InfiniteTalk — lip-sync and talking video from image

★ 6.5 · content

infinitetalk is a Claude Code skill that converts audio into synchronized talking-head video, supporting both Image-to-Video (from a single photo) and Video-to-Video (re-dubbing an existing clip) modes with precise lip, head, body, and facial expression alignment. The skill relies on Wan2.1-I2V-14B-480P and InfiniteTalk model weights (~30 GB total); the main script `scripts/infer_infinitetalk.py` accepts resolution options (480P/720P), a `clip` or `streaming` mode for unlimited-length output, and an audio guidance scale parameter. A built-in TTS engine powered by Kokoro-82M lets you supply raw text instead of a pre-recorded audio file, with two-speaker dialogue support via `(s1)`/`(s2)` markers. A 16 GB+ VRAM GPU is recommended, though int8 quantization and model offloading are available for lower-end hardware. The skill is aimed at creators building digital avatars, virtual presenters, or any lip-sync video generation workflow.