipeline(DiffusionPipeline):
         else:
             batch_size = prompt_embeds.shape[0]
 
-        # 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`
-        # corresponds to doing no classifier free guidance.
-        do_classifier_free_guidance = guidance_scale > 1.0
-
         # 3. Encode input prompt
         prompt_embeds, negative_prompt_embeds = self.encode_prompt(
             prompt,
             device,
             num_images_per_prompt,
-            do_classifier_free_guidance,
+            self.do_classifier_free_guidance,
             negative_prompt,
             prompt_embeds=prompt_embeds,
             negative_prompt_embeds=negative_prompt_embeds,
@@ -761,12 +849,11 @@ class StableDiffusionAdapterPipeline(DiffusionPipeline):
         # 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
-        if do_classifier_free_guidance:
+        if self.do_classifier_free_guidance:
             prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds])
 
         # 4. Prepare timesteps
-        self.scheduler.set_timesteps(num_inference_steps, device=device)
-        timesteps = self.scheduler.timesteps
+        timesteps, num_inference_steps = retrieve_timesteps(self.scheduler, num_inference_steps, device, timesteps)
 
         # 5. Prepare latent variables
         num_channels_latents = self.unet.config.in_channels
@@ -784,6 +871,14 @@ class StableDiffusionAdapterPipeline(DiffusionPipeline):
         # 6. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
         extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
 
+        # 6.5 Optionally get Guidance Scale Embedding
+        timestep_cond = None
+        if self.unet.config.time_cond_proj_dim is not None:
+            guidance_scale_tensor = torch.tensor(self.guidance_scale - 1).repeat(batch_size * num_images_per_prompt)
+            timestep_cond = self.get_guidance_scale_embedding(
+                guidance_scale_tensor, embedding_dim=self.unet.config.time_cond_proj_dim
+            ).to(device=device, dtype=latents.dtype)
+
         # 7. Denoising loop
         if isinstance(self.adapter, MultiAdapter):
             adapter_state = self.adapter(adapter_input, adapter_conditioning_scale)
@@ -796,7 +891,7 @@ class StableDiffusionAdapterPipeline(DiffusionPipeline):
         if num_images_per_prompt > 1:
             for k, v in enumerate(adapter_state):
                 adapter_state[k] = v.repeat(num_images_per_prompt, 1, 1, 1)
-        if do_classifier_free_guidance:
+        if self.do_classifier_free_guidance:
             for k, v in enumerate(adapter_state):
                 adapter_state[k] = torch.cat([v] * 2, dim=0)
 
@@ -804,7 +899,7 @@ class StableDiffusionAdapterPipeline(DiffusionPipeline):
         with self.progress_bar(total=num_inference_steps) as progress_bar:
             for i, t in enumerate(timesteps):
                 # expand the latents if we are doing classifier free guidance
-                latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
+                latent_model_input = torch.cat([latents] * 2) if self.do_classifier_free_guidance else latents
                 latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
 
                 # predict the noise residual
@@ -812,13 +907,14 @@ class StableDiffusionAdapterPipeline(DiffusionPipeline):
                     latent_model_input,
                     t,
                     encoder_hidden_states=prompt_embeds,
+                    timestep_cond=timestep_cond,
                     cross_attention_kwargs=cross_attention_kwargs,
                     down_intrablock_additional_residuals=[state.clone() for state in adapter_state],
                     return_dict=False,
                 )[0]
 
                 # perform guidance
-                if do_classifier_free_guidance:
+                if self.do_classifier_free_guidance:
                     noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
                     noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
 

diffusers/pipelines/t2i_adapter/pipeline_stable_diffusion_xl_adapter.py

@@ -123,6 +123,51 @@ def rescale_noise_cfg(noise_cfg, noise_pred_text, guidance_rescale=0.0):
     return noise_cfg
 
 
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.retrieve_timesteps
+def retrieve_timesteps(
+    scheduler,
+    num_inference_steps: Optional[int] = None,
+    device: Optional[Union[str, torch.device]] = None,
+    timesteps: Optional[List[int]] = None,
+    **kwargs,
+):
+    """
+    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`.
+
+    Args:
+        scheduler (`SchedulerMixin`):
+            The scheduler to get timesteps from.
+        num_inference_steps (`int`):
+            The number of diffusion steps used when generating samples with a pre-trained model. If used,
+            `timesteps` must be `None`.
+        device (`str` or `torch.device`, *optional*):
+            The device to which the timesteps should be moved to. If `None`, the timesteps are not moved.
+        timesteps (`List[int]`, *optional*):
+                Custom timesteps used to support arbitrary spacing between timesteps. If `None`, then the default
+                timestep spacing strategy of the scheduler is used. If `timesteps` is passed, `num_inference_steps`
+                must be `None`.
+
+    Returns:
+        `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the
+        second element is the number of inference steps.
+    """
+    if timesteps is not None:
+        accepts_timesteps = "timesteps" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+        if not accepts_timesteps:
+            raise ValueError(
+                f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+                f" timestep schedules. Please check whether you are using the correct scheduler."
+            )
+        scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+        num_inference_steps = len(timesteps)
+    else:
+        scheduler.set_timesteps(num_inference_steps, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+    return timesteps, num_inference_steps
+
+
 class StableDiffusionXLAdapterPipeline(
     DiffusionPipeline, FromSingleFileMixin, StableDiffusionXLLoraLoaderMixin, TextualInversionLoaderMixin
 ):
@@ -159,6 +204,7 @@ class StableDiffusionXLAdapterPipeline(
         feature_extractor ([`CLIPFeatureExtractor`]):
             Model that extracts features from generated images to be used as inputs for the `safety_checker`.
     """
+
     model_cpu_offload_seq = "text_encoder->text_encoder_2->unet->vae"
     _optional_components = ["tokenizer", "tokenizer_2", "text_encoder", "text_encoder_2"]
 
@@ -670,6 +716,46 @@ class StableDiffusionXLAdapterPipeline(
         """Disables the FreeU mechanism if enabled."""
         self.unet.disable_freeu()
 
+    # Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
+    def get_guidance_scale_embedding(self, w, embedding_dim=512, dtype=torch.float32):
+        """
+        See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298
+
+        Args:
+            timesteps (`torch.Tensor`):
+                generate embedding vectors at these timesteps
+            embedding_dim (`int`, *optional*, defaults to 512):
+                dimension of the embeddings to generate
+            dtype:
+                data type of the generated embeddings
+
+        Returns:
+            `torch.FloatTensor`: Embedding vectors with shape `(len(timesteps), embedding_dim)`
+        """
+        assert len(w.shape) == 1
+        w = w * 1000.0
+
+        half_dim = embedding_dim // 2
+        emb = torch.log(torch.tensor(10000.0)) / (half_dim - 1)
+        emb = torch.exp(torch.arange(half_dim, dtype=dtype) * -emb)
+        emb = w.to(dtype)[:, None] * emb[None, :]
+        emb = torch.cat([torch.sin(emb), torch.cos(emb)], dim=1)
+        if embedding_dim % 2 == 1:  # zero pad
+            emb = torch.nn.functional.pad(emb, (0, 1))
+        assert emb.shape == (w.shape[0], embedding_dim)
+        return emb
+
+    @property
+    def guidance_scale(self):
+        return self._guidance_scale
+
+    # 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`
+    # corresponds to doing no classifier free guidance.
+    @property
+    def do_classifier_free_guidance(self):
+        return self._guidance_scale > 1 and self.unet.config.time_cond_proj_dim is None
+
     @torch.no_grad()
     @replace_example_docstring(EXAMPLE_DOC_STRING)
     def __call__(
@@ -680,6 +766,7 @@ class StableDiffusionXLAdapterPipeline(
         height: Optional[int] = None,
         width: Optional[int] = None,
         num_inference_steps: int = 50,
+        timesteps: List[int] = None,
         denoising_end: Optional[float] = None,
         guidance_scale: float = 5.0,
         negative_prompt: Optional[Union[str, List[str]]] = None,
@@ -733,6 +820,10 @@ class StableDiffusionXLAdapterPipeline(
             num_inference_steps (`int`, *optional*, defaults to 50):
                 The number of denoising steps. More denoising steps usually lead to a higher quality image at the
                 expense of slower inference.
+            timesteps (`List[int]`, *optional*):
+                Custom timesteps to use for the denoising process with schedulers which support a `timesteps` argument
+                in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is
+                passed will be used. Must be in descending order.
             denoising_end (`float`, *optional*):
                 When specified, determines the fraction (between 0.0 and 1.0) of the total denoising process to be
                 completed before it is intentionally prematurely terminated. As a result, the returned sample will
@@ -882,6 +973,8 @@ class StableDiffusionXLAdapterPipeline(
             negative_pooled_prompt_embeds,
         )
 
+        self._guidance_scale = guidance_scale
+
         # 2. Define call parameters
         if prompt is not None and isinstance(prompt, str):
             batch_size = 1
@@ -892,11 +985,6 @@ class StableDiffusionXLAdapterPipeline(
 
         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`
-        # corresponds to doing no classifier free guidance.
-        do_classifier_free_guidance = guidance_scale > 1.0
-
         # 3. Encode input prompt
         (
             prompt_embeds,
@@ -908,7 +996,7 @@ class StableDiffusionXLAdapterPipeline(
             prompt_2=prompt_2,
             device=device,
             num_images_per_prompt=num_images_per_prompt,
-            do_classifier_free_guidance=do_classifier_free_guidance,
+            do_classifier_free_guidance=self.do_classifier_free_guidance,
             negative_prompt=negative_prompt,
             negative_prompt_2=negative_prompt_2,
             prompt_embeds=prompt_embeds,
@@ -919,9 +1007,7 @@ class StableDiffusionXLAdapterPipeline(
         )
 
         # 4. Prepare timesteps
-        self.scheduler.set_timesteps(num_inference_steps, device=device)
-
-        timesteps = self.scheduler.timesteps
+        timesteps, num_inference_steps = retrieve_timesteps(self.scheduler, num_inference_steps, device, timesteps)
 
         # 5. Prepare latent variables
         num_channels_latents = self.unet.config.in_channels
@@ -939,6 +1025,14 @@ class StableDiffusionXLAdapterPipeline(
         # 6. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
         extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
 
+        # 6.5 Optionally get Guidance Scale Embedding
+        timestep_cond = None
+        if self.unet.config.time_cond_proj_dim is not None:
+            guidance_scale_tensor = torch.tensor(self.guidance_scale - 1).repeat(batch_size * num_images_per_prompt)
+            timestep_cond = self.get_guidance_scale_embedding(
+                guidance_scale_tensor, embedding_dim=self.unet.config.time_cond_proj_dim
+            ).to(device=device, dtype=latents.dtype)
+
         # 7. Prepare added time ids & embeddings & adapter features
         if isinstance(self.adapter, MultiAdapter):
             adapter_state = self.adapter(adapter_input, adapter_conditioning_scale)
@@ -951,7 +1045,7 @@ class StableDiffusionXLAdapterPipeline(
         if num_images_per_prompt > 1:
             for k, v in enumerate(adapter_state):
                 adapter_state[k] = v.repeat(num_images_per_prompt, 1, 1, 1)
-        if do_classifier_free_guidance:
+        if self.do_classifier_free_guidance:
             for k, v in enumerate(adapter_state):
                 adapter_state[k] = torch.cat([v] * 2, dim=0)
 
@@ -979,7 +1073,7 @@ class StableDiffusionXLAdapterPipeline(
         else:
             negative_add_time_ids = add_time_ids
 
-        if do_classifier_free_guidance:
+        if self.do_classifier_free_guidance:
             prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0)
             add_text_embeds = torch.cat([negative_pooled_prompt_embeds, add_text_embeds], dim=0)
             add_time_ids = torch.cat([negative_add_time_ids, add_time_ids], dim=0)
@@ -1005,7 +1099,7 @@ class StableDiffusionXLAdapterPipeline(
         with self.progress_bar(total=num_inference_steps) as progress_bar:
             for i, t in enumerate(timesteps):
                 # expand the latents if we are doing classifier free guidance
-                latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
+                latent_model_input = torch.cat([latents] * 2) if self.do_classifier_free_guidance else latents
 
                 latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
 
@@ -1021,6 +1115,7 @@ class StableDiffusionXLAdapterPipeline(
                     latent_model_input,
                     t,
                     encoder_hidden_states=prompt_embeds,
+                    timestep_cond=timestep_cond,
                     cross_attention_kwargs=cross_attention_kwargs,
                     added_cond_kwargs=added_cond_kwargs,
                     return_dict=False,
@@ -1028,11 +1123,11 @@ class StableDiffusionXLAdapterPipeline(
                 )[0]
 
                 # perform guidance
-                if do_classifier_free_guidance:
+                if self.do_classifier_free_guidance:
                     noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
                     noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
 
-                if do_classifier_free_guidance and guidance_rescale > 0.0:
+                if self.do_classifier_free_guidance and guidance_rescale > 0.0:
                     # Based on 3.4. in https://arxiv.org/pdf/2305.08891.pdf
                     noise_pred = rescale_noise_cfg(noise_pred, noise_pred_text, guidance_rescale=guidance_rescale)
 

diffusers/pipelines/text_to_video_synthesis/__init__.py

@@ -25,6 +25,7 @@ else:
     _import_structure["pipeline_text_to_video_synth"] = ["TextToVideoSDPipeline"]
     _import_structure["pipeline_text_to_video_synth_img2img"] = ["VideoToVideoSDPipeline"]
     _import_structure["pipeline_text_to_video_zero"] = ["TextToVideoZeroPipeline"]
+    _import_structure["pipeline_text_to_video_zero_sdxl"] = ["TextToVideoZeroSDXLPipeline"]
 
 
 if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
@@ -38,6 +39,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
         from .pipeline_text_to_video_synth import TextToVideoSDPipeline
         from .pipeline_text_to_video_synth_img2img import VideoToVideoSDPipeline
         from .pipeline_text_to_video_zero import TextToVideoZeroPipeline
+        from .pipeline_text_to_video_zero_sdxl import TextToVideoZeroSDXLPipeline
 
 else:
     import sys

diffusers/pipelines/text_to_video_synthesis/pipeline_text_to_video_synth.py

@@ -96,6 +96,7 @@ class TextToVideoSDPipeline(DiffusionPipeline, TextualInversionLoaderMixin, Lora
             A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of
             [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`].
     """
+
     model_cpu_offload_seq = "text_encoder->unet->vae"
 
     def __init__(

diffusers/pipelines/text_to_video_synthesis/pipeline_text_to_video_synth_img2img.py

@@ -79,6 +79,20 @@ EXAMPLE_DOC_STRING = """
 """
 
 
+# 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")
+
+
 def tensor2vid(video: torch.Tensor, mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5]) -> List[np.ndarray]:
     # This code is copied from https://github.com/modelscope/modelscope/blob/1509fdb973e5871f37148a4b5e5964cafd43e64d/modelscope/pipelines/multi_modal/text_to_video_synthesis_pipeline.py#L78
     # reshape to ncfhw
@@ -158,6 +172,7 @@ class VideoToVideoSDPipeline(DiffusionPipeline, TextualInversionLoaderMixin, Lor
             A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of
             [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`].
     """
+
     model_cpu_offload_seq = "text_encoder->unet->vae"
 
     def __init__(
@@ -546,14 +561,14 @@ class VideoToVideoSDPipeline(DiffusionPipeline, TextualInversionLoaderMixin, Lor
                     f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
                     f" size of {batch_size}. Make sure the batch size matches the length of the generators."
                 )
-
             elif isinstance(generator, list):
                 init_latents = [
-                    self.vae.encode(video[i : i + 1]).latent_dist.sample(generator[i]) for i in range(batch_size)
+                    retrieve_latents(self.vae.encode(video[i : i + 1]), generator=generator[i])
+                    for i in range(batch_size)
                 ]
                 init_latents = torch.cat(init_latents, dim=0)
             else:
-                init_latents = self.vae.encode(video).latent_dist.sample(generator)
+                init_latents = retrieve_latents(self.vae.encode(video), generator=generator)
 
             init_latents = self.vae.config.scaling_factor * init_latents
 

diffusers/pipelines/text_to_video_synthesis/pipeline_text_to_video_zero.py

@@ -13,6 +13,7 @@ from diffusers.models import AutoencoderKL, UNet2DConditionModel
 from diffusers.pipelines.stable_diffusion import StableDiffusionPipeline, StableDiffusionSafetyChecker
 from diffusers.schedulers import KarrasDiffusionSchedulers
 from diffusers.utils import BaseOutput
+from diffusers.utils.torch_utils import randn_tensor
 
 
 def rearrange_0(tensor, f):
@@ -135,7 +136,7 @@ class CrossFrameAttnProcessor2_0:
 
         # Cross Frame Attention
         if not is_cross_attention:
-            video_length = key.size()[0] // self.batch_size
+            video_length = max(1, key.size()[0] // self.batch_size)
             first_frame_index = [0] * video_length
 
             # rearrange keys to have batch and frames in the 1st and 2nd dims respectively
@@ -183,6 +184,7 @@ class TextToVideoPipelineOutput(BaseOutput):
             List indicating whether the corresponding generated image contains "not-safe-for-work" (nsfw) content or
             `None` if safety checking could not be performed.
     """
+
     images: Union[List[PIL.Image.Image], np.ndarray]
     nsfw_content_detected: Optional[List[bool]]
 
@@ -338,7 +340,7 @@ class TextToVideoZeroPipeline(StableDiffusionPipeline):
             x_t1:
                 Forward process applied to x_t0 from time t0 to t1.
         """
-        eps = torch.randn(x_t0.size(), generator=generator, dtype=x_t0.dtype, device=x_t0.device)
+        eps = randn_tensor(x_t0.size(), generator=generator, dtype=x_t0.dtype, device=x_t0.device)
         alpha_vec = torch.prod(self.scheduler.alphas[t0:t1])
         x_t1 = torch.sqrt(alpha_vec) * x_t0 + torch.sqrt(1 - alpha_vec) * eps
         return x_t1