The photo
- One image: JPEG, PNG, WEBP, HEIC, BMP, TIFF or GIF, under 30 MB.
- Each side between 300 and 6000 pixels, width-to-height ratio between 0.4 and 2.5.
- A clear, front-facing face works best. Busy group photos make it guess who talks.
UlazAI - AI Image & Video Tools
New talking photo model
OmniHuman 2.0: one photo, one voice file, a talking video
Upload a portrait, add a voice recording, and OmniHuman 2.0 makes that person say it. It costs 14 credits per second, where 1.5 costs 27, and it takes audio up to five minutes long instead of one.
14 credits per second of voice audio. The price follows the length of your audio file, which the Studio measures before it starts.
Clips up to 5 minutes from one photo. 1.5 stops at 60 seconds.
A short prompt now steers motion, scene and camera next to the voice.
A 7-second clip cost us 98 credits on 2.0 and 189 on 1.5.
Same photo, same voice, two models
OmniHuman 1.5 against 2.0
We gave both models the same street portrait and the same 7-second voice clip, with the same prompt. Press play to start both at once.
Sound comes from the right-hand video.
- 2.0 kept her eyes on the camera for the whole clip, as the prompt asked. 1.5 opened with her head turned away.
- 1.5 is sharper: its 1080p mode gave us 1248×1664 pixels against 864×1072 for 2.0.
- 2.0 cost 98 credits for these 7 seconds, 1.5 cost 189.
Prompt used for both
She talks warmly to the camera, small natural head movements, city street behind her.
What you upload
Two files and an optional line of direction. The voice decides the length of the video and how the mouth moves; the photo decides who is speaking.
The voice file
- WAV or MP3, under 15 MB, between 4 seconds and 5 minutes.
- The Studio measures the file before it starts and charges for that exact length.
- Clean speech without a music bed gives the steadiest mouth movement.
The prompt
- Optional, up to 300 characters.
- Use it for posture, expression, hand movement or a slow camera push.
- Write it in English, Chinese, Japanese, Korean, Spanish or Indonesian.
Where 2.0 changes the job
| Voice length | OmniHuman 2.0 (720p) | OmniHuman 1.5 (1080p) |
|---|---|---|
| 7s | 98 credits | 189 credits |
| 15s | 210 credits | 405 credits |
| 30s | 420 credits | 810 credits |
| 60s | 840 credits | - |
| 5 min | 4200 credits | - |
Jobs that fit a talking photo
OmniHuman is for videos where the person is the message. If you need a new scene built around them, use a text-to-video model instead.
A founder or expert explaining something
One good portrait and a recorded script. With five minutes of audio you can now do a full product walkthrough in one take.
The same face in several languages
Record or generate the script in each language and reuse the photo. The face stays the same, the mouth follows each track.
Course and training clips
A fixed presenter for a series of short lessons, without booking a studio for every update.
Characters and mascots
Illustrated or 3D characters work too, as long as the face is clearly visible.
Studio routes
Photo + voice to talking video
omnihuman_2_0
4 seconds to 5 minutes, set by the audio file
One portrait image, one WAV or MP3 voice file, and an optional prompt of up to 300 characters.
Open routeIs 1.5 still worth using?
Only if you need 1080p output. 1.5 renders 1080p, 2.0 renders 720p. For everything else 2.0 is cheaper per second, accepts five times longer audio and takes a prompt for motion and camera.
A first clip in four steps
- Pick a sharp portrait where the face is clearly visible.
- Record the script as WAV or MP3, or make it with the voiceover tool. Start with 10 to 20 seconds.
- Add one line of direction, such as "calm, small nods, slow push in".
- Check the price the Studio shows for your audio length, then generate.
FAQ
What does OmniHuman 2.0 cost?
14 credits per second of audio. A 30-second clip costs 420 credits. OmniHuman 1.5 costs 27 credits per second.
How long can the video be?
As long as the voice file: from 4 seconds up to 5 minutes.
What resolution do I get?
720p. If you need 1080p, OmniHuman 1.5 is still available.
Do I need a prompt?
No. The photo and the voice are enough. A prompt helps when you want specific motion or a camera move.
Is there a watermark?
No. UlazAI turns the provider watermark off for every OmniHuman 2.0 render.
Can I use a voice I generated with AI?
Yes. Any clean WAV or MP3 works, including files from the UlazAI voiceover tool.