wanx-troopers
Wan-Move
Wan-Move support has been added to native ComfyUI. New nodes:
WanMoveTracksFromCoordsWanMoveTracksToVideo(note: mask is “for optional mask if you want parts of the track to not affect the output”)
If you want the points to start at any given frame, give the points a mask yes; it’s binary only; point is either visible or not; if no mask is provided when creating points, it’s all ones
To add a stationary point:
Splitting splines between generations to facilitate continuations: GetTracksRange node from KJNodes:
Idea: use SVI-film with it? The one with 5 frames overlap, Not SVI 2.0.
Works with Context Windows.
Summary
HF:Ruihang/Wan-Move-14B-480P features all-new and shiny Wan-Move model. The model provides abilities reminiscent of ATI and VACE. In essence the motion is taken from driving video and applied to a starting image. Driving motion can be supplied as a simple moving dot.
The model bears official Wan logo suggesting a link with the makers of the original AI video models. Based on Wan 2.1 I2V.
The model already has some support in kijai/ComfyUI-WanVideoWrapper
WanVideo Add WanMove Tracks,WanVideo Draw WanMove Tracks- model weights, 16Gb: HF:Kijai/WanVideo_comfy_fp8_scaled:WanMove
- example workflow
- GH:filliptm/ComfyUI_FL-Path-Animator can be used to build paths
T2V LoRA worked
WanMove is technically identical to Wan 2.1 I2V
As a LoRA
Gleb reported being able to make Wan Move work as LoRA:
He reported not being able to make it work with Wan 2.1 but working with 2.2