OmniHuman 1.5

OmniHuman 1.5 turns one photo and one voice track into a presenter whose head, shoulders and hands move with the speech.

OmniHuman 1.5 is ByteDance’s talking-head model, and the pick when the result has to look like a person delivering a line rather than a still with an animated mouth. It drives head, shoulders and gesture from the audio, so pauses and emphasis read in the body too.

It takes a single image and a voice track: no reference video, no rig, no capture session. Give it a generated headshot and a voiceover from the Audio tab and you have a spokesperson who never existed.

At a glance

  • Type: Image + audio → video
  • Inputs: One still portrait, one audio track
  • Resolution: 720p · 1080p
  • Length: Follows the audio track
  • Pricing: Per second of output

What it is best at

  • Body and gesture motion as well as lip movement
  • Runs the full length of the track, with no fixed cap
  • A turbo mode for faster, cheaper drafts

What it costs

About 272 tokens ($2.72) per 10 s of audio; the exact price depends on the length of what you upload.

SettingsTokensPrice
10 s of audio272$2.72
30 s of audio787$7.87
60 s of audio1,556$15.56

Prices come from the live catalog at $0.0100 per token, the pay-as-you-go rate; subscriptions include tokens at a lower rate. The exact price is shown before you generate, and a failed generation is refunded.

Questions

What do I need to give OmniHuman 1.5?
A single still portrait and an audio track, both from your Assets. The photo can be a headshot, an illustration or something you generated on the Image tab.
How is OmniHuman 1.5 different from lip sync on a video?
Lip sync starts from footage of someone talking and swaps the voice. OmniHuman starts from one photo, so there is no footage to begin with and the performance is generated in full.
How long can OmniHuman 1.5 output be?
As long as the audio you give it. The model follows the track rather than capping at a fixed length, and the token price is quoted before you run.

Use it inside

Related models