ComfyUI-WanVideoWrapper

Fills the background (masked area) with Gaussian noise sampled using the mean and variance of the subject (unmasked) region.

Helper node to generate a list of floats that can be used to schedule cfg scale for the steps, outside the set range cfg is set to 1.0

Helper node to generate a list of floats that can be used to schedule things like cfg and lora scale per step

Draw face landmarks on an image for visualization/debugging

Helper node to create empty Wan model to use with BasicScheduler -node to get sigmas

Extracts the first N frames from a video sequence for continuations.

Input for MoCha model: https://github.com/Orange-3DV-Team/MoCha

Visualizes the camera poses, from Animatediff-Evolved CameraCtrl Pose or a .txt file with RealEstate camera intrinsics and coordinates, in a 3D plot.

Adds latent and timestep noise multipliers when using flowmatch_pusa

https://github.com/time-to-move/TTM

Adds NAG prompt embeds to original prompt embeds: 'https://github.com/ChenDarYen/Normalized-Attention-Guidance'

Comma separated list of blocks to apply block swap to, can also use ranges like '0-5' or '0,2,3-5' etc., can be connected to the dense_blocks input of 'WanVideoSetRadialAttention' node

Loads Wan clip_vision model from 'ComfyUI/models/clip_vision'

Context options for WanVideo, allows splitting the video into context windows and attemps blending them for longer generations than the model and memory otherwise would allow.

Loads ControlNet model from 'https://huggingface.co/collections/TheDenk/wan21-controlnets-68302b430411dafc0d74d2fc'

EasyCache for WanVideoWrapper, source https://github.com/H-EmbodVis/EasyCache

Encodes a batch of images individually to create a latent video batch where each video is a single frame, useful for I2V init purposes, for example as multiple context window inits

https://github.com/NUS-HPC-AI-Lab/Enhance-A-Video

Experimental stuff

Extra model to load and add to the main model, ie. VACE or MTV Crafter 'ComfyUI/models/diffusion_models'

Loads Wan VAE model from 'ComfyUI/models/vae'

https://github.com/TianxingWu/FreeInit; FreeInit, a concise yet effective method to improve temporal consistency of videos generated by diffusion models

Resizes image to the closest supported resolution based on aspect ratio and max pixels, according to the original code

Rescale latents to match the expected range for encoding or decoding between native ComfyUI VAE and the WanVideoWrapper VAE.

Enables Multi/InfiniteTalk long video generation sampling method, the video is created in windows with overlapping frames. Not compatible or necessary to be used with context windows and many other features besides Multi/InfiniteTalk.

Enables Multi/InfiniteTalk long video generation sampling method, the video is created in windows with overlapping frames. Not compatible or necessary to be used with context windows and many other features besides Multi/InfiniteTalk.

Looping through latent shift as shown in https://github.com/YisuiTT/Mobius/

Select a LoRA model from ComfyUI/models/loras

Select a LoRA model from ComfyUI/models/loras

MagCache for WanVideoWrapper, source https://github.com/Zehong-Ma/MagCache

Gets possible already decoded images from the samples dictionary, used with Multi/InfiniteTalk sampling

https://github.com/KwaiVGI/ReCamMaster

https://github.com/KwaiVGI/ReCamMaster

https://github.com/KwaiVGI/ReCamMaster

Utility node with no other functionality than to look cleaner, useful for the live preview as the main sampler node has become a messy monster

Node to output all settings and inputs for the WanVideoSamplerFromSettings -node

Override the attention mode for the model for specific step and/or block range

Sets the LoRA weights to be used directly in linear layers of the model, this does NOT merge LoRAs

Sets radial attention parameters, dense attention refers to normal attention

Simply passes a float value as an integer, used to set start/end steps with sigma threshold

Skips uncond on the selected blocks

Loads Wan text_encoder model from 'ComfyUI/models/LLM'

Patch WanVideo model to use TeaCache. Speeds up inference by caching the output and applying it instead of doing the step. Best results are achieved by choosing the appropriate coefficients for the model. Early steps should never be skipped, with too aggressive values this can happen and the motion suffers. Starting later can help with that too. When NOT using coefficients, the threshold value should be about 10 times smaller than the value used with coefficients. Official recommended values https://github.com/ali-vilab/TeaCache/tree/main/TeaCache4Wan2.1

Bridge between ComfyUI native text embedding and WanVideoWrapper text embedding

Encodes text prompts into text embeddings. For rudimentary prompt travel you can input multiple prompts separated by '|', they will be equally spread over the video length

Encodes text prompts into text embeddings. This node loads and completely unloads the T5 after done, leaving no VRAM or RAM imprint. If prompts have been cached before T5 is not loaded at all. negative output is meant to be used with NAG, it contains only negative prompt embeddings. Additionally you can provide a Qwen LLM model to extend the positive prompt with either one of the original Wan templates or a custom system prompt.

Encodes text prompt into text embedding.

Loads Wan VAE model from 'ComfyUI/models/vae_approx'

torch.compile settings, when connected to the model loader, torch.compile of the selected layers is attempted. Requires Triton and torch > 2.7.0 is recommended

Set UltraVico parameters, attention mode still needs to be set to sageattn_ultravico, https://github.com/thu-ml/DiT-Extrapolation

VACE model to use when not using model that has it included, loaded from 'ComfyUI/models/diffusion_models'

Helper node to create start/end frame batch and masks for VACE

Loads Wan VAE model from 'ComfyUI/models/vae'

Loads MMAudio VAE for Ovi audio generation

Adds Ovi negative text embeddings and audio CFG scale to the text embeddings dictionary