Two-hand finger-frame gesture becomes a window into an AI character-swapped version of the same footage — generated entirely locally, no API key, no cloud billing.
Local ComfyUI port of the gesture-tracking pipeline from sophiamyang's finger-frame-effect-ai. Where the original sends your clip to Gemini per generation, this workflow runs on your own GPU with Wan2.1 SCAIL-2: it generates a cropped region covering everywhere the window travels (not the whole frame), pastes it back with head-drift auto-align, and reveals it through the tracked quad with rim-only refraction.
Requires: ComfyUI-WanVideoWrapper, mediapipe (auto-installed), Wan2.1 14B SCAIL-2 checkpoint + LoRAs (see workflow notes). Tested on RTX 4090 24GB.
What it's NOT good at: multiple gesture sessions in one clip, non-continuous takes.
==
Usage notes: Pick a head image and a body image of the character you want, load your own driving video, and optionally describe the appearance/outfit in the prompt. Auto-resolution in the pipeline works with both horizontal and vertical video. No length limit on the clip — tested with 30+ second videos. Just feed the input at a format/resolution ComfyUI can handle. That's it — the pipeline is fully automated, nothing else to configure. works with anime and real characters. Requires the ComfyUI-FingerWindow-Local custom node: https://github.com/stark622/ComfyUI-FingerWindow-Local-SCAIL2 (see README for install — copy the ComfyUI-FingerWindow-Local subfolder into your custom_nodes)
-tools\converter.bat if you need convert iphone or another type video to comfyui standart in one click