Diffusers documentation
SkyReels-V2: Infinite-length Film Generative model
SkyReels-V2: Infinite-length Film Generative model
SkyReels-V2 by the SkyReels Team.
Recent advances in video generation have been driven by diffusion models and autoregressive frameworks, yet critical challenges persist in harmonizing prompt adherence, visual quality, motion dynamics, and duration: compromises in motion dynamics to enhance temporal visual quality, constrained video duration (5-10 seconds) to prioritize resolution, and inadequate shot-aware generation stemming from general-purpose MLLMs’ inability to interpret cinematic grammar, such as shot composition, actor expressions, and camera motions. These intertwined limitations hinder realistic long-form synthesis and professional film-style generation. To address these limitations, we propose SkyReels-V2, an Infinite-length Film Generative Model, that synergizes Multi-modal Large Language Model (MLLM), Multi-stage Pretraining, Reinforcement Learning, and Diffusion Forcing Framework. Firstly, we design a comprehensive structural representation of video that combines the general descriptions by the Multi-modal LLM and the detailed shot language by sub-expert models. Aided with human annotation, we then train a unified Video Captioner, named SkyCaptioner-V1, to efficiently label the video data. Secondly, we establish progressive-resolution pretraining for the fundamental video generation, followed by a four-stage post-training enhancement: Initial concept-balanced Supervised Fine-Tuning (SFT) improves baseline quality; Motion-specific Reinforcement Learning (RL) training with human-annotated and synthetic distortion data addresses dynamic artifacts; Our diffusion forcing framework with non-decreasing noise schedules enables long-video synthesis in an efficient search space; Final high-quality SFT refines visual fidelity. All the code and models are available at this https URL.
You can find all the original SkyReels-V2 checkpoints under the Skywork organization.
The following SkyReels-V2 models are supported in Diffusers:
- SkyReels-V2 DF 1.3B - 540P
- SkyReels-V2 DF 14B - 540P
- SkyReels-V2 DF 14B - 720P
- SkyReels-V2 T2V 14B - 540P
- SkyReels-V2 T2V 14B - 720P
- SkyReels-V2 I2V 1.3B - 540P
- SkyReels-V2 I2V 14B - 540P
- SkyReels-V2 I2V 14B - 720P
- SkyReels-V2 FLF2V 1.3B - 540P
Click on the SkyReels-V2 models in the right sidebar for more examples of video generation.
A Visual Demonstration
An example with these parameters: base_num_frames=97, num_frames=97, num_inference_steps=30, ar_step=5, causal_block_size=5
vae_scale_factor_temporal -> 4 num_latent_frames: (97-1)//vae_scale_factor_temporal+1 = 25 frames -> 5 blocks of 5 frames each
base_num_latent_frames = (97-1)//vae_scale_factor_temporal+1 = 25 → blocks = 25//5 = 5 blocks This 5 blocks means the maximum context length of the model is 25 frames in the latent space.
Asynchronous Processing Timeline: ┌─────────────────────────────────────────────────────────────────┐ │ Steps: 1 6 11 16 21 26 31 36 41 46 50 │ │ Block 1: [■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■] │ │ Block 2: [■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■] │ │ Block 3: [■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■] │ │ Block 4: [■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■] │ │ Block 5: [■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■] │ └─────────────────────────────────────────────────────────────────┘
For Long Videos (num_frames > base_num_frames): base_num_frames acts as the “sliding window size” for processing long videos.
Example: 257-frame video with base_num_frames=97, overlap_history=17 ┌──── Iteration 1 (frames 1-97) ────┐ │ Processing window: 97 frames │ → 5 blocks, async processing │ Generates: frames 1-97 │ └───────────────────────────────────┘ ┌────── Iteration 2 (frames 81-177) ──────┐ │ Processing window: 97 frames │ │ Overlap: 17 frames (81-97) from prev │ → 5 blocks, async processing │ Generates: frames 98-177 │ └─────────────────────────────────────────┘ ┌────── Iteration 3 (frames 161-257) ──────┐ │ Processing window: 97 frames │ │ Overlap: 17 frames (161-177) from prev │ → 5 blocks, async processing │ Generates: frames 178-257 │ └──────────────────────────────────────────┘
Each iteration independently runs the asynchronous processing with its own 5 blocks. base_num_frames controls:
Memory usage (larger window = more VRAM)
Model context length (must match training constraints)
Number of blocks per iteration (base_num_latent_frames // causal_block_size)
Each block takes 30 steps to complete denoising. Block N starts at step: 1 + (N-1) x ar_step Total steps: 30 + (5-1) x 5 = 50 steps
Synchronous mode (ar_step=0) would process all blocks/frames simultaneously: ┌──────────────────────────────────────────────┐ │ Steps: 1 … 30 │ │ All blocks: [■■■■■■■■■■■■■■■■■■■■■■■■■■■■■■] │ └──────────────────────────────────────────────┘ Total steps: 30 steps
An example on how the step matrix is constructed for asynchronous processing: Given the parameters: (num_inference_steps=30, flow_shift=8, num_frames=97, ar_step=5, causal_block_size=5)
num_latent_frames = (97 frames - 1) // (4 temporal downsampling) + 1 = 25
step_template = [999, 995, 991, 986, 980, 975, 969, 963, 956, 948, 941, 932, 922, 912, 901, 888, 874, 859, 841, 822, 799, 773, 743, 708, 666, 615, 551, 470, 363, 216]
The algorithm creates a 50x25 step_matrix where:
Row 1: [999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999]
Row 2: [995, 995, 995, 995, 995, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999]
Row 3: [991, 991, 991, 991, 991, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999]
…
Row 7: [969, 969, 969, 969, 969, 995, 995, 995, 995, 995, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999, 999]
…
Row 21: [799, 799, 799, 799, 799, 888, 888, 888, 888, 888, 941, 941, 941, 941, 941, 975, 975, 975, 975, 975, 999, 999, 999, 999, 999]
…
Row 35: [ 0, 0, 0, 0, 0, 216, 216, 216, 216, 216, 666, 666, 666, 666, 666, 822, 822, 822, 822, 822, 901, 901, 901, 901, 901]
…
Row 42: [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 551, 551, 551, 551, 551, 773, 773, 773, 773, 773]
…
Row 50: [ 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 216, 216, 216, 216, 216]
Detailed Row 6 Analysis:
step_matrix[5]: [ 975, 975, 975, 975, 975, 999, 999, 999, 999, 999, 999, …, 999]
step_index[5]: [ 6, 6, 6, 6, 6, 1, 1, 1, 1, 1, 0, …, 0]
step_update_mask[5]: [True,True,True,True,True,True,True,True,True,True,False, …,False]
valid_interval[5]: (0, 25)
Key Pattern: Block i lags behind Block i-1 by exactly ar_step=5 timesteps, creating the staggered “diffusion forcing” effect where later blocks condition on cleaner earlier blocks.
Text-to-Video Generation
The example below demonstrates how to generate a video from text.
Refer to the Reduce memory usage guide for more details about the various memory saving techniques.
From the original repo:
You can use —ar_step 5 to enable asynchronous inference. When asynchronous inference, —causal_block_size 5 is recommended while it is not supposed to be set for synchronous generation… Asynchronous inference will take more steps to diffuse the whole sequence which means it will be SLOWER than synchronous mode. In our experiments, asynchronous inference may improve the instruction following and visual consistent performance.
# pip install ftfy
import torch
from diffusers import AutoModel, SkyReelsV2DiffusionForcingPipeline, UniPCMultistepScheduler
from diffusers.utils import export_to_video
vae = AutoModel.from_pretrained("Skywork/SkyReels-V2-DF-14B-540P-Diffusers", subfolder="vae", torch_dtype=torch.float32)
transformer = AutoModel.from_pretrained("Skywork/SkyReels-V2-DF-14B-540P-Diffusers", subfolder="transformer", torch_dtype=torch.bfloat16)
pipeline = SkyReelsV2DiffusionForcingPipeline.from_pretrained(
"Skywork/SkyReels-V2-DF-14B-540P-Diffusers",
vae=vae,
transformer=transformer,
torch_dtype=torch.bfloat16
)
flow_shift = 8.0 # 8.0 for T2V, 5.0 for I2V
pipeline.scheduler = UniPCMultistepScheduler.from_config(pipeline.scheduler.config, flow_shift=flow_shift)
pipeline = pipeline.to("cuda")
prompt = "A cat and a dog baking a cake together in a kitchen. The cat is carefully measuring flour, while the dog is stirring the batter with a wooden spoon. The kitchen is cozy, with sunlight streaming through the window."
output = pipeline(
prompt=prompt,
num_inference_steps=30,
height=544, # 720 for 720P
width=960, # 1280 for 720P
num_frames=97,
base_num_frames=97, # 121 for 720P
ar_step=5, # Controls asynchronous inference (0 for synchronous mode)
causal_block_size=5, # Number of frames in each block for asynchronous processing
overlap_history=None, # Number of frames to overlap for smooth transitions in long videos; 17 for long video generations
addnoise_condition=20, # Improves consistency in long video generation
).frames[0]
export_to_video(output, "T2V.mp4", fps=24, quality=8)First-Last-Frame-to-Video Generation
The example below demonstrates how to use the image-to-video pipeline to generate a video using a text description, a starting frame, and an ending frame.
import numpy as np
import torch
import torchvision.transforms.functional as TF
from diffusers import AutoencoderKLWan, SkyReelsV2DiffusionForcingImageToVideoPipeline, UniPCMultistepScheduler
from diffusers.utils import export_to_video, load_image
model_id = "Skywork/SkyReels-V2-DF-14B-720P-Diffusers"
vae = AutoencoderKLWan.from_pretrained(model_id, subfolder="vae", torch_dtype=torch.float32)
pipeline = SkyReelsV2DiffusionForcingImageToVideoPipeline.from_pretrained(
model_id, vae=vae, torch_dtype=torch.bfloat16
)
flow_shift = 5.0 # 8.0 for T2V, 5.0 for I2V
pipeline.scheduler = UniPCMultistepScheduler.from_config(pipeline.scheduler.config, flow_shift=flow_shift)
pipeline.to("cuda")
first_frame = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/flf2v_input_first_frame.png")
last_frame = load_image("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/flf2v_input_last_frame.png")
def aspect_ratio_resize(image, pipeline, max_area=720 * 1280):
aspect_ratio = image.height / image.width
mod_value = pipeline.vae_scale_factor_spatial * pipeline.transformer.config.patch_size[1]
height = round(np.sqrt(max_area * aspect_ratio)) // mod_value * mod_value
width = round(np.sqrt(max_area / aspect_ratio)) // mod_value * mod_value
image = image.resize((width, height))
return image, height, width
def center_crop_resize(image, height, width):
# Calculate resize ratio to match first frame dimensions
resize_ratio = max(width / image.width, height / image.height)
# Resize the image
width = round(image.width * resize_ratio)
height = round(image.height * resize_ratio)
size = [width, height]
image = TF.center_crop(image, size)
return image, height, width
first_frame, height, width = aspect_ratio_resize(first_frame, pipeline)
if last_frame.size != first_frame.size:
last_frame, _, _ = center_crop_resize(last_frame, height, width)
prompt = "CG animation style, a small blue bird takes off from the ground, flapping its wings. The bird's feathers are delicate, with a unique pattern on its chest. The background shows a blue sky with white clouds under bright sunshine. The camera follows the bird upward, capturing its flight and the vastness of the sky from a close-up, low-angle perspective."
output = pipeline(
image=first_frame, last_image=last_frame, prompt=prompt, height=height, width=width, guidance_scale=5.0
).frames[0]
export_to_video(output, "output.mp4", fps=24, quality=8)Video-to-Video Generation
SkyReelsV2DiffusionForcingVideoToVideoPipeline extends a given video.
import numpy as np
import torch
import torchvision.transforms.functional as TF
from diffusers import AutoencoderKLWan, SkyReelsV2DiffusionForcingVideoToVideoPipeline, UniPCMultistepScheduler
from diffusers.utils import export_to_video, load_video
model_id = "Skywork/SkyReels-V2-DF-14B-540P-Diffusers"
vae = AutoencoderKLWan.from_pretrained(model_id, subfolder="vae", torch_dtype=torch.float32)
pipeline = SkyReelsV2DiffusionForcingVideoToVideoPipeline.from_pretrained(
model_id, vae=vae, torch_dtype=torch.bfloat16
)
flow_shift = 5.0 # 8.0 for T2V, 5.0 for I2V
pipeline.scheduler = UniPCMultistepScheduler.from_config(pipeline.scheduler.config, flow_shift=flow_shift)
pipeline.to("cuda")
video = load_video("input_video.mp4")
prompt = "CG animation style, a small blue bird takes off from the ground, flapping its wings. The bird's feathers are delicate, with a unique pattern on its chest. The background shows a blue sky with white clouds under bright sunshine. The camera follows the bird upward, capturing its flight and the vastness of the sky from a close-up, low-angle perspective."
output = pipeline(
video=video, prompt=prompt, height=544, width=960, guidance_scale=5.0,
num_inference_steps=30, num_frames=257, base_num_frames=97#, ar_step=5, causal_block_size=5,
).frames[0]
export_to_video(output, "output.mp4", fps=24, quality=8)
# Total frames will be the number of frames of given video + 257Notes
SkyReels-V2 supports LoRAs with load_lora_weights().
Show example code
# pip install ftfy import torch from diffusers import AutoModel, SkyReelsV2DiffusionForcingPipeline from diffusers.utils import export_to_video vae = AutoModel.from_pretrained( "Skywork/SkyReels-V2-DF-1.3B-540P-Diffusers", subfolder="vae", torch_dtype=torch.float32 ) pipeline = SkyReelsV2DiffusionForcingPipeline.from_pretrained( "Skywork/SkyReels-V2-DF-1.3B-540P-Diffusers", vae=vae, torch_dtype=torch.bfloat16 ) pipeline.to("cuda") pipeline.load_lora_weights("benjamin-paine/steamboat-willie-1.3b", adapter_name="steamboat-willie") pipeline.set_adapters("steamboat-willie") pipeline.enable_model_cpu_offload() # use "steamboat willie style" to trigger the LoRA prompt = """ steamboat willie style, golden era animation, The camera rushes from far to near in a low-angle shot, revealing a white ferret on a log. It plays, leaps into the water, and emerges, as the camera zooms in for a close-up. Water splashes berry bushes nearby, while moss, snow, and leaves blanket the ground. Birch trees and a light blue sky frame the scene, with ferns in the foreground. Side lighting casts dynamic shadows and warm highlights. Medium composition, front view, low angle, with depth of field. """ output = pipeline( prompt=prompt, num_frames=97, guidance_scale=6.0, ).frames[0] export_to_video(output, "output.mp4", fps=24)
SkyReelsV2DiffusionForcingPipeline
class diffusers.SkyReelsV2DiffusionForcingPipeline
< source >( tokenizer: AutoTokenizer text_encoder: UMT5EncoderModel transformer: SkyReelsV2Transformer3DModel vae: AutoencoderKLWan scheduler: UniPCMultistepScheduler )
Parameters
- tokenizer (
AutoTokenizer) — Tokenizer from T5, specifically the google/umt5-xxl variant. - text_encoder (
UMT5EncoderModel) — T5, specifically the google/umt5-xxl variant. - transformer (SkyReelsV2Transformer3DModel) — Conditional Transformer to denoise the encoded image latents.
- scheduler (UniPCMultistepScheduler) —
A scheduler to be used in combination with
transformerto denoise the encoded image latents. - vae (AutoencoderKLWan) — Variational Auto-Encoder (VAE) Model to encode and decode videos to and from latent representations.
Pipeline for Text-to-Video (t2v) generation using SkyReels-V2 with diffusion forcing.
This model inherits from DiffusionPipeline. Check the superclass documentation for the generic methods implemented for all pipelines (downloading, saving, running on a specific device, etc.).
__call__
< source >( prompt: typing.Union[str, typing.List[str]] negative_prompt: typing.Union[str, typing.List[str]] = None height: int = 544 width: int = 960 num_frames: int = 97 num_inference_steps: int = 50 guidance_scale: float = 6.0 num_videos_per_prompt: typing.Optional[int] = 1 generator: typing.Union[torch._C.Generator, typing.List[torch._C.Generator], NoneType] = None latents: typing.Optional[torch.Tensor] = None prompt_embeds: typing.Optional[torch.Tensor] = None negative_prompt_embeds: typing.Optional[torch.Tensor] = None output_type: typing.Optional[str] = 'np' return_dict: bool = True attention_kwargs: typing.Optional[typing.Dict[str, typing.Any]] = None callback_on_step_end: typing.Union[typing.Callable[[int, int, typing.Dict], NoneType], diffusers.callbacks.PipelineCallback, diffusers.callbacks.MultiPipelineCallbacks, NoneType] = None callback_on_step_end_tensor_inputs: typing.List[str] = ['latents'] max_sequence_length: int = 512 overlap_history: typing.Optional[int] = None addnoise_condition: float = 0 base_num_frames: int = 97 ar_step: int = 0 causal_block_size: typing.Optional[int] = None fps: int = 24 ) → ~SkyReelsV2PipelineOutput or tuple
Parameters
- prompt (
strorList[str], optional) — The prompt or prompts to guide the image generation. If not defined, one has to passprompt_embeds. instead. - negative_prompt (
strorList[str], optional) — The prompt or prompts not to guide the image generation. If not defined, one has to passnegative_prompt_embedsinstead. Ignored when not using guidance (i.e., ignored ifguidance_scaleis less than1). - height (
int, defaults to544) — The height of the generated video. - width (
int, defaults to960) — The width of the generated video. - num_frames (
int, defaults to97) — The number of frames in the generated video. - num_inference_steps (
int, defaults to50) — The number of denoising steps. More denoising steps usually lead to a higher quality image at the expense of slower inference. - guidance_scale (
float, defaults to6.0) — Guidance scale as defined in Classifier-Free Diffusion Guidance.guidance_scaleis defined aswof equation 2. of Imagen Paper. Guidance scale is enabled by settingguidance_scale > 1. Higher guidance scale encourages to generate images that are closely linked to the textprompt, usually at the expense of lower image quality. (6.0 for T2V, 5.0 for I2V) - num_videos_per_prompt (
int, optional, defaults to 1) — The number of images to generate per prompt. - generator (
torch.GeneratororList[torch.Generator], optional) — Atorch.Generatorto make generation deterministic. - latents (
torch.Tensor, optional) — Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image generation. Can be used to tweak the same generation with different prompts. If not provided, a latents tensor is generated by sampling using the supplied randomgenerator. - prompt_embeds (
torch.Tensor, optional) — Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not provided, text embeddings are generated from thepromptinput argument. - negative_prompt_embeds (
torch.Tensor, optional) — Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not provided, text embeddings are generated from thenegative_promptinput argument. - output_type (
str, optional, defaults to"np") — The output format of the generated image. Choose betweenPIL.Imageornp.array. - return_dict (
bool, optional, defaults toTrue) — Whether or not to return aSkyReelsV2PipelineOutputinstead of a plain tuple. - attention_kwargs (
dict, optional) — A kwargs dictionary that if specified is passed along to theAttentionProcessoras defined underself.processorin diffusers.models.attention_processor. - callback_on_step_end (
Callable,PipelineCallback,MultiPipelineCallbacks, optional) — A function or a subclass ofPipelineCallbackorMultiPipelineCallbacksthat is called at the end of each denoising step during the inference. with the following arguments:callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict).callback_kwargswill include a list of all tensors as specified bycallback_on_step_end_tensor_inputs. - callback_on_step_end_tensor_inputs (
List, optional) — The list of tensor inputs for thecallback_on_step_endfunction. The tensors specified in the list will be passed ascallback_kwargsargument. You will only be able to include variables listed in the._callback_tensor_inputsattribute of your pipeline class. - max_sequence_length (
int, optional, defaults to512) — The maximum sequence length of the prompt. - overlap_history (
int, optional, defaults toNone) — Number of frames to overlap for smooth transitions in long videos. IfNone, the pipeline assumes short video generation mode, and no overlap is applied. 17 and 37 are recommended to set. - addnoise_condition (
float, optional, defaults to0) — This is used to help smooth the long video generation by adding some noise to the clean condition. Too large noise can cause the inconsistency as well. 20 is a recommended value, and you may try larger ones, but it is recommended to not exceed 50. - base_num_frames (
int, optional, defaults to97) — 97 or 121 | Base frame count (97 for 540P, 121 for 720P) - ar_step (
int, optional, defaults to0) — Controls asynchronous inference (0 for synchronous mode) You can setar_step=5to enable asynchronous inference. When asynchronous inference,causal_block_size=5is recommended while it is not supposed to be set for synchronous generation. Asynchronous inference will take more steps to diffuse the whole sequence which means it will be SLOWER than synchronous mode. In our experiments, asynchronous inference may improve the instruction following and visual consistent performance. - causal_block_size (
int, optional, defaults toNone) — The number of frames in each block/chunk. Recommended when using asynchronous inference (when ar_step > 0) - fps (
int, optional, defaults to24) — Frame rate of the generated video
Returns
~SkyReelsV2PipelineOutput or tuple
If return_dict is True, SkyReelsV2PipelineOutput is returned, otherwise a tuple is returned
where the first element is a list with the generated images and the second element is a list of bools
indicating whether the corresponding generated image contains “not-safe-for-work” (nsfw) content.
The call function to the pipeline for generation.
Examples:
>>> import torch
>>> from diffusers import (
... SkyReelsV2DiffusionForcingPipeline,
... UniPCMultistepScheduler,
... AutoencoderKLWan,
... )
>>> from diffusers.utils import export_to_video
>>> # Load the pipeline
>>> # Available models:
>>> # - Skywork/SkyReels-V2-DF-1.3B-540P-Diffusers
>>> # - Skywork/SkyReels-V2-DF-14B-540P-Diffusers
>>> # - Skywork/SkyReels-V2-DF-14B-720P-Diffusers
>>> vae = AutoencoderKLWan.from_pretrained(
... "Skywork/SkyReels-V2-DF-14B-720P-Diffusers",
... subfolder="vae",
... torch_dtype=torch.float32,
... )
>>> pipe = SkyReelsV2DiffusionForcingPipeline.from_pretrained(
... "Skywork/SkyReels-V2-DF-14B-720P-Diffusers",
... vae=vae,
... torch_dtype=torch.bfloat16,
... )
>>> flow_shift = 8.0 # 8.0 for T2V, 5.0 for I2V
>>> pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config, flow_shift=flow_shift)
>>> pipe = pipe.to("cuda")
>>> prompt = "A cat and a dog baking a cake together in a kitchen. The cat is carefully measuring flour, while the dog is stirring the batter with a wooden spoon. The kitchen is cozy, with sunlight streaming through the window."
>>> output = pipe(
... prompt=prompt,
... num_inference_steps=30,
... height=544,
... width=960,
... guidance_scale=6.0, # 6.0 for T2V, 5.0 for I2V
... num_frames=97,
... ar_step=5, # Controls asynchronous inference (0 for synchronous mode)
... causal_block_size=5, # Number of frames processed together in a causal block
... overlap_history=None, # Number of frames to overlap for smooth transitions in long videos
... addnoise_condition=20, # Improves consistency in long video generation
... ).frames[0]
>>> export_to_video(output, "video.mp4", fps=24, quality=8)encode_prompt
< source >( prompt: typing.Union[str, typing.List[str]] negative_prompt: typing.Union[str, typing.List[str], NoneType] = None do_classifier_free_guidance: bool = True num_videos_per_prompt: int = 1 prompt_embeds: typing.Optional[torch.Tensor] = None negative_prompt_embeds: typing.Optional[torch.Tensor] = None max_sequence_length: int = 226 device: typing.Optional[torch.device] = None dtype: typing.Optional[torch.dtype] = None )
Parameters
- prompt (
strorList[str], optional) — prompt to be encoded - negative_prompt (
strorList[str], optional) — The prompt or prompts not to guide the image generation. If not defined, one has to passnegative_prompt_embedsinstead. Ignored when not using guidance (i.e., ignored ifguidance_scaleis less than1). - do_classifier_free_guidance (
bool, optional, defaults toTrue) — Whether to use classifier free guidance or not. - num_videos_per_prompt (
int, optional, defaults to 1) — Number of videos that should be generated per prompt. torch device to place the resulting embeddings on - prompt_embeds (
torch.Tensor, 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 frompromptinput argument. - negative_prompt_embeds (
torch.Tensor, optional) — Pre-generated negative text embeddings. Can be used to easily tweak text inputs, e.g. prompt weighting. If not provided, negative_prompt_embeds will be generated fromnegative_promptinput argument. - device — (
torch.device, optional): torch device - dtype — (
torch.dtype, optional): torch dtype
Encodes the prompt into text encoder hidden states.
generate_timestep_matrix
< source >( num_latent_frames: int step_template: Tensor base_num_latent_frames: int ar_step: int = 5 num_pre_ready: int = 0 causal_block_size: int = 1 shrink_interval_with_mask: bool = False ) → tuple containing
Parameters
- num_latent_frames (int) — Total number of latent frames to generate
- step_template (torch.Tensor) — Base timestep schedule (e.g., [1000, 800, 600, …, 0])
- base_num_latent_frames (int) — Maximum frames the model can process in one forward pass
- ar_step (int, optional) — Autoregressive step size for temporal lag. 0 = synchronous, >0 = asynchronous. Defaults to 5.
- num_pre_ready (int, optional) — Number of frames already denoised (e.g., from prefix in a video2video task). Defaults to 0.
- causal_block_size (int, optional) — Number of frames processed as a causal block. Defaults to 1.
- shrink_interval_with_mask (bool, optional) — Whether to optimize processing intervals. Defaults to False.
Returns
tuple containing
- step_matrix (torch.Tensor): Matrix of timesteps for each frame at each iteration Shape: [num_iterations, num_latent_frames]
- step_index (torch.Tensor): Index matrix for timestep lookup Shape: [num_iterations, num_latent_frames]
- step_update_mask (torch.Tensor): Boolean mask indicating which frames to update Shape: [num_iterations, num_latent_frames]
- valid_interval (list[tuple]): List of (start, end) intervals for each iteration
Raises
ValueError
ValueError— If ar_step is too small for the given configuration
This function implements the core diffusion forcing algorithm that creates a coordinated denoising schedule across temporal frames. It supports both synchronous and asynchronous generation modes:
Synchronous Mode (ar_step=0, causal_block_size=1):
- All frames are denoised simultaneously at each timestep
- Each frame follows the same denoising trajectory: [1000, 800, 600, …, 0]
- Simpler but may have less temporal consistency for long videos
Asynchronous Mode (ar_step>0, causal_block_size>1):
- Frames are grouped into causal blocks and processed block/chunk-wise
- Each block is denoised in a staggered pattern creating a “denoising wave”
- Earlier blocks are more denoised, later blocks lag behind by ar_step timesteps
- Creates stronger temporal dependencies and better consistency
SkyReelsV2DiffusionForcingImageToVideoPipeline
class diffusers.SkyReelsV2DiffusionForcingImageToVideoPipeline
< source >( tokenizer: AutoTokenizer text_encoder: UMT5EncoderModel transformer: SkyReelsV2Transformer3DModel vae: AutoencoderKLWan scheduler: UniPCMultistepScheduler )
Parameters
- tokenizer (
AutoTokenizer) — Tokenizer from T5, specifically the google/umt5-xxl variant. - text_encoder (
UMT5EncoderModel) — T5, specifically the google/umt5-xxl variant. - transformer (SkyReelsV2Transformer3DModel) — Conditional Transformer to denoise the encoded image latents.
- scheduler (UniPCMultistepScheduler) —
A scheduler to be used in combination with
transformerto denoise the encoded image latents. - vae (AutoencoderKLWan) — Variational Auto-Encoder (VAE) Model to encode and decode videos to and from latent representations.
Pipeline for Image-to-Video (i2v) generation using SkyReels-V2 with diffusion forcing.
This model inherits from DiffusionPipeline. Check the superclass documentation for the generic methods implemented for all pipelines (downloading, saving, running on a specific device, etc.).
__call__
< source >( image: typing.Union[PIL.Image.Image, numpy.ndarray, torch.Tensor, typing.List[PIL.Image.Image], typing.List[numpy.ndarray], typing.List[torch.Tensor]] prompt: typing.Union[str, typing.List[str]] = None negative_prompt: typing.Union[str, typing.List[str]] = None height: int = 544 width: int = 960 num_frames: int = 97 num_inference_steps: int = 50 guidance_scale: float = 5.0 num_videos_per_prompt: typing.Optional[int] = 1 generator: typing.Union[torch._C.Generator, typing.List[torch._C.Generator], NoneType] = None latents: typing.Optional[torch.Tensor] = None prompt_embeds: typing.Optional[torch.Tensor] = None negative_prompt_embeds: typing.Optional[torch.Tensor] = None image_embeds: typing.Optional[torch.Tensor] = None last_image: typing.Optional[torch.Tensor] = None output_type: typing.Optional[str] = 'np' return_dict: bool = True attention_kwargs: typing.Optional[typing.Dict[str, typing.Any]] = None callback_on_step_end: typing.Union[typing.Callable[[int, int, typing.Dict], NoneType], diffusers.callbacks.PipelineCallback, diffusers.callbacks.MultiPipelineCallbacks, NoneType] = None callback_on_step_end_tensor_inputs: typing.List[str] = ['latents'] max_sequence_length: int = 512 overlap_history: typing.Optional[int] = None addnoise_condition: float = 0 base_num_frames: int = 97 ar_step: int = 0 causal_block_size: typing.Optional[int] = None fps: int = 24 ) → ~SkyReelsV2PipelineOutput or tuple
Parameters
- image (
PipelineImageInput) — The input image to condition the generation on. Must be an image, a list of images or atorch.Tensor. - prompt (
strorList[str], optional) — The prompt or prompts to guide the image generation. If not defined, one has to passprompt_embeds. instead. - negative_prompt (
strorList[str], optional) — The prompt or prompts not to guide the image generation. If not defined, one has to passnegative_prompt_embedsinstead. Ignored when not using guidance (i.e., ignored ifguidance_scaleis less than1). - height (
int, defaults to544) — The height of the generated video. - width (
int, defaults to960) — The width of the generated video. - num_frames (
int, defaults to97) — The number of frames in the generated video. - num_inference_steps (
int, defaults to50) — The number of denoising steps. More denoising steps usually lead to a higher quality image at the expense of slower inference. - guidance_scale (
float, defaults to5.0) — Guidance scale as defined in Classifier-Free Diffusion Guidance.guidance_scaleis defined aswof equation 2. of Imagen Paper. Guidance scale is enabled by settingguidance_scale > 1. Higher guidance scale encourages to generate images that are closely linked to the textprompt, usually at the expense of lower image quality. (6.0 for T2V, 5.0 for I2V) - num_videos_per_prompt (
int, optional, defaults to 1) — The number of images to generate per prompt. - generator (
torch.GeneratororList[torch.Generator], optional) — Atorch.Generatorto make generation deterministic. - latents (
torch.Tensor, optional) — Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image generation. Can be used to tweak the same generation with different prompts. If not provided, a latents tensor is generated by sampling using the supplied randomgenerator. - prompt_embeds (
torch.Tensor, optional) — Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not provided, text embeddings are generated from thepromptinput argument. - negative_prompt_embeds (
torch.Tensor, optional) — Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not provided, text embeddings are generated from thenegative_promptinput argument. - image_embeds (
torch.Tensor, optional) — Pre-generated image embeddings. Can be used to easily tweak image inputs (weighting). If not provided, image embeddings are generated from theimageinput argument. - last_image (
torch.Tensor, optional) — Pre-generated image embeddings. Can be used to easily tweak image inputs (weighting). If not provided, image embeddings are generated from theimageinput argument. - output_type (
str, optional, defaults to"np") — The output format of the generated image. Choose betweenPIL.Imageornp.array. - return_dict (
bool, optional, defaults toTrue) — Whether or not to return aSkyReelsV2PipelineOutputinstead of a plain tuple. - attention_kwargs (
dict, optional) — A kwargs dictionary that if specified is passed along to theAttentionProcessoras defined underself.processorin diffusers.models.attention_processor. - callback_on_step_end (
Callable,PipelineCallback,MultiPipelineCallbacks, optional) — A function or a subclass ofPipelineCallbackorMultiPipelineCallbacksthat is called at the end of each denoising step during the inference. with the following arguments:callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict).callback_kwargswill include a list of all tensors as specified bycallback_on_step_end_tensor_inputs. - callback_on_step_end_tensor_inputs (
List, optional) — The list of tensor inputs for thecallback_on_step_endfunction. The tensors specified in the list will be passed ascallback_kwargsargument. You will only be able to include variables listed in the._callback_tensor_inputsattribute of your pipeline class. - max_sequence_length (
int, optional, defaults to512) — The maximum sequence length of the prompt. - overlap_history (
int, optional, defaults toNone) — Number of frames to overlap for smooth transitions in long videos. IfNone, the pipeline assumes short video generation mode, and no overlap is applied. 17 and 37 are recommended to set. - addnoise_condition (
float, optional, defaults to0) — This is used to help smooth the long video generation by adding some noise to the clean condition. Too large noise can cause the inconsistency as well. 20 is a recommended value, and you may try larger ones, but it is recommended to not exceed 50. - base_num_frames (
int, optional, defaults to97) — 97 or 121 | Base frame count (97 for 540P, 121 for 720P) - ar_step (
int, optional, defaults to0) — Controls asynchronous inference (0 for synchronous mode) You can setar_step=5to enable asynchronous inference. When asynchronous inference,causal_block_size=5is recommended while it is not supposed to be set for synchronous generation. Asynchronous inference will take more steps to diffuse the whole sequence which means it will be SLOWER than synchronous mode. In our experiments, asynchronous inference may improve the instruction following and visual consistent performance. - causal_block_size (
int, optional, defaults toNone) — The number of frames in each block/chunk. Recommended when using asynchronous inference (when ar_step > 0) - fps (
int, optional, defaults to24) — Frame rate of the generated video
Returns
~SkyReelsV2PipelineOutput or tuple
If return_dict is True, SkyReelsV2PipelineOutput is returned, otherwise a tuple is returned
where the first element is a list with the generated images and the second element is a list of bools
indicating whether the corresponding generated image contains “not-safe-for-work” (nsfw) content.
The call function to the pipeline for generation.
Examples:
>>> import torch
>>> from diffusers import (
... SkyReelsV2DiffusionForcingImageToVideoPipeline,
... UniPCMultistepScheduler,
... AutoencoderKLWan,
... )
>>> from diffusers.utils import export_to_video
>>> from PIL import Image
>>> # Load the pipeline
>>> # Available models:
>>> # - Skywork/SkyReels-V2-DF-1.3B-540P-Diffusers
>>> # - Skywork/SkyReels-V2-DF-14B-540P-Diffusers
>>> # - Skywork/SkyReels-V2-DF-14B-720P-Diffusers
>>> vae = AutoencoderKLWan.from_pretrained(
... "Skywork/SkyReels-V2-DF-14B-720P-Diffusers",
... subfolder="vae",
... torch_dtype=torch.float32,
... )
>>> pipe = SkyReelsV2DiffusionForcingImageToVideoPipeline.from_pretrained(
... "Skywork/SkyReels-V2-DF-14B-720P-Diffusers",
... vae=vae,
... torch_dtype=torch.bfloat16,
... )
>>> flow_shift = 5.0 # 8.0 for T2V, 5.0 for I2V
>>> pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config, flow_shift=flow_shift)
>>> pipe = pipe.to("cuda")
>>> prompt = "A cat and a dog baking a cake together in a kitchen. The cat is carefully measuring flour, while the dog is stirring the batter with a wooden spoon. The kitchen is cozy, with sunlight streaming through the window."
>>> image = Image.open("path/to/image.png")
>>> output = pipe(
... image=image,
... prompt=prompt,
... num_inference_steps=50,
... height=544,
... width=960,
... guidance_scale=5.0, # 6.0 for T2V, 5.0 for I2V
... num_frames=97,
... ar_step=0, # Controls asynchronous inference (0 for synchronous mode)
... overlap_history=None, # Number of frames to overlap for smooth transitions in long videos
... addnoise_condition=20, # Improves consistency in long video generation
... ).frames[0]
>>> export_to_video(output, "video.mp4", fps=24, quality=8)encode_prompt
< source >( prompt: typing.Union[str, typing.List[str]] negative_prompt: typing.Union[str, typing.List[str], NoneType] = None do_classifier_free_guidance: bool = True num_videos_per_prompt: int = 1 prompt_embeds: typing.Optional[torch.Tensor] = None negative_prompt_embeds: typing.Optional[torch.Tensor] = None max_sequence_length: int = 226 device: typing.Optional[torch.device] = None dtype: typing.Optional[torch.dtype] = None )
Parameters
- prompt (
strorList[str], optional) — prompt to be encoded - negative_prompt (
strorList[str], optional) — The prompt or prompts not to guide the image generation. If not defined, one has to passnegative_prompt_embedsinstead. Ignored when not using guidance (i.e., ignored ifguidance_scaleis less than1). - do_classifier_free_guidance (
bool, optional, defaults toTrue) — Whether to use classifier free guidance or not. - num_videos_per_prompt (
int, optional, defaults to 1) — Number of videos that should be generated per prompt. torch device to place the resulting embeddings on - prompt_embeds (
torch.Tensor, 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 frompromptinput argument. - negative_prompt_embeds (
torch.Tensor, optional) — Pre-generated negative text embeddings. Can be used to easily tweak text inputs, e.g. prompt weighting. If not provided, negative_prompt_embeds will be generated fromnegative_promptinput argument. - device — (
torch.device, optional): torch device - dtype — (
torch.dtype, optional): torch dtype
Encodes the prompt into text encoder hidden states.
generate_timestep_matrix
< source >( num_latent_frames: int step_template: Tensor base_num_latent_frames: int ar_step: int = 5 num_pre_ready: int = 0 causal_block_size: int = 1 shrink_interval_with_mask: bool = False ) → tuple containing
Parameters
- num_latent_frames (int) — Total number of latent frames to generate
- step_template (torch.Tensor) — Base timestep schedule (e.g., [1000, 800, 600, …, 0])
- base_num_latent_frames (int) — Maximum frames the model can process in one forward pass
- ar_step (int, optional) — Autoregressive step size for temporal lag. 0 = synchronous, >0 = asynchronous. Defaults to 5.
- num_pre_ready (int, optional) — Number of frames already denoised (e.g., from prefix in a video2video task). Defaults to 0.
- causal_block_size (int, optional) — Number of frames processed as a causal block. Defaults to 1.
- shrink_interval_with_mask (bool, optional) — Whether to optimize processing intervals. Defaults to False.
Returns
tuple containing
- step_matrix (torch.Tensor): Matrix of timesteps for each frame at each iteration Shape: [num_iterations, num_latent_frames]
- step_index (torch.Tensor): Index matrix for timestep lookup Shape: [num_iterations, num_latent_frames]
- step_update_mask (torch.Tensor): Boolean mask indicating which frames to update Shape: [num_iterations, num_latent_frames]
- valid_interval (list[tuple]): List of (start, end) intervals for each iteration
Raises
ValueError
ValueError— If ar_step is too small for the given configuration
This function implements the core diffusion forcing algorithm that creates a coordinated denoising schedule across temporal frames. It supports both synchronous and asynchronous generation modes:
Synchronous Mode (ar_step=0, causal_block_size=1):
- All frames are denoised simultaneously at each timestep
- Each frame follows the same denoising trajectory: [1000, 800, 600, …, 0]
- Simpler but may have less temporal consistency for long videos
Asynchronous Mode (ar_step>0, causal_block_size>1):
- Frames are grouped into causal blocks and processed block/chunk-wise
- Each block is denoised in a staggered pattern creating a “denoising wave”
- Earlier blocks are more denoised, later blocks lag behind by ar_step timesteps
- Creates stronger temporal dependencies and better consistency
SkyReelsV2DiffusionForcingVideoToVideoPipeline
class diffusers.SkyReelsV2DiffusionForcingVideoToVideoPipeline
< source >( tokenizer: AutoTokenizer text_encoder: UMT5EncoderModel transformer: SkyReelsV2Transformer3DModel vae: AutoencoderKLWan scheduler: UniPCMultistepScheduler )
Parameters
- tokenizer (
AutoTokenizer) — Tokenizer from T5, specifically the google/umt5-xxl variant. - text_encoder (
UMT5EncoderModel) — T5, specifically the google/umt5-xxl variant. - transformer (SkyReelsV2Transformer3DModel) — Conditional Transformer to denoise the encoded image latents.
- scheduler (UniPCMultistepScheduler) —
A scheduler to be used in combination with
transformerto denoise the encoded image latents. - vae (AutoencoderKLWan) — Variational Auto-Encoder (VAE) Model to encode and decode videos to and from latent representations.
Pipeline for Video-to-Video (v2v) generation using SkyReels-V2 with diffusion forcing.
This model inherits from DiffusionPipeline. Check the superclass documentation for the generic methods implemented for all pipelines (downloading, saving, running on a specific device, etc.).
__call__
< source >( video: typing.List[PIL.Image.Image] prompt: typing.Union[str, typing.List[str]] = None negative_prompt: typing.Union[str, typing.List[str]] = None height: int = 544 width: int = 960 num_frames: int = 120 num_inference_steps: int = 50 guidance_scale: float = 6.0 num_videos_per_prompt: typing.Optional[int] = 1 generator: typing.Union[torch._C.Generator, typing.List[torch._C.Generator], NoneType] = None latents: typing.Optional[torch.Tensor] = None prompt_embeds: typing.Optional[torch.Tensor] = None negative_prompt_embeds: typing.Optional[torch.Tensor] = None output_type: typing.Optional[str] = 'np' return_dict: bool = True attention_kwargs: typing.Optional[typing.Dict[str, typing.Any]] = None callback_on_step_end: typing.Union[typing.Callable[[int, int, typing.Dict], NoneType], diffusers.callbacks.PipelineCallback, diffusers.callbacks.MultiPipelineCallbacks, NoneType] = None callback_on_step_end_tensor_inputs: typing.List[str] = ['latents'] max_sequence_length: int = 512 overlap_history: typing.Optional[int] = None addnoise_condition: float = 0 base_num_frames: int = 97 ar_step: int = 0 causal_block_size: typing.Optional[int] = None fps: int = 24 ) → ~SkyReelsV2PipelineOutput or tuple
Parameters
- video (
List[Image.Image]) — The video to guide the video generation. - prompt (
strorList[str], optional) — The prompt or prompts to guide the video generation. If not defined, one has to passprompt_embeds. instead. - negative_prompt (
strorList[str], optional) — The prompt or prompts not to guide the video generation. If not defined, one has to passnegative_prompt_embedsinstead. Ignored when not using guidance (i.e., ignored ifguidance_scaleis less than1). - height (
int, defaults to544) — The height of the generated video. - width (
int, defaults to960) — The width of the generated video. - num_frames (
int, defaults to120) — The number of frames in the generated video. - num_inference_steps (
int, defaults to50) — The number of denoising steps. More denoising steps usually lead to a higher quality image at the expense of slower inference. - guidance_scale (
float, defaults to6.0) — Guidance scale as defined in Classifier-Free Diffusion Guidance.guidance_scaleis defined aswof equation 2. of Imagen Paper. Guidance scale is enabled by settingguidance_scale > 1. Higher guidance scale encourages to generate images that are closely linked to the textprompt, usually at the expense of lower image quality. (6.0 for T2V, 5.0 for I2V) - num_videos_per_prompt (
int, optional, defaults to 1) — The number of images to generate per prompt. - generator (
torch.GeneratororList[torch.Generator], optional) — Atorch.Generatorto make generation deterministic. - latents (
torch.Tensor, optional) — Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image generation. Can be used to tweak the same generation with different prompts. If not provided, a latents tensor is generated by sampling using the supplied randomgenerator. - prompt_embeds (
torch.Tensor, optional) — Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not provided, text embeddings are generated from thepromptinput argument. - negative_prompt_embeds (
torch.Tensor, optional) — Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not provided, text embeddings are generated from thenegative_promptinput argument. - output_type (
str, optional, defaults to"np") — The output format of the generated image. Choose betweenPIL.Imageornp.array. - return_dict (
bool, optional, defaults toTrue) — Whether or not to return aSkyReelsV2PipelineOutputinstead of a plain tuple. - attention_kwargs (
dict, optional) — A kwargs dictionary that if specified is passed along to theAttentionProcessoras defined underself.processorin diffusers.models.attention_processor. - callback_on_step_end (
Callable,PipelineCallback,MultiPipelineCallbacks, optional) — A function or a subclass ofPipelineCallbackorMultiPipelineCallbacksthat is called at the end of each denoising step during the inference. with the following arguments:callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict).callback_kwargswill include a list of all tensors as specified bycallback_on_step_end_tensor_inputs. - callback_on_step_end_tensor_inputs (
List, optional) — The list of tensor inputs for thecallback_on_step_endfunction. The tensors specified in the list will be passed ascallback_kwargsargument. You will only be able to include variables listed in the._callback_tensor_inputsattribute of your pipeline class. - max_sequence_length (
int, optional, defaults to512) — The maximum sequence length of the prompt. - overlap_history (
int, optional, defaults toNone) — Number of frames to overlap for smooth transitions in long videos. IfNone, the pipeline assumes short video generation mode, and no overlap is applied. 17 and 37 are recommended to set. - addnoise_condition (
float, optional, defaults to0) — This is used to help smooth the long video generation by adding some noise to the clean condition. Too large noise can cause the inconsistency as well. 20 is a recommended value, and you may try larger ones, but it is recommended to not exceed 50. - base_num_frames (
int, optional, defaults to97) — 97 or 121 | Base frame count (97 for 540P, 121 for 720P) - ar_step (
int, optional, defaults to0) — Controls asynchronous inference (0 for synchronous mode) You can setar_step=5to enable asynchronous inference. When asynchronous inference,causal_block_size=5is recommended while it is not supposed to be set for synchronous generation. Asynchronous inference will take more steps to diffuse the whole sequence which means it will be SLOWER than synchronous mode. In our experiments, asynchronous inference may improve the instruction following and visual consistent performance. - causal_block_size (
int, optional, defaults toNone) — The number of frames in each block/chunk. Recommended when using asynchronous inference (when ar_step > 0) - fps (
int, optional, defaults to24) — Frame rate of the generated video
Returns
~SkyReelsV2PipelineOutput or tuple
If return_dict is True, SkyReelsV2PipelineOutput is returned, otherwise a tuple is returned
where the first element is a list with the generated images and the second element is a list of bools
indicating whether the corresponding generated image contains “not-safe-for-work” (nsfw) content.
The call function to the pipeline for generation.
Examples:
>>> import torch
>>> from diffusers import (
... SkyReelsV2DiffusionForcingVideoToVideoPipeline,
... UniPCMultistepScheduler,
... AutoencoderKLWan,
... )
>>> from diffusers.utils import export_to_video
>>> # Load the pipeline
>>> # Available models:
>>> # - Skywork/SkyReels-V2-DF-1.3B-540P-Diffusers
>>> # - Skywork/SkyReels-V2-DF-14B-540P-Diffusers
>>> # - Skywork/SkyReels-V2-DF-14B-720P-Diffusers
>>> vae = AutoencoderKLWan.from_pretrained(
... "Skywork/SkyReels-V2-DF-14B-720P-Diffusers",
... subfolder="vae",
... torch_dtype=torch.float32,
... )
>>> pipe = SkyReelsV2DiffusionForcingVideoToVideoPipeline.from_pretrained(
... "Skywork/SkyReels-V2-DF-14B-720P-Diffusers",
... vae=vae,
... torch_dtype=torch.bfloat16,
... )
>>> flow_shift = 8.0 # 8.0 for T2V, 5.0 for I2V
>>> pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config, flow_shift=flow_shift)
>>> pipe = pipe.to("cuda")
>>> prompt = "A cat and a dog baking a cake together in a kitchen. The cat is carefully measuring flour, while the dog is stirring the batter with a wooden spoon. The kitchen is cozy, with sunlight streaming through the window."
>>> output = pipe(
... prompt=prompt,
... num_inference_steps=50,
... height=544,
... width=960,
... guidance_scale=6.0, # 6.0 for T2V, 5.0 for I2V
... num_frames=97,
... ar_step=0, # Controls asynchronous inference (0 for synchronous mode)
... overlap_history=None, # Number of frames to overlap for smooth transitions in long videos
... addnoise_condition=20, # Improves consistency in long video generation
... ).frames[0]
>>> export_to_video(output, "video.mp4", fps=24, quality=8)encode_prompt
< source >( prompt: typing.Union[str, typing.List[str]] negative_prompt: typing.Union[str, typing.List[str], NoneType] = None do_classifier_free_guidance: bool = True num_videos_per_prompt: int = 1 prompt_embeds: typing.Optional[torch.Tensor] = None negative_prompt_embeds: typing.Optional[torch.Tensor] = None max_sequence_length: int = 226 device: typing.Optional[torch.device] = None dtype: typing.Optional[torch.dtype] = None )
Parameters
- prompt (
strorList[str], optional) — prompt to be encoded - negative_prompt (
strorList[str], optional) — The prompt or prompts not to guide the image generation. If not defined, one has to passnegative_prompt_embedsinstead. Ignored when not using guidance (i.e., ignored ifguidance_scaleis less than1). - do_classifier_free_guidance (
bool, optional, defaults toTrue) — Whether to use classifier free guidance or not. - num_videos_per_prompt (
int, optional, defaults to 1) — Number of videos that should be generated per prompt. torch device to place the resulting embeddings on - prompt_embeds (
torch.Tensor, 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 frompromptinput argument. - negative_prompt_embeds (
torch.Tensor, optional) — Pre-generated negative text embeddings. Can be used to easily tweak text inputs, e.g. prompt weighting. If not provided, negative_prompt_embeds will be generated fromnegative_promptinput argument. - device — (
torch.device, optional): torch device - dtype — (
torch.dtype, optional): torch dtype
Encodes the prompt into text encoder hidden states.
generate_timestep_matrix
< source >( num_latent_frames: int step_template: Tensor base_num_latent_frames: int ar_step: int = 5 num_pre_ready: int = 0 causal_block_size: int = 1 shrink_interval_with_mask: bool = False ) → tuple containing
Parameters
- num_latent_frames (int) — Total number of latent frames to generate
- step_template (torch.Tensor) — Base timestep schedule (e.g., [1000, 800, 600, …, 0])
- base_num_latent_frames (int) — Maximum frames the model can process in one forward pass
- ar_step (int, optional) — Autoregressive step size for temporal lag. 0 = synchronous, >0 = asynchronous. Defaults to 5.
- num_pre_ready (int, optional) — Number of frames already denoised (e.g., from prefix in a video2video task). Defaults to 0.
- causal_block_size (int, optional) — Number of frames processed as a causal block. Defaults to 1.
- shrink_interval_with_mask (bool, optional) — Whether to optimize processing intervals. Defaults to False.
Returns
tuple containing
- step_matrix (torch.Tensor): Matrix of timesteps for each frame at each iteration Shape: [num_iterations, num_latent_frames]
- step_index (torch.Tensor): Index matrix for timestep lookup Shape: [num_iterations, num_latent_frames]
- step_update_mask (torch.Tensor): Boolean mask indicating which frames to update Shape: [num_iterations, num_latent_frames]
- valid_interval (list[tuple]): List of (start, end) intervals for each iteration
Raises
ValueError
ValueError— If ar_step is too small for the given configuration
This function implements the core diffusion forcing algorithm that creates a coordinated denoising schedule across temporal frames. It supports both synchronous and asynchronous generation modes:
Synchronous Mode (ar_step=0, causal_block_size=1):
- All frames are denoised simultaneously at each timestep
- Each frame follows the same denoising trajectory: [1000, 800, 600, …, 0]
- Simpler but may have less temporal consistency for long videos
Asynchronous Mode (ar_step>0, causal_block_size>1):
- Frames are grouped into causal blocks and processed block/chunk-wise
- Each block is denoised in a staggered pattern creating a “denoising wave”
- Earlier blocks are more denoised, later blocks lag behind by ar_step timesteps
- Creates stronger temporal dependencies and better consistency
SkyReelsV2Pipeline
class diffusers.SkyReelsV2Pipeline
< source >( tokenizer: AutoTokenizer text_encoder: UMT5EncoderModel transformer: SkyReelsV2Transformer3DModel vae: AutoencoderKLWan scheduler: UniPCMultistepScheduler )
Parameters
- tokenizer (
T5Tokenizer) — Tokenizer from T5, specifically the google/umt5-xxl variant. - text_encoder (
T5EncoderModel) — T5, specifically the google/umt5-xxl variant. - transformer (SkyReelsV2Transformer3DModel) — Conditional Transformer to denoise the input latents.
- scheduler (UniPCMultistepScheduler) —
A scheduler to be used in combination with
transformerto denoise the encoded image latents. - vae (AutoencoderKLWan) — Variational Auto-Encoder (VAE) Model to encode and decode videos to and from latent representations.
Pipeline for Text-to-Video (t2v) generation using SkyReels-V2.
This model inherits from DiffusionPipeline. Check the superclass documentation for the generic methods implemented for all pipelines (downloading, saving, running on a particular device, etc.).
__call__
< source >( prompt: typing.Union[str, typing.List[str]] = None negative_prompt: typing.Union[str, typing.List[str]] = None height: int = 544 width: int = 960 num_frames: int = 97 num_inference_steps: int = 50 guidance_scale: float = 6.0 num_videos_per_prompt: typing.Optional[int] = 1 generator: typing.Union[torch._C.Generator, typing.List[torch._C.Generator], NoneType] = None latents: typing.Optional[torch.Tensor] = None prompt_embeds: typing.Optional[torch.Tensor] = None negative_prompt_embeds: typing.Optional[torch.Tensor] = None output_type: typing.Optional[str] = 'np' return_dict: bool = True attention_kwargs: typing.Optional[typing.Dict[str, typing.Any]] = None callback_on_step_end: typing.Union[typing.Callable[[int, int, typing.Dict], NoneType], diffusers.callbacks.PipelineCallback, diffusers.callbacks.MultiPipelineCallbacks, NoneType] = None callback_on_step_end_tensor_inputs: typing.List[str] = ['latents'] max_sequence_length: int = 512 ) → ~SkyReelsV2PipelineOutput or tuple
Parameters
- prompt (
strorList[str], optional) — The prompt or prompts to guide the image generation. If not defined, one has to passprompt_embeds. instead. - height (
int, defaults to544) — The height in pixels of the generated image. - width (
int, defaults to960) — The width in pixels of the generated image. - num_frames (
int, defaults to97) — The number of frames in the generated video. - num_inference_steps (
int, defaults to50) — The number of denoising steps. More denoising steps usually lead to a higher quality image at the expense of slower inference. - guidance_scale (
float, defaults to6.0) — Guidance scale as defined in Classifier-Free Diffusion Guidance.guidance_scaleis defined aswof equation 2. of Imagen Paper. Guidance scale is enabled by settingguidance_scale > 1. Higher guidance scale encourages to generate images that are closely linked to the textprompt, usually at the expense of lower image quality. - num_videos_per_prompt (
int, optional, defaults to 1) — The number of images to generate per prompt. - generator (
torch.GeneratororList[torch.Generator], optional) — Atorch.Generatorto make generation deterministic. - latents (
torch.Tensor, optional) — Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image generation. Can be used to tweak the same generation with different prompts. If not provided, a latents tensor is generated by sampling using the supplied randomgenerator. - prompt_embeds (
torch.Tensor, optional) — Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not provided, text embeddings are generated from thepromptinput argument. - output_type (
str, optional, defaults to"np") — The output format of the generated image. Choose betweenPIL.Imageornp.array. - return_dict (
bool, optional, defaults toTrue) — Whether or not to return aSkyReelsV2PipelineOutputinstead of a plain tuple. - attention_kwargs (
dict, optional) — A kwargs dictionary that if specified is passed along to theAttentionProcessoras defined underself.processorin diffusers.models.attention_processor. - callback_on_step_end (
Callable,PipelineCallback,MultiPipelineCallbacks, optional) — A function or a subclass ofPipelineCallbackorMultiPipelineCallbacksthat is called at the end of each denoising step during the inference. with the following arguments:callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict).callback_kwargswill include a list of all tensors as specified bycallback_on_step_end_tensor_inputs. - callback_on_step_end_tensor_inputs (
List, optional) — The list of tensor inputs for thecallback_on_step_endfunction. The tensors specified in the list will be passed ascallback_kwargsargument. You will only be able to include variables listed in the._callback_tensor_inputsattribute of your pipeline class. - max_sequence_length (
int, optional, defaults to512) — The maximum sequence length for the text encoder.
Returns
~SkyReelsV2PipelineOutput or tuple
If return_dict is True, SkyReelsV2PipelineOutput is returned, otherwise a tuple is returned
where the first element is a list with the generated images and the second element is a list of bools
indicating whether the corresponding generated image contains “not-safe-for-work” (nsfw) content.
The call function to the pipeline for generation.
Examples:
>>> import torch
>>> from diffusers import (
... SkyReelsV2Pipeline,
... UniPCMultistepScheduler,
... AutoencoderKLWan,
... )
>>> from diffusers.utils import export_to_video
>>> # Load the pipeline
>>> # Available models:
>>> # - Skywork/SkyReels-V2-T2V-14B-540P-Diffusers
>>> # - Skywork/SkyReels-V2-T2V-14B-720P-Diffusers
>>> vae = AutoencoderKLWan.from_pretrained(
... "Skywork/SkyReels-V2-T2V-14B-720P-Diffusers",
... subfolder="vae",
... torch_dtype=torch.float32,
... )
>>> pipe = SkyReelsV2Pipeline.from_pretrained(
... "Skywork/SkyReels-V2-T2V-14B-720P-Diffusers",
... vae=vae,
... torch_dtype=torch.bfloat16,
... )
>>> flow_shift = 8.0 # 8.0 for T2V, 5.0 for I2V
>>> pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config, flow_shift=flow_shift)
>>> pipe = pipe.to("cuda")
>>> prompt = "A cat and a dog baking a cake together in a kitchen. The cat is carefully measuring flour, while the dog is stirring the batter with a wooden spoon. The kitchen is cozy, with sunlight streaming through the window."
>>> output = pipe(
... prompt=prompt,
... num_inference_steps=50,
... height=544,
... width=960,
... guidance_scale=6.0, # 6.0 for T2V, 5.0 for I2V
... num_frames=97,
... ).frames[0]
>>> export_to_video(output, "video.mp4", fps=24, quality=8)encode_prompt
< source >( prompt: typing.Union[str, typing.List[str]] negative_prompt: typing.Union[str, typing.List[str], NoneType] = None do_classifier_free_guidance: bool = True num_videos_per_prompt: int = 1 prompt_embeds: typing.Optional[torch.Tensor] = None negative_prompt_embeds: typing.Optional[torch.Tensor] = None max_sequence_length: int = 226 device: typing.Optional[torch.device] = None dtype: typing.Optional[torch.dtype] = None )
Parameters
- prompt (
strorList[str], optional) — prompt to be encoded - negative_prompt (
strorList[str], optional) — The prompt or prompts not to guide the image generation. If not defined, one has to passnegative_prompt_embedsinstead. Ignored when not using guidance (i.e., ignored ifguidance_scaleis less than1). - do_classifier_free_guidance (
bool, optional, defaults toTrue) — Whether to use classifier free guidance or not. - num_videos_per_prompt (
int, optional, defaults to 1) — Number of videos that should be generated per prompt. torch device to place the resulting embeddings on - prompt_embeds (
torch.Tensor, 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 frompromptinput argument. - negative_prompt_embeds (
torch.Tensor, optional) — Pre-generated negative text embeddings. Can be used to easily tweak text inputs, e.g. prompt weighting. If not provided, negative_prompt_embeds will be generated fromnegative_promptinput argument. - device — (
torch.device, optional): torch device - dtype — (
torch.dtype, optional): torch dtype
Encodes the prompt into text encoder hidden states.
SkyReelsV2ImageToVideoPipeline
class diffusers.SkyReelsV2ImageToVideoPipeline
< source >( tokenizer: AutoTokenizer text_encoder: UMT5EncoderModel image_encoder: CLIPVisionModelWithProjection image_processor: CLIPProcessor transformer: SkyReelsV2Transformer3DModel vae: AutoencoderKLWan scheduler: UniPCMultistepScheduler )
Parameters
- tokenizer (
T5Tokenizer) — Tokenizer from T5, specifically the google/umt5-xxl variant. - text_encoder (
T5EncoderModel) — T5, specifically the google/umt5-xxl variant. - image_encoder (
CLIPVisionModelWithProjection) — CLIP, specifically the clip-vit-huge-patch14 variant. - transformer (SkyReelsV2Transformer3DModel) — Conditional Transformer to denoise the input latents.
- scheduler (UniPCMultistepScheduler) —
A scheduler to be used in combination with
transformerto denoise the encoded image latents. - vae (AutoencoderKLWan) — Variational Auto-Encoder (VAE) Model to encode and decode videos to and from latent representations.
Pipeline for Image-to-Video (i2v) generation using SkyReels-V2.
This model inherits from DiffusionPipeline. Check the superclass documentation for the generic methods implemented for all pipelines (downloading, saving, running on a particular device, etc.).
__call__
< source >( image: typing.Union[PIL.Image.Image, numpy.ndarray, torch.Tensor, typing.List[PIL.Image.Image], typing.List[numpy.ndarray], typing.List[torch.Tensor]] prompt: typing.Union[str, typing.List[str]] = None negative_prompt: typing.Union[str, typing.List[str]] = None height: int = 544 width: int = 960 num_frames: int = 97 num_inference_steps: int = 50 guidance_scale: float = 5.0 num_videos_per_prompt: typing.Optional[int] = 1 generator: typing.Union[torch._C.Generator, typing.List[torch._C.Generator], NoneType] = None latents: typing.Optional[torch.Tensor] = None prompt_embeds: typing.Optional[torch.Tensor] = None negative_prompt_embeds: typing.Optional[torch.Tensor] = None image_embeds: typing.Optional[torch.Tensor] = None last_image: typing.Optional[torch.Tensor] = None output_type: typing.Optional[str] = 'np' return_dict: bool = True attention_kwargs: typing.Optional[typing.Dict[str, typing.Any]] = None callback_on_step_end: typing.Union[typing.Callable[[int, int, typing.Dict], NoneType], diffusers.callbacks.PipelineCallback, diffusers.callbacks.MultiPipelineCallbacks, NoneType] = None callback_on_step_end_tensor_inputs: typing.List[str] = ['latents'] max_sequence_length: int = 512 ) → ~SkyReelsV2PipelineOutput or tuple
Parameters
- image (
PipelineImageInput) — The input image to condition the generation on. Must be an image, a list of images or atorch.Tensor. - prompt (
strorList[str], optional) — The prompt or prompts to guide the image generation. If not defined, one has to passprompt_embeds. instead. - negative_prompt (
strorList[str], optional) — The prompt or prompts not to guide the image generation. If not defined, one has to passnegative_prompt_embedsinstead. Ignored when not using guidance (i.e., ignored ifguidance_scaleis less than1). - height (
int, defaults to544) — The height of the generated video. - width (
int, defaults to960) — The width of the generated video. - num_frames (
int, defaults to97) — The number of frames in the generated video. - num_inference_steps (
int, defaults to50) — The number of denoising steps. More denoising steps usually lead to a higher quality image at the expense of slower inference. - guidance_scale (
float, defaults to5.0) — Guidance scale as defined in Classifier-Free Diffusion Guidance.guidance_scaleis defined aswof equation 2. of Imagen Paper. Guidance scale is enabled by settingguidance_scale > 1. Higher guidance scale encourages to generate images that are closely linked to the textprompt, usually at the expense of lower image quality. - num_videos_per_prompt (
int, optional, defaults to 1) — The number of images to generate per prompt. - generator (
torch.GeneratororList[torch.Generator], optional) — Atorch.Generatorto make generation deterministic. - latents (
torch.Tensor, optional) — Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image generation. Can be used to tweak the same generation with different prompts. If not provided, a latents tensor is generated by sampling using the supplied randomgenerator. - prompt_embeds (
torch.Tensor, optional) — Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not provided, text embeddings are generated from thepromptinput argument. - negative_prompt_embeds (
torch.Tensor, optional) — Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not provided, text embeddings are generated from thenegative_promptinput argument. - image_embeds (
torch.Tensor, optional) — Pre-generated image embeddings. Can be used to easily tweak image inputs (weighting). If not provided, image embeddings are generated from theimageinput argument. - output_type (
str, optional, defaults to"np") — The output format of the generated image. Choose betweenPIL.Imageornp.array. - return_dict (
bool, optional, defaults toTrue) — Whether or not to return aWanPipelineOutputinstead of a plain tuple. - attention_kwargs (
dict, optional) — A kwargs dictionary that if specified is passed along to theAttentionProcessoras defined underself.processorin diffusers.models.attention_processor. - callback_on_step_end (
Callable,PipelineCallback,MultiPipelineCallbacks, optional) — A function or a subclass ofPipelineCallbackorMultiPipelineCallbacksthat is called at the end of each denoising step during the inference. with the following arguments:callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict).callback_kwargswill include a list of all tensors as specified bycallback_on_step_end_tensor_inputs. - callback_on_step_end_tensor_inputs (
List, optional) — The list of tensor inputs for thecallback_on_step_endfunction. The tensors specified in the list will be passed ascallback_kwargsargument. You will only be able to include variables listed in the._callback_tensor_inputsattribute of your pipeline class. - max_sequence_length (
int, optional, defaults to512) — The maximum sequence length of the prompt.
Returns
~SkyReelsV2PipelineOutput or tuple
If return_dict is True, SkyReelsV2PipelineOutput is returned, otherwise a tuple is returned
where the first element is a list with the generated images and the second element is a list of bools
indicating whether the corresponding generated image contains “not-safe-for-work” (nsfw) content.
The call function to the pipeline for generation.
Examples:
>>> import torch
>>> from diffusers import (
... SkyReelsV2ImageToVideoPipeline,
... UniPCMultistepScheduler,
... AutoencoderKLWan,
... )
>>> from diffusers.utils import export_to_video
>>> from PIL import Image
>>> # Load the pipeline
>>> # Available models:
>>> # - Skywork/SkyReels-V2-I2V-1.3B-540P-Diffusers
>>> # - Skywork/SkyReels-V2-I2V-14B-540P-Diffusers
>>> # - Skywork/SkyReels-V2-I2V-14B-720P-Diffusers
>>> vae = AutoencoderKLWan.from_pretrained(
... "Skywork/SkyReels-V2-I2V-14B-720P-Diffusers",
... subfolder="vae",
... torch_dtype=torch.float32,
... )
>>> pipe = SkyReelsV2ImageToVideoPipeline.from_pretrained(
... "Skywork/SkyReels-V2-I2V-14B-720P-Diffusers",
... vae=vae,
... torch_dtype=torch.bfloat16,
... )
>>> flow_shift = 5.0 # 8.0 for T2V, 5.0 for I2V
>>> pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config, flow_shift=flow_shift)
>>> pipe = pipe.to("cuda")
>>> prompt = "A cat and a dog baking a cake together in a kitchen. The cat is carefully measuring flour, while the dog is stirring the batter with a wooden spoon. The kitchen is cozy, with sunlight streaming through the window."
>>> image = Image.open("path/to/image.png")
>>> output = pipe(
... image=image,
... prompt=prompt,
... num_inference_steps=50,
... height=544,
... width=960,
... guidance_scale=5.0, # 6.0 for T2V, 5.0 for I2V
... num_frames=97,
... ).frames[0]
>>> export_to_video(output, "video.mp4", fps=24, quality=8)encode_prompt
< source >( prompt: typing.Union[str, typing.List[str]] negative_prompt: typing.Union[str, typing.List[str], NoneType] = None do_classifier_free_guidance: bool = True num_videos_per_prompt: int = 1 prompt_embeds: typing.Optional[torch.Tensor] = None negative_prompt_embeds: typing.Optional[torch.Tensor] = None max_sequence_length: int = 226 device: typing.Optional[torch.device] = None dtype: typing.Optional[torch.dtype] = None )
Parameters
- prompt (
strorList[str], optional) — prompt to be encoded - negative_prompt (
strorList[str], optional) — The prompt or prompts not to guide the image generation. If not defined, one has to passnegative_prompt_embedsinstead. Ignored when not using guidance (i.e., ignored ifguidance_scaleis less than1). - do_classifier_free_guidance (
bool, optional, defaults toTrue) — Whether to use classifier free guidance or not. - num_videos_per_prompt (
int, optional, defaults to 1) — Number of videos that should be generated per prompt. torch device to place the resulting embeddings on - prompt_embeds (
torch.Tensor, 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 frompromptinput argument. - negative_prompt_embeds (
torch.Tensor, optional) — Pre-generated negative text embeddings. Can be used to easily tweak text inputs, e.g. prompt weighting. If not provided, negative_prompt_embeds will be generated fromnegative_promptinput argument. - device — (
torch.device, optional): torch device - dtype — (
torch.dtype, optional): torch dtype
Encodes the prompt into text encoder hidden states.
SkyReelsV2PipelineOutput
class diffusers.pipelines.skyreels_v2.pipeline_output.SkyReelsV2PipelineOutput
< source >( frames: Tensor )
Parameters
- frames (
torch.Tensor,np.ndarray, or List[List[PIL.Image.Image]]) — List of video outputs - It can be a nested list of lengthbatch_size,with each sub-list containing denoised PIL image sequences of lengthnum_frames.It can also be a NumPy array or Torch tensor of shape(batch_size, num_frames, channels, height, width).
Output class for SkyReelsV2 pipelines.