OmniHuman: a talking photo that moves its whole body
ByteDance's OmniHuman turns one still picture and a voice into a speaking clip. The head, shoulders and hands move with the words, not just the lips.
Real results
Made in solorender with these models. Nothing retouched.
Best for
- ✓Expressive presenters: head, shoulders and hands move along with the speech
- ✓Characters that aren't people. Our T-rex in a dressing gown talks with his jaw
- ✓Scenes with more going on than a face. Props stay in hand and other people in the shot keep moving
- ✓A clip from a typed script. Pick a voice in the Talk lane and solorender makes the voiceover for you first
Where it struggles
- —The voice can be up to 30 seconds. For a longer speech, split it into parts and join the clips in the Video editor
- —One size only. VEED Fabric offers 480p or 720p
- —The clip is as long as the voice, so a one-line voice gives a clip of a few seconds
Prompts that work
Each one opens the studio with the prompt already in the box. They are also one click away inside the app, under the prompt field.
Interview, in character
Try it →The T-rex in the dressing gown speaks to an interviewer just off camera, weary and offended, mouth opening and closing with each word, head tilting with diva disdain. The makeup artist keeps gently dabbing powder on his snout. Subtle handheld documentary camera.
Say how they feel and where they look. Mention anyone else in the shot so they keep moving too.
Quiet, emotional line
Try it →An elderly grandmother on a sofa, deeply moved, eyes wet behind her red glasses, turning her head gently toward her grandson beside her and speaking softly to him with wonder, small gentle movements, holding her teacup.
"Small gentle movements" keeps a soft line soft. Name the prop so it stays in her hand.
What you give it
A picture and a voice. Open the Talk lane, pick OmniHuman, and add the picture under Face. Then either pick a voiceover from your Library, or type a script and choose who reads it.
The prompt is optional. Leave it empty and OmniHuman still speaks the words. Add one when you want a mood or a gesture: "leans back smugly", "small embarrassed shrug", "head tilting with diva disdain".
OmniHuman or VEED Fabric?
Both make a photo talk. VEED Fabric is quicker, often done in about 20 seconds, and costs a little less a second at 480p. OmniHuman moves the head, hands and body with the speech, which reads as a performance rather than a talking head.
See both on the same kind of job in our talking photo comparison.
Questions
How long can an OmniHuman clip be?
As long as the voice, up to 30 seconds.
How is OmniHuman priced?
By the second of voice, measured from the file itself. The price is at the top of this page, read live from our price list.
Does it work on cartoons and animals?
Our T-rex example above came out well. Try one short line first before you write a long script.
Run these for real
Every model in this guide is live in solorender. Credits never expire.
Open OmniHuman →