Upload a portrait photo and an audio track (up to 35s), and OmniHuman 1.5 generates a talking avatar video with synced lips and natural motion.
A single portrait plus a script becomes a lifelike spokesperson — synced lips, expressions, and gestures. No camera, no studio.
Hey everyone! I am so excited to share my new favorite foundation. It blends in seamlessly for a flawless, natural finish, and it actually feels lightweight all day. A little goes a long way, so one bottle lasts for months.
A complete AI avatar (digital human) generator: upload a portrait and an audio clip — or type a script and pick a voice — and get a lifelike talking video with synced lips, natural expressions, and gestures. Powered by ByteDance OmniHuman 1.5, no camera, actor, or studio required.
No filming, no actor, no green screen. Upload one clear photo of a person and turn it into a presenter that speaks your script. Create a reusable digital host for product explainers, course intros, announcements, and ads — and regenerate new lines anytime without reshooting.
Already have a voice recording? Upload it and the avatar lip-syncs to it. No audio? Switch to text mode — type what the avatar should say, choose a voice from the built-in text-to-speech library, and the script is synthesized into the driving audio for you. The estimated length and credit cost are shown before you generate.
OmniHuman 1.5 reads the rhythm, prosody, and meaning of the audio to drive head movement, posture, and hand gestures alongside accurate lip-sync, while preserving the person’s identity and lighting. The result performs like a real presenter rather than a static portrait with an animated mouth.
Spin up spokesperson clips for ads and landing pages, narrated lessons and onboarding for training, talking-head posts for social, or multilingual versions of the same message. Every generation is saved to history with its inputs, so you can iterate and produce variations fast.
Presenter-style portraits generated on CreateVision AI — any one of them can drive a talking avatar video once you add audio.
OmniHuman 1.5 drives the lip sync, expression, and body motion on this page.
| Input | One portrait photo plus either an uploaded audio file or a typed script. |
|---|---|
| Portrait guidance | Front-facing, well lit, whole face visible and in focus. Sunglasses and heavy shadow reduce lip-sync accuracy. |
| Audio length | Up to 35 seconds per generation — roughly 85 spoken words. |
| Voice library | 60 built-in voices across 10 languages, with previews available before you generate. |
| Model | OmniHuman 1.5, which drives lip sync, facial expression, and natural head and body motion. |
| Cost | Billed per second of output like other video generation; shown before you generate. |
| Output | A downloadable clip, saved to your history with the source portrait and script attached. |
One portrait, one voice track, one clip.
A clear, front-facing photo works best. The AI avatar video generator reads the face from this single frame.
Upload an audio file, or type a script and pick from the built-in voice library. Audio can run up to 35 seconds.
The model syncs lips, expression, and natural head motion to the audio, and returns a clip you can download or extend.
There is no image prompt here — the portrait and the audio do the work. What you write is the script, and script choices decide whether the result looks natural.
Start with something short. The first second is where lip sync is most visible.
Hey — quick one for you.
Keep sentences under about 15 words. Long clauses produce unnatural pauses.
This serum has two ingredients. That is the whole formula.
Commas and full stops become breaths. Use them where you would actually pause.
It works — and it takes ten seconds.
Roughly 2.5 words per second. Write to the clip length you want.
~85 words for a 35-second clip
Our revolutionary product utilises cutting-edge technology to deliver unparalleled results for discerning customers seeking excellence.
This is the part people always ask about. Two ingredients, no filler. You will see the difference in about a week.
The first is one 19-word sentence of abstractions — the avatar delivers it in a flat, breathless monotone because there is nowhere to pause. The second is three short sentences with natural stress points, which is what makes the delivery read as human.
Hello and welcome to our channel where today we will be discussing the five most important things you need to know about skincare routines.
Five things about skincare. Number one is the one people get wrong.
Long run-on openings are where lip sync drifts most, because the model has no punctuation to anchor to. A short hook gives it clean stress points and holds attention better.
Talking avatars are one output. Image and video generation run from the same account.
Where avatar generation sits among the wider video engines.
Producing the portrait you will hand to the avatar generator.
What per-second video pricing looks like across engines.
Generating a source portrait from scratch if you do not have one.
The engine to use for the b-roll around your avatar clip.
Avatar videos bill per second like other video output. Free accounts get 80 credits a day to try it.
Perfect for getting started
Billed annually · $240/year
Billed annually · $380/year
An AI avatar is a generated talking video of a person built from a single photo and a voice. Instead of filming, you provide a portrait and audio (or a script + voice), and the model animates a lifelike speaking performance — used as a digital presenter, spokesperson, or narrator.
A clear, front-facing portrait photo, plus either an audio clip (mp3/wav) or — in text mode — the script you want spoken and a voice picked from the built-in library. That is enough to generate a talking avatar video.
Yes. Upload your own audio recording and the avatar lip-syncs to it. If you do not have a recording, switch to text mode and choose a voice from the text-to-speech library instead.
Up to 35 seconds of audio per generation. For longer scripts, split them into multiple clips. The estimated duration is shown as you type, and you are billed by the second of audio.
Common uses include product explainers and ads, course and onboarding narration, social talking-head clips, announcements, and multilingual versions of a message — anywhere you would otherwise need to film a presenter.
CreateVision AI uses ByteDance’s OmniHuman 1.5, which drives lip-sync plus head motion and gestures from the audio. See the OmniHuman 1.5 model page for the technical details.
A well-lit, front-facing portrait where the whole face is visible and in focus. Sunglasses, heavy shadow across the face, extreme angles, or a very small face in frame all reduce lip-sync accuracy.
No. You can type a script and pick a voice from the built-in library, which covers 60 voices across 10 languages. Uploading your own recording is supported if you would rather use it.
Up to 35 seconds of audio per generation. For longer presentations, generate several segments and join them — keeping the same source portrait keeps the presenter consistent across clips.
