InfiniteTalk drives talking-head lip-sync from an audio track — either animating a still directly, or layering lip-sync on top of an existing motion clip. It's talking-head motion specifically, not full-body performance.
What it needs: ~12 GB VRAM, GGUF-quantized, plus an audio track.
What it's good for: making a character talk at $0 per shot, entirely locally — no hosted lip-sync API, no per-second billing.
In FlixML: `infinitetalk_i2v` (image-to-video lip-sync) and `infinitetalk_v2v` (lip-sync applied over a driving motion clip).