tongflow-modal-infinitetalk
June 17, 2026 · View on GitHub
Official TongFlow plugin. Audio-driven lip-sync with InfiniteTalk (MeiGen-AI/InfiniteTalk, on Wan-AI/Wan2.1-I2V-14B-480P with a wav2vec audio encoder), running on a GPU via Modal. Turns an audio track plus a video (or image) into a talking-head video.
Capabilities
- Lip sync (
audio-video-lip-sync) — drive a subject's lips from an audio track (audio + video → video).
Credentials
Add in TongFlow Settings (gear icon, top-right):
| Key | Required | Notes |
|---|---|---|
MODAL_TOKEN_ID | ✅ | Create at modal.com/settings/tokens. |
MODAL_TOKEN_SECRET | ✅ | Paired with MODAL_TOKEN_ID. |
HF_TOKEN | ✅ | Required to fetch the Wan2.1 / InfiniteTalk weights from Hugging Face. |
Weights (Hugging Face)
The plugin injects HF_TOKEN from your TongFlow Settings into the Modal download job at deploy time — no manual modal secret create needed. Without it the weight download fails.