ComfyUI-WanVideoWrapper

Change Wan's CFG per step, not per run

Schedule CFG or LoRA strength per step

The 3D pose estimator behind Wan Animate

The audio encoder that makes lip-sync possible

Check what your face-ID pipeline is actually seeing

Fill a masked region with subject-matched noise

Render NLF pose predictions into a control image

A fake model just to get the right sigmas

Grab the first N frames to seed the next clip

Turn pose keypoints into a face mask

Pull expression embeds from a driving face for Wan

Load the expression-driver for multi-face Wan animation

Load the audio-driven talking-portrait model for Wan

Drive a Wan talking head from an audio clip

Audio-driven talking-head video on Wan

Render facial landmarks into a keypoints image

Load the Lynx identity adapter for Wan

Load the NLF 3D pose model for Wan pose control

Load the motion tokenizer that drives MTVCrafter animation

The misnamed node that actually loads CLIP vision

The umT5 loader for Kijai's Wan wrapper

Turn audio into a talking-avatar drive signal for Wan

Single-photo face identity for Wan video

Crop the face Lynx needs from your reference

Swap a character into a video with MoCha and Wan

Turn detected poses into motion tokens for Wan

The model behind audio-driven talking video

Placeholder silent embeds for MultiTalk frames

Driving Wan lip-sync from audio (one voice or several)

3D human pose from video, for pose-driven Wan control

Even out your audio before it drives Wan

Loading the audio VAE for Ovi's video-plus-sound generation

Load the Qwen LLM that powers Wan prompt extension

See your camera path before you render it

Encode a text-plus-image prompt through Qwen-VL

Drag-a-point motion control for native Wan

Subject-consistent Wan video from reference images

Add control-video conditioning to Wan-Fun

Combine dense and sparse control on Wan

Inject extra latents into Wan image embeds

Drive facial expression on a Wan clip

Feed low-res frames into Wan's FlashVSR upscaler

Wiring Decart's video editor into your Wan graph

Lock a face into a Wan video with Lynx

Inject MTVCrafter motion into your Wan generation

Sliding-window extension for longer Wan clips

Alignment-free pose transfer for Wan characters

Hold one reference image across the whole clip

Fusing Ovi's audio and video latents into one clip

The multiplier node for kijai's Pusa LoRA

Audio- and pose-driven conditioning for Wan S2V

Layer pose control into a SCAIL motion-transfer job

Point SCAIL-2 motion at your character

Lock a face into your Wan clip from one photo

Pose-driven Wan animation from pose latents

Giving Wan a memory across shots

Time-to-Move motion control, drag-an-object animation for Wan

Drawing motion instead of describing it

Drive a character with pose, face, and reference

Get your negative prompt back at CFG 1 on Wan

Trajectory-based motion control, model-side

Draw motion paths and make Wan follow them

See your trajectories before you render

Name the transformer blocks for swap or radial attention

How to fit Wan 14B on a 12-16GB card

Turning your reference image into Wan I2V conditioning

Merge two sets of Wan image embeds into one

Sliding windows for longer-than-81-frame clips

Scheduling latent control over part of the clip

Depth and pose control for Wan generations

An actual ControlNet, not VACE or a LoRA

Latents to frames, with tiling for when VRAM runs out

Turning generated audio latents into sound

The sampler built for chaining clips

Skip redundant Wan steps — but not with speed LoRAs

The text-to-video starting point in a WanVideoWrapper graph

The blank audio canvas for Ovi's audio+video generation

Turn images into Wan latents for I2V and V2V

Turn a stack of images into per-frame Wan latents

Feeding existing audio into an Ovi generation

A free-ish quality bump for Wan motion

The experimental sampling grab-bag for Wan

Stacking VACE or MTV Crafter onto your base model

The fast upscale VAE for Wan

FreeInit for Wan, and when the 3x cost is worth it

Camera-trajectory control for Wan-Fun

The old I2V encode and what replaced it

Snap your I2V input to a resolution Wan actually likes

The I2V starting point, and how to add an end frame

The windowed engine for talking-head video

Windowed long-form audio-driven video

The fix when mixing native and wrapper VAE breaks your output

Seamless looping Wan clips via latent shift

Turn a LoRA off in specific transformer blocks

Stacking LoRAs (and speed LoRAs) into a Wan graph

How you actually load a LoRA in this pack

Stack Wan LoRAs, and where speed LoRAs belong

A third caching option, after TeaCache got deprecated

Fast object and watermark removal from video

The front door to Kijai's Wan wrapper

Separate audio guidance for Wan's audio-video Ovi mode

Grab already-decoded frames from Wan samples

Drop specific subjects into a Wan video

Crack open an embeds bundle so you can see what you actually built

Rewrite short prompts into what Wan actually wants

Let an LLM rewrite your terse prompt into what Wan wants

SMPL + hand-aware pose control for dance

Turn camera poses into conditioning the sampler can actually use

Preset camera moves for ReCamMaster

Dial in an orbit shot by degrees instead of picking a preset

Pick Wan's RoPE implementation (and save VRAM)

The node that actually generates the video

The one node that bundles every Wan sampler add-on

The tidy face of Wan's monster sampler

Bundle your sampler config to reuse it

The engine of a WanVideoWrapper graph

Build the Wan noise schedule outside the sampler

The same Wan scheduler, one clean connection

Switch attention backends mid-generation

Apply block-swap args to a model you already loaded

Apply Wan LoRAs live in the linear layers, no merge

Sparse attention that pays off on long Wan clips

Turn a sigma threshold into a step number

Skip Layer Guidance for Wan, the nearly-free quality bump

The anchor node behind minute-long Wan clips

Skip redundant Wan steps for a real speedup

Use native ComfyUI conditioning inside WanVideoWrapper

Where your prompt goes in a Wan-wrapper graph

Encode your Wan prompt, then get the T5 out of VRAM

Encode exactly one prompt when the bundled encoder gives you too much

Load a tiny VAE for fast Wan previews

Free speed, if you can get Triton working

Extrapolation tuning for Wan, and a feature with basically no track record yet

Camera and 3D control for Wan

Camera-move control for Wan video

Skeletons for driving Wan character motion

Pose-drive a Wan character with UniAnimate

Pack a subject and a new backdrop for UniLumos-style relighting

Where your control video and reference images actually become VACE conditioning

Attach a separate VACE module to your Wan model

Build the frame batch and masks VACE needs

The small file that decodes your Wan latents

The aggressive offload option for tight VRAM

Visualize the motion tracks driving a Wan-Move generation

Load the audio brain for Wan talking-head video

Load Whisper for audio-driven Wan workflows