BytePlus launches OmniHuman 2.0 for five-minute talking videos from one photo

What happened

A single OmniHuman 2.0 take can run from 30 to 300 seconds. The previous limit was 60 seconds, and the longer clips are generated in one pass with no stitching.

The model takes three inputs: one reference image, an audio track (speech, music or dialogue) and an optional text prompt. Output is native 720p. Multi-person scenes are supported, and the text prompt can be used to direct gestures and camera.

The performance claims are BytePlus's own. Early reviewers flagged weak spots, including teeth in close-ups and over-acting.

Why it matters for creators

Longer audio-driven avatars expand what creators can do with lip sync for explainers, music clips and UGC-style ads. A five-minute single take covers formats that previously needed several clips joined together.

Prompt control over gestures and camera gives some direction over the performance, and multi-person support allows scenes with more than one speaker.

The weak spots reviewers noted still matter. Close-ups and expressive deliveries are worth checking carefully before a clip is published.

KreatorFlow offers OmniHuman lip sync.