InfiniteTalk Fast Video-To-Video

infinitetalk-fast Video-to-Video

What is InfiniteTalk?

InfiniteTalk creates new videos by combining an input silent video and an audio track. It ensures precise lip synchronization while aligning head, face, and body movements with the audio. With optional masking and prompting, you can control which areas move and how the scene appears. The model also maintains visual identity for natural and consistent results.

Why it looks great

• Accurate lip synchronization: matches lip motion precisely to the audio.
• Full-body coherence: aligns head pose, facial expressions, and posture with speech.
• Mask control: optional mask images let you define which regions can move.
• Instruction following: prompts can guide style, pose, or behavior.
• Identity preservation: ensures consistent visual identity across all frames.

How to Use

1. Upload the audio file.
2. Upload a video as the base.
3. (Optional) Upload a mask image to control which regions can move.
4. (Optional) Write a prompt to guide the style, pose, or expressions.
5. Set the seed if you want reproducibility.
6. Submit the job and download the generated video.

Note

• Max clip length per job: 10 minutes
• Processing speed: ~10–30 seconds of wall time per 1 second of video (varies by resolution and queue load)
• Mask safety tip: Do not upload the full image as mask_image. The mask should only cover the regions you want to animate—otherwise the result may render as fully black.

More Versions

• Single-character version
• Multi-character version