- Home
- AI Video Generator
- OmniHuman 2.0
OmniHuman 2.0: Turn One Photo and Audio Into a 5-Minute Performance
OmniHuman 2.0 is ByteDance's audio-driven avatar video model. Give it one image and one voice or song track, and it generates the whole performance, with lips, expression, gestures, and body movement following the tone and meaning of the audio, in a single take of up to 5 minutes. Watch official examples, then use OmniHuman 2.0 on Nereo at 5 credits per second.
Talking Photo
OmniHuman 2.0 Features and Official Examples
Direct the Scene With a Text Prompt
Beyond the audio, a short prompt can ask for a specific action, a camera move, or a prop that is not in the photo. It is the difference between a talking head and a staged scene.
Official BytePlus example of prompt control: the character's actions and camera movement follow the text prompt while the voice drives the performance.
A Singer From a Single Image
Give OmniHuman 2.0 a sung track and the face carries the lyrics, held notes, and breaths, with movement that follows the rhythm. Keep the vocal clear and loud in the mix.
Official BytePlus example of a rhythmic singing performance generated from one image and a song.
Cinematic Emotion, Driven by the Words
The model reads the context of the audio, not only its sounds, so tension, anger, or relief show up in gestures and expression the way an actor would deliver the line.
Official BytePlus example of emotional, cinematic performance driven by the dialogue.
Animals and Stylized Characters
OmniHuman 2.0 is not limited to real people. Mascots, 3D characters, and animals can speak or sing when their eyes and mouth are clearly drawn.
Official BytePlus example of an animal character brought to life with audio.
Audio Comprehension
Expression changes with the delivery: the same face looks different when the line is sung softly, belted, or spoken with a smile.
Official BytePlus example showing the performance following the feeling of the audio.
Characters That Interact With the Scene
Prompts can describe how a character moves within the environment, so the shot feels like a scene rather than a portrait with a moving mouth.
Official BytePlus example of environmental interaction described in the prompt.
How to Use OmniHuman 2.0
The photo and the audio carry most of the result. A short prompt is optional.
Choose One Clear Image
Use a JPG, PNG, or WebP image up to 20 MB with one face looking toward the camera, a plain or softly blurred background, and a relaxed mouth. Frame it the way you want the video, and match vertical or wide to where it will play.
Prepare the Audio
Upload an MP3 or WAV voice, song, or dialogue track up to 15 MB and between 4 and 299 seconds long. Keep the voice clean and trim silence; every second of audio becomes a second of video.
Add a Short Direction and Generate
Optionally describe mood, gestures, or camera movement in one or two sentences. Test a short excerpt first, then run the full track.
What Creators Use OmniHuman 2.0 For
Every use case starts from one image and one recording.
Courses and Lessons
Run a full lesson in one take, so the instructor looks the same at minute four as at minute one and nothing needs cutting together.
Spokesperson and UGC Ads
Turn a presenter photo into product explainers and testimonial-style ads, then make new versions by swapping the voiceover.
Music Performances
Give a sung track to a portrait or character and get a performance that follows held notes, breaths, and rhythm.
Mascots and Characters
Let a brand mascot, a 3D character, or an animal deliver lines for ads, kids' content, and channel intros.
Interviews and Dialogue
Stage two-person conversations and sketches driven by one dialogue track, then check that each line lands on the right face.
Localized Messages
Pair the same spokesperson image with translated voice tracks to deliver one message in several languages.
OmniHuman 2.0 vs OmniHuman 1.5
| Feature | OmniHuman 2.0 | OmniHuman 1.5 |
|---|---|---|
| Inputs | 1 image + audio + optional prompt | 1 image + audio + optional prompt |
| Max audio / video length | 4–300 seconds | Under 60 seconds |
| Output resolution | Native 720p | 720p or 1080p |
| Audio formats | WAV, MP3 (under 15 MB) | MP3, WAV, M4A on Nereo |
| Multi-person scenes | Core feature | Through extra API steps |
| BytePlus API price | $0.07 per second | $0.12 per second |
| On Nereo | Live, 5 credits per second | Live, 8.5 credits per second |
Planning an OmniHuman Job on Nereo
OmniHuman 2.0 costs 5 credits per second of video on Nereo. The model selector and credit estimate in the tool show the exact cost before you generate.
Billed by Output Seconds
OmniHuman 2.0 costs 5 credits per second of video and OmniHuman 1.5 costs 8.5, so the length of your audio sets the cost.
Test a Short Excerpt First
Run 10 to 15 seconds before a long take to check lip timing, hands, and background stability.
Trim Silence
Cut quiet seconds at the start and end of the track; they become billed video too.
Pick the Right Model
Use OmniHuman 2.0 for long takes and expressive performances, and switch to OmniHuman 1.5 for crisp 1080p short clips.
Simplify the Photo
One face and a plain background reduce drift and other faces moving on their own.
Check Rights and Consent
Only animate images and voices you own or have permission to use, and follow platform AI disclosure rules.
OmniHuman 2.0: Strengths and Practical Considerations
Pros
- Long single takes — Up to 5 minutes without joins, which suits lessons, full songs, and long explainers.
- Expressive, audio-aware performance — Gestures and expression follow the tone and meaning of the recording, not only its sounds.
- Lower per-second price — BytePlus lists 2.0 at $0.07 per second, about 40% less than 1.5.
Cons
- 720p output — 2.0 is documented as native 720p, so large-screen use may need upscaling.
- No published benchmarks yet — BytePlus shares showcases and specs but no quantitative comparison, and busy backgrounds can still drift.
OmniHuman 2.0 FAQ
Find answers to common questions about our AI video generation service.
OmniHuman 2.0 is ByteDance's audio-driven avatar video model, released through BytePlus on October 15, 2026. It turns one image and one audio track, plus an optional text prompt, into a video of that character performing the audio.
The biggest changes are length and cost: audio up to 300 seconds instead of under 60, and $0.07 per second instead of $0.12 on BytePlus. 2.0 also lists multi-person scenes and dialogue as core features. 1.5 can output 1080p, while 2.0 outputs native 720p.
The audio can be 4 to 300 seconds long, and the video matches the audio length, so one take can run up to 5 minutes.
Audio in WAV or MP3 under 15 MB. Images in JPEG, PNG, WebP, BMP, TIFF, GIF, HEIC, or HEIF under 30 MB, 300 to 6,000 pixels per side, with an aspect ratio between 0.4 and 2.5.
Yes. It accepts music as well as speech and follows lyrics, held notes, and rhythm. Keep the vocal clear and up front in the mix for the best mouth movement.
Yes. BytePlus shows animal characters and 3D-style characters among its examples. Results depend on how clearly the eyes and mouth are drawn, so test a short clip first.
The optional text prompt supports Chinese, English, Japanese, Korean, Mexican Spanish, and Indonesian. The spoken audio itself drives the performance.
Yes. OmniHuman 2.0 is live in Nereo's AI Talking Photo Generator, which the button on this page opens, and it is selected by default at 5 credits per second of output. OmniHuman 1.5, Kling AI Avatar, and MiniMax H3 Max Lip Sync are in the same model menu if you want to compare.
