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

diffusers/pipelines/audioldm2/modeling_audioldm2.py

@@ -1112,7 +1112,7 @@ class CrossAttnDownBlock2D(nn.Module):
         )
 
         for i in range(num_layers):
-            if self.training and self.gradient_checkpointing:
+            if torch.is_grad_enabled() and self.gradient_checkpointing:
 
                 def create_custom_forward(module, return_dict=None):
                     def custom_forward(*inputs):
@@ -1290,7 +1290,7 @@ class UNetMidBlock2DCrossAttn(nn.Module):
         )
 
         for i in range(len(self.resnets[1:])):
-            if self.training and self.gradient_checkpointing:
+            if torch.is_grad_enabled() and self.gradient_checkpointing:
 
                 def create_custom_forward(module, return_dict=None):
                     def custom_forward(*inputs):
@@ -1464,7 +1464,7 @@ class CrossAttnUpBlock2D(nn.Module):
             res_hidden_states_tuple = res_hidden_states_tuple[:-1]
             hidden_states = torch.cat([hidden_states, res_hidden_states], dim=1)
 
-            if self.training and self.gradient_checkpointing:
+            if torch.is_grad_enabled() and self.gradient_checkpointing:
 
                 def create_custom_forward(module, return_dict=None):
                     def custom_forward(*inputs):

diffusers/pipelines/aura_flow/pipeline_aura_flow.py

@@ -53,7 +53,7 @@ def retrieve_timesteps(
     sigmas: Optional[List[float]] = None,
     **kwargs,
 ):
-    """
+    r"""
     Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
     custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
 
@@ -387,7 +387,6 @@ class AuraFlowPipeline(DiffusionPipeline):
         prompt: Union[str, List[str]] = None,
         negative_prompt: Union[str, List[str]] = None,
         num_inference_steps: int = 50,
-        timesteps: List[int] = None,
         sigmas: List[float] = None,
         guidance_scale: float = 3.5,
         num_images_per_prompt: Optional[int] = 1,
@@ -424,10 +423,6 @@ class AuraFlowPipeline(DiffusionPipeline):
             sigmas (`List[float]`, *optional*):
                 Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed,
                 `num_inference_steps` and `timesteps` must be `None`.
-            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 5.0):
                 Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598).
                 `guidance_scale` is defined as `w` of equation 2. of [Imagen
@@ -522,9 +517,7 @@ class AuraFlowPipeline(DiffusionPipeline):
         # 4. Prepare timesteps
 
         # sigmas = np.linspace(1.0, 1 / num_inference_steps, num_inference_steps)
-        timesteps, num_inference_steps = retrieve_timesteps(
-            self.scheduler, num_inference_steps, device, timesteps, sigmas
-        )
+        timesteps, num_inference_steps = retrieve_timesteps(self.scheduler, num_inference_steps, device, sigmas=sigmas)
 
         # 5. Prepare latents.
         latent_channels = self.transformer.config.in_channels

diffusers/pipelines/auto_pipeline.py

@@ -18,8 +18,10 @@ from collections import OrderedDict
 from huggingface_hub.utils import validate_hf_hub_args
 
 from ..configuration_utils import ConfigMixin
+from ..models.controlnets import ControlNetUnionModel
 from ..utils import is_sentencepiece_available
 from .aura_flow import AuraFlowPipeline
+from .cogview3 import CogView3PlusPipeline
 from .controlnet import (
     StableDiffusionControlNetImg2ImgPipeline,
     StableDiffusionControlNetInpaintPipeline,
@@ -27,9 +29,22 @@ from .controlnet import (
     StableDiffusionXLControlNetImg2ImgPipeline,
     StableDiffusionXLControlNetInpaintPipeline,
     StableDiffusionXLControlNetPipeline,
+    StableDiffusionXLControlNetUnionImg2ImgPipeline,
+    StableDiffusionXLControlNetUnionInpaintPipeline,
+    StableDiffusionXLControlNetUnionPipeline,
 )
 from .deepfloyd_if import IFImg2ImgPipeline, IFInpaintingPipeline, IFPipeline
-from .flux import FluxPipeline
+from .flux import (
+    FluxControlImg2ImgPipeline,
+    FluxControlInpaintPipeline,
+    FluxControlNetImg2ImgPipeline,
+    FluxControlNetInpaintPipeline,
+    FluxControlNetPipeline,
+    FluxControlPipeline,
+    FluxImg2ImgPipeline,
+    FluxInpaintPipeline,
+    FluxPipeline,
+)
 from .hunyuandit import HunyuanDiTPipeline
 from .kandinsky import (
     KandinskyCombinedPipeline,
@@ -49,12 +64,18 @@ from .kandinsky2_2 import (
 )
 from .kandinsky3 import Kandinsky3Img2ImgPipeline, Kandinsky3Pipeline
 from .latent_consistency_models import LatentConsistencyModelImg2ImgPipeline, LatentConsistencyModelPipeline
+from .lumina import LuminaText2ImgPipeline
 from .pag import (
     HunyuanDiTPAGPipeline,
     PixArtSigmaPAGPipeline,
+    StableDiffusion3PAGImg2ImgPipeline,
     StableDiffusion3PAGPipeline,
+    StableDiffusionControlNetPAGInpaintPipeline,
     StableDiffusionControlNetPAGPipeline,
+    StableDiffusionPAGImg2ImgPipeline,
+    StableDiffusionPAGInpaintPipeline,
     StableDiffusionPAGPipeline,
+    StableDiffusionXLControlNetPAGImg2ImgPipeline,
     StableDiffusionXLControlNetPAGPipeline,
     StableDiffusionXLPAGImg2ImgPipeline,
     StableDiffusionXLPAGInpaintPipeline,
@@ -94,6 +115,7 @@ AUTO_TEXT2IMAGE_PIPELINES_MAPPING = OrderedDict(
         ("kandinsky3", Kandinsky3Pipeline),
         ("stable-diffusion-controlnet", StableDiffusionControlNetPipeline),
         ("stable-diffusion-xl-controlnet", StableDiffusionXLControlNetPipeline),
+        ("stable-diffusion-xl-controlnet-union", StableDiffusionXLControlNetUnionPipeline),
         ("wuerstchen", WuerstchenCombinedPipeline),
         ("cascade", StableCascadeCombinedPipeline),
         ("lcm", LatentConsistencyModelPipeline),
@@ -106,6 +128,10 @@ AUTO_TEXT2IMAGE_PIPELINES_MAPPING = OrderedDict(
         ("pixart-sigma-pag", PixArtSigmaPAGPipeline),
         ("auraflow", AuraFlowPipeline),
         ("flux", FluxPipeline),
+        ("flux-control", FluxControlPipeline),
+        ("flux-controlnet", FluxControlNetPipeline),
+        ("lumina", LuminaText2ImgPipeline),
+        ("cogview3", CogView3PlusPipeline),
     ]
 )
 
@@ -114,14 +140,21 @@ AUTO_IMAGE2IMAGE_PIPELINES_MAPPING = OrderedDict(
         ("stable-diffusion", StableDiffusionImg2ImgPipeline),
         ("stable-diffusion-xl", StableDiffusionXLImg2ImgPipeline),
         ("stable-diffusion-3", StableDiffusion3Img2ImgPipeline),
+        ("stable-diffusion-3-pag", StableDiffusion3PAGImg2ImgPipeline),
         ("if", IFImg2ImgPipeline),
         ("kandinsky", KandinskyImg2ImgCombinedPipeline),
         ("kandinsky22", KandinskyV22Img2ImgCombinedPipeline),
         ("kandinsky3", Kandinsky3Img2ImgPipeline),
         ("stable-diffusion-controlnet", StableDiffusionControlNetImg2ImgPipeline),
+        ("stable-diffusion-pag", StableDiffusionPAGImg2ImgPipeline),
         ("stable-diffusion-xl-controlnet", StableDiffusionXLControlNetImg2ImgPipeline),
+        ("stable-diffusion-xl-controlnet-union", StableDiffusionXLControlNetUnionImg2ImgPipeline),
         ("stable-diffusion-xl-pag", StableDiffusionXLPAGImg2ImgPipeline),
+        ("stable-diffusion-xl-controlnet-pag", StableDiffusionXLControlNetPAGImg2ImgPipeline),
         ("lcm", LatentConsistencyModelImg2ImgPipeline),
+        ("flux", FluxImg2ImgPipeline),
+        ("flux-controlnet", FluxControlNetImg2ImgPipeline),
+        ("flux-control", FluxControlImg2ImgPipeline),
     ]
 )
 
@@ -134,8 +167,14 @@ AUTO_INPAINT_PIPELINES_MAPPING = OrderedDict(
         ("kandinsky", KandinskyInpaintCombinedPipeline),
         ("kandinsky22", KandinskyV22InpaintCombinedPipeline),
         ("stable-diffusion-controlnet", StableDiffusionControlNetInpaintPipeline),
+        ("stable-diffusion-controlnet-pag", StableDiffusionControlNetPAGInpaintPipeline),
         ("stable-diffusion-xl-controlnet", StableDiffusionXLControlNetInpaintPipeline),
+        ("stable-diffusion-xl-controlnet-union", StableDiffusionXLControlNetUnionInpaintPipeline),
         ("stable-diffusion-xl-pag", StableDiffusionXLPAGInpaintPipeline),
+        ("flux", FluxInpaintPipeline),
+        ("flux-controlnet", FluxControlNetInpaintPipeline),
+        ("flux-control", FluxControlInpaintPipeline),
+        ("stable-diffusion-pag", StableDiffusionPAGInpaintPipeline),
     ]
 )
 
@@ -161,12 +200,12 @@ _AUTO_INPAINT_DECODER_PIPELINES_MAPPING = OrderedDict(
 )
 
 if is_sentencepiece_available():
-    from .kolors import KolorsPipeline
+    from .kolors import KolorsImg2ImgPipeline, KolorsPipeline
     from .pag import KolorsPAGPipeline
 
     AUTO_TEXT2IMAGE_PIPELINES_MAPPING["kolors"] = KolorsPipeline
     AUTO_TEXT2IMAGE_PIPELINES_MAPPING["kolors-pag"] = KolorsPAGPipeline
-    AUTO_IMAGE2IMAGE_PIPELINES_MAPPING["kolors"] = KolorsPipeline
+    AUTO_IMAGE2IMAGE_PIPELINES_MAPPING["kolors"] = KolorsImg2ImgPipeline
 
 SUPPORTED_TASKS_MAPPINGS = [
     AUTO_TEXT2IMAGE_PIPELINES_MAPPING,
@@ -368,13 +407,20 @@ class AutoPipelineForText2Image(ConfigMixin):
 
         config = cls.load_config(pretrained_model_or_path, **load_config_kwargs)
         orig_class_name = config["_class_name"]
+        if "ControlPipeline" in orig_class_name:
+            to_replace = "ControlPipeline"
+        else:
+            to_replace = "Pipeline"
 
         if "controlnet" in kwargs:
-            orig_class_name = config["_class_name"].replace("Pipeline", "ControlNetPipeline")
+            if isinstance(kwargs["controlnet"], ControlNetUnionModel):
+                orig_class_name = config["_class_name"].replace(to_replace, "ControlNetUnionPipeline")
+            else:
+                orig_class_name = config["_class_name"].replace(to_replace, "ControlNetPipeline")
         if "enable_pag" in kwargs:
             enable_pag = kwargs.pop("enable_pag")
             if enable_pag:
-                orig_class_name = orig_class_name.replace("Pipeline", "PAGPipeline")
+                orig_class_name = orig_class_name.replace(to_replace, "PAGPipeline")
 
         text_2_image_cls = _get_task_class(AUTO_TEXT2IMAGE_PIPELINES_MAPPING, orig_class_name)
 
@@ -656,12 +702,29 @@ class AutoPipelineForImage2Image(ConfigMixin):
         config = cls.load_config(pretrained_model_or_path, **load_config_kwargs)
         orig_class_name = config["_class_name"]
 
+        # the `orig_class_name` can be:
+        # `- *Pipeline` (for regular text-to-image checkpoint)
+        #  - `*ControlPipeline` (for Flux tools specific checkpoint)
+        # `- *Img2ImgPipeline` (for refiner checkpoint)
+        if "Img2Img" in orig_class_name:
+            to_replace = "Img2ImgPipeline"
+        elif "ControlPipeline" in orig_class_name:
+            to_replace = "ControlPipeline"
+        else:
+            to_replace = "Pipeline"
+
         if "controlnet" in kwargs:
-            orig_class_name = config["_class_name"].replace("Pipeline", "ControlNetPipeline")
+            if isinstance(kwargs["controlnet"], ControlNetUnionModel):
+                orig_class_name = orig_class_name.replace(to_replace, "ControlNetUnion" + to_replace)
+            else:
+                orig_class_name = orig_class_name.replace(to_replace, "ControlNet" + to_replace)
         if "enable_pag" in kwargs:
             enable_pag = kwargs.pop("enable_pag")
             if enable_pag:
-                orig_class_name = orig_class_name.replace("Pipeline", "PAGPipeline")
+                orig_class_name = orig_class_name.replace(to_replace, "PAG" + to_replace)
+
+        if to_replace == "ControlPipeline":
+            orig_class_name = orig_class_name.replace(to_replace, "ControlImg2ImgPipeline")
 
         image_2_image_cls = _get_task_class(AUTO_IMAGE2IMAGE_PIPELINES_MAPPING, orig_class_name)
 
@@ -948,13 +1011,28 @@ class AutoPipelineForInpainting(ConfigMixin):
         config = cls.load_config(pretrained_model_or_path, **load_config_kwargs)
         orig_class_name = config["_class_name"]
 
+        # The `orig_class_name`` can be:
+        # `- *InpaintPipeline` (for inpaint-specific checkpoint)
+        #  - `*ControlPipeline` (for Flux tools specific checkpoint)
+        #  - or *Pipeline (for regular text-to-image checkpoint)
+        if "Inpaint" in orig_class_name:
+            to_replace = "InpaintPipeline"
+        elif "ControlPipeline" in orig_class_name:
+            to_replace = "ControlPipeline"
+        else:
+            to_replace = "Pipeline"
+
         if "controlnet" in kwargs:
-            orig_class_name = config["_class_name"].replace("Pipeline", "ControlNetPipeline")
+            if isinstance(kwargs["controlnet"], ControlNetUnionModel):
+                orig_class_name = orig_class_name.replace(to_replace, "ControlNetUnion" + to_replace)
+            else:
+                orig_class_name = orig_class_name.replace(to_replace, "ControlNet" + to_replace)
         if "enable_pag" in kwargs:
             enable_pag = kwargs.pop("enable_pag")
             if enable_pag:
-                orig_class_name = config["_class_name"].replace("Pipeline", "PAGPipeline")
-
+                orig_class_name = orig_class_name.replace(to_replace, "PAG" + to_replace)
+        if to_replace == "ControlPipeline":
+            orig_class_name = orig_class_name.replace(to_replace, "ControlInpaintPipeline")
         inpainting_cls = _get_task_class(AUTO_INPAINT_PIPELINES_MAPPING, orig_class_name)
 
         kwargs = {**load_config_kwargs, **kwargs}

diffusers/pipelines/blip_diffusion/modeling_blip2.py

@@ -167,7 +167,7 @@ class Blip2QFormerEncoder(nn.Module):
             layer_head_mask = head_mask[i] if head_mask is not None else None
             past_key_value = past_key_values[i] if past_key_values is not None else None
 
-            if getattr(self.config, "gradient_checkpointing", False) and self.training:
+            if getattr(self.config, "gradient_checkpointing", False) and torch.is_grad_enabled():
                 if use_cache:
                     logger.warning(
                         "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."

diffusers/pipelines/cogvideo/__init__.py

@@ -23,6 +23,7 @@ except OptionalDependencyNotAvailable:
     _dummy_objects.update(get_objects_from_module(dummy_torch_and_transformers_objects))
 else:
     _import_structure["pipeline_cogvideox"] = ["CogVideoXPipeline"]
+    _import_structure["pipeline_cogvideox_fun_control"] = ["CogVideoXFunControlPipeline"]
     _import_structure["pipeline_cogvideox_image2video"] = ["CogVideoXImageToVideoPipeline"]
     _import_structure["pipeline_cogvideox_video2video"] = ["CogVideoXVideoToVideoPipeline"]
 
@@ -35,6 +36,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
         from ...utils.dummy_torch_and_transformers_objects import *
     else:
         from .pipeline_cogvideox import CogVideoXPipeline
+        from .pipeline_cogvideox_fun_control import CogVideoXFunControlPipeline
         from .pipeline_cogvideox_image2video import CogVideoXImageToVideoPipeline
         from .pipeline_cogvideox_video2video import CogVideoXVideoToVideoPipeline
 

diffusers/pipelines/cogvideo/pipeline_cogvideox.py

@@ -15,12 +15,13 @@
 
 import inspect
 import math
-from typing import Callable, Dict, List, Optional, Tuple, Union
+from typing import Any, Callable, Dict, List, Optional, Tuple, Union
 
 import torch
 from transformers import T5EncoderModel, T5Tokenizer
 
 from ...callbacks import MultiPipelineCallbacks, PipelineCallback
+from ...loaders import CogVideoXLoraLoaderMixin
 from ...models import AutoencoderKLCogVideoX, CogVideoXTransformer3DModel
 from ...models.embeddings import get_3d_rotary_pos_embed
 from ...pipelines.pipeline_utils import DiffusionPipeline
@@ -85,7 +86,7 @@ def retrieve_timesteps(
     sigmas: Optional[List[float]] = None,
     **kwargs,
 ):
-    """
+    r"""
     Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
     custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
 
@@ -136,7 +137,7 @@ def retrieve_timesteps(
     return timesteps, num_inference_steps
 
 
-class CogVideoXPipeline(DiffusionPipeline):
+class CogVideoXPipeline(DiffusionPipeline, CogVideoXLoraLoaderMixin):
     r"""
     Pipeline for text-to-video generation using CogVideoX.
 
@@ -187,6 +188,9 @@ class CogVideoXPipeline(DiffusionPipeline):
         self.vae_scale_factor_temporal = (
             self.vae.config.temporal_compression_ratio if hasattr(self, "vae") and self.vae is not None else 4
         )
+        self.vae_scaling_factor_image = (
+            self.vae.config.scaling_factor if hasattr(self, "vae") and self.vae is not None else 0.7
+        )
 
         self.video_processor = VideoProcessor(vae_scale_factor=self.vae_scale_factor_spatial)
 
@@ -316,6 +320,12 @@ class CogVideoXPipeline(DiffusionPipeline):
     def prepare_latents(
         self, batch_size, num_channels_latents, num_frames, height, width, dtype, device, generator, latents=None
     ):
+        if isinstance(generator, list) and len(generator) != batch_size:
+            raise ValueError(
+                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."
+            )
+
         shape = (
             batch_size,
             (num_frames - 1) // self.vae_scale_factor_temporal + 1,
@@ -323,11 +333,6 @@ class CogVideoXPipeline(DiffusionPipeline):
             height // self.vae_scale_factor_spatial,
             width // self.vae_scale_factor_spatial,
         )
-        if isinstance(generator, list) and len(generator) != batch_size:
-            raise ValueError(
-                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."
-            )
 
         if latents is None:
             latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
@@ -340,7 +345,7 @@ class CogVideoXPipeline(DiffusionPipeline):
 
     def decode_latents(self, latents: torch.Tensor) -> torch.Tensor:
         latents = latents.permute(0, 2, 1, 3, 4)  # [batch_size, num_channels, num_frames, height, width]
-        latents = 1 / self.vae.config.scaling_factor * latents
+        latents = 1 / self.vae_scaling_factor_image * latents
 
         frames = self.vae.decode(latents).sample
         return frames
@@ -437,22 +442,39 @@ class CogVideoXPipeline(DiffusionPipeline):
     ) -> Tuple[torch.Tensor, torch.Tensor]:
         grid_height = height // (self.vae_scale_factor_spatial * self.transformer.config.patch_size)
         grid_width = width // (self.vae_scale_factor_spatial * self.transformer.config.patch_size)
-        base_size_width = 720 // (self.vae_scale_factor_spatial * self.transformer.config.patch_size)
-        base_size_height = 480 // (self.vae_scale_factor_spatial * self.transformer.config.patch_size)
 
-        grid_crops_coords = get_resize_crop_region_for_grid(
-            (grid_height, grid_width), base_size_width, base_size_height
-        )
-        freqs_cos, freqs_sin = get_3d_rotary_pos_embed(
-            embed_dim=self.transformer.config.attention_head_dim,
-            crops_coords=grid_crops_coords,
-            grid_size=(grid_height, grid_width),
-            temporal_size=num_frames,
-            use_real=True,
-        )
+        p = self.transformer.config.patch_size
+        p_t = self.transformer.config.patch_size_t
+
+        base_size_width = self.transformer.config.sample_width // p
+        base_size_height = self.transformer.config.sample_height // p
+
+        if p_t is None:
+            # CogVideoX 1.0
+            grid_crops_coords = get_resize_crop_region_for_grid(
+                (grid_height, grid_width), base_size_width, base_size_height
+            )
+            freqs_cos, freqs_sin = get_3d_rotary_pos_embed(
+                embed_dim=self.transformer.config.attention_head_dim,
+                crops_coords=grid_crops_coords,
+                grid_size=(grid_height, grid_width),
+                temporal_size=num_frames,
+                device=device,
+            )
+        else:
+            # CogVideoX 1.5
+            base_num_frames = (num_frames + p_t - 1) // p_t
+
+            freqs_cos, freqs_sin = get_3d_rotary_pos_embed(
+                embed_dim=self.transformer.config.attention_head_dim,
+                crops_coords=None,
+                grid_size=(grid_height, grid_width),
+                temporal_size=base_num_frames,
+                grid_type="slice",
+                max_size=(base_size_height, base_size_width),
+                device=device,
+            )
 
-        freqs_cos = freqs_cos.to(device=device)
-        freqs_sin = freqs_sin.to(device=device)
         return freqs_cos, freqs_sin
 
     @property
@@ -463,6 +485,10 @@ class CogVideoXPipeline(DiffusionPipeline):
     def num_timesteps(self):
         return self._num_timesteps
 
+    @property
+    def attention_kwargs(self):
+        return self._attention_kwargs
+
     @property
     def interrupt(self):
         return self._interrupt
@@ -473,9 +499,9 @@ class CogVideoXPipeline(DiffusionPipeline):
         self,
         prompt: Optional[Union[str, List[str]]] = None,
         negative_prompt: Optional[Union[str, List[str]]] = None,
-        height: int = 480,
-        width: int = 720,
-        num_frames: int = 49,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
+        num_frames: Optional[int] = None,
         num_inference_steps: int = 50,
         timesteps: Optional[List[int]] = None,
         guidance_scale: float = 6,
@@ -488,6 +514,7 @@ class CogVideoXPipeline(DiffusionPipeline):
         negative_prompt_embeds: Optional[torch.FloatTensor] = None,
         output_type: str = "pil",
         return_dict: bool = True,
+        attention_kwargs: Optional[Dict[str, Any]] = None,
         callback_on_step_end: Optional[
             Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks]
         ] = None,
@@ -505,14 +532,14 @@ class CogVideoXPipeline(DiffusionPipeline):
                 The prompt or prompts not to guide the image generation. If not defined, one has to pass
                 `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
                 less than `1`).
-            height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
-                The height in pixels of the generated image. This is set to 1024 by default for the best results.
-            width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
-                The width in pixels of the generated image. This is set to 1024 by default for the best results.
+            height (`int`, *optional*, defaults to self.transformer.config.sample_height * self.vae_scale_factor_spatial):
+                The height in pixels of the generated image. This is set to 480 by default for the best results.
+            width (`int`, *optional*, defaults to self.transformer.config.sample_height * self.vae_scale_factor_spatial):
+                The width in pixels of the generated image. This is set to 720 by default for the best results.
             num_frames (`int`, defaults to `48`):
                 Number of frames to generate. Must be divisible by self.vae_scale_factor_temporal. Generated video will
                 contain 1 extra frame because CogVideoX is conditioned with (num_seconds * fps + 1) frames where
-                num_seconds is 6 and fps is 4. However, since videos can be saved at any fps, the only condition that
+                num_seconds is 6 and fps is 8. However, since videos can be saved at any fps, the only condition that
                 needs to be satisfied is that of divisibility mentioned above.
             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
@@ -549,6 +576,10 @@ class CogVideoXPipeline(DiffusionPipeline):
             return_dict (`bool`, *optional*, defaults to `True`):
                 Whether or not to return a [`~pipelines.stable_diffusion_xl.StableDiffusionXLPipelineOutput`] instead
                 of a plain tuple.
+            attention_kwargs (`dict`, *optional*):
+                A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
+                `self.processor` in
+                [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
             callback_on_step_end (`Callable`, *optional*):
                 A function that calls at the end of each denoising steps during the inference. The function is called
                 with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int,
@@ -570,16 +601,13 @@ class CogVideoXPipeline(DiffusionPipeline):
             `tuple`. When returning a tuple, the first element is a list with the generated images.
         """
 
-        if num_frames > 49:
-            raise ValueError(
-                "The number of frames must be less than 49 for now due to static positional embeddings. This will be updated in the future to remove this limitation."
-            )
-
         if isinstance(callback_on_step_end, (PipelineCallback, MultiPipelineCallbacks)):
             callback_on_step_end_tensor_inputs = callback_on_step_end.tensor_inputs
 
-        height = height or self.transformer.config.sample_size * self.vae_scale_factor_spatial
-        width = width or self.transformer.config.sample_size * self.vae_scale_factor_spatial
+        height = height or self.transformer.config.sample_height * self.vae_scale_factor_spatial
+        width = width or self.transformer.config.sample_width * self.vae_scale_factor_spatial
+        num_frames = num_frames or self.transformer.config.sample_frames
+
         num_videos_per_prompt = 1
 
         # 1. Check inputs. Raise error if not correct
@@ -593,6 +621,7 @@ class CogVideoXPipeline(DiffusionPipeline):
             negative_prompt_embeds,
         )
         self._guidance_scale = guidance_scale
+        self._attention_kwargs = attention_kwargs
         self._interrupt = False
 
         # 2. Default call parameters
@@ -628,7 +657,16 @@ class CogVideoXPipeline(DiffusionPipeline):
         timesteps, num_inference_steps = retrieve_timesteps(self.scheduler, num_inference_steps, device, timesteps)
         self._num_timesteps = len(timesteps)
 
-        # 5. Prepare latents.
+        # 5. Prepare latents
+        latent_frames = (num_frames - 1) // self.vae_scale_factor_temporal + 1
+
+        # For CogVideoX 1.5, the latent frames should be padded to make it divisible by patch_size_t
+        patch_size_t = self.transformer.config.patch_size_t
+        additional_frames = 0
+        if patch_size_t is not None and latent_frames % patch_size_t != 0:
+            additional_frames = patch_size_t - latent_frames % patch_size_t
+            num_frames += additional_frames * self.vae_scale_factor_temporal
+
         latent_channels = self.transformer.config.in_channels
         latents = self.prepare_latents(
             batch_size * num_videos_per_prompt,
@@ -674,6 +712,7 @@ class CogVideoXPipeline(DiffusionPipeline):
                     encoder_hidden_states=prompt_embeds,
                     timestep=timestep,
                     image_rotary_emb=image_rotary_emb,
+                    attention_kwargs=attention_kwargs,
                     return_dict=False,
                 )[0]
                 noise_pred = noise_pred.float()
@@ -717,6 +756,8 @@ class CogVideoXPipeline(DiffusionPipeline):
                     progress_bar.update()
 
         if not output_type == "latent":
+            # Discard any padding frames that were added for CogVideoX 1.5
+            latents = latents[:, additional_frames:]
             video = self.decode_latents(latents)
             video = self.video_processor.postprocess_video(video=video, output_type=output_type)
         else: