Why AI video models do not lip sync to a voiceover, and what to use

A text-to-video or image-to-video model does not lip sync to a voiceover you add afterward, and Sume's models docs say so directly: video models do not lip-sync to generated TTS or to a later voice-over, so a talking face is never a video-model clip with narration laid underneath. For speech, use a path built for it: Avatar video from a script, or the lip-sync endpoint that takes a still and an audio clip.

This is from the Models overview, read 2026-09-29.

Which path fits which job?

What about video models that generate audio?

Several catalog models can generate audio with generate_audio, and some accept audio references. That is native sound made with the video. It is not the same as syncing lips to a specific recording you supply, and the docs do not promise lip sync from it. Check each model's generate_audio and supported_input_references at GET /v1/videos/models.

How do I put speech under B-roll then?

Cut B-roll for the parts where no one is on camera and let the avatar or lip-sync clip carry the on-camera lines. Join them with a Timeline render, using the speech as the audio spine. See talking head vs B-roll.

Where do I read the details?

Lip sync from a photo and audio covers the still-plus-audio call. Avatar, lip sync or motion control helps you pick one.

How do I check what a video model accepts?

GET /v1/videos/models lists each model with these fields.