What is OmniHuman 1.5?
OmniHuman 1.5 is a ByteDance Intelligent Creation research system for audio-driven human video generation. This page turns one human image plus a voice or performance track into an expressive avatar workflow.
Image + audio to expressive avatar video
OmniHuman 1.5 AI Avatar Generator
Upload one image and up to 35 seconds of MP3 or WAV audio to create an expressive talking-avatar video with natural lip sync, emotion, gestures, and camera motion.
Built from the published OmniHuman 1.5 research workflow and delivered here through EvoLink API access.
Last updated:
Upload voice or performance audio
MP3 or WAV · up to 35 seconds · up to 20 MB
Generation settings
Use faster prompt enhancement for quicker preparation.
Automatically isolate the main human subject.
Check that the uploaded image contains a usable person.
Add audio to calculate credits
Upload only media you own or have permission to use. Do not impersonate, deceive, or violate another person’s rights.
OmniHuman 1.5 quick answers
The essentials for deciding whether this is the right image-and-audio avatar workflow.
What do I need to upload?
Upload one JPG, PNG, or WebP image with a clear person and one MP3 or WAV audio file. A text prompt is optional and can guide emotion, gesture, or camera behavior.
How long can the audio be?
This SeedVideo AI integration accepts audio up to 35 seconds. The research project shows longer examples, but those research capabilities are not the same as this API limit.
How is an OmniHuman generation priced?
The base price is 10 credits per started audio second. A 10.1-second file bills as 11 seconds, or 110 base credits, before any plan discount.
OmniHuman 1.5 input and API specifications
These are the limits enforced by this SeedVideo AI integration, based on the EvoLink API contract.
| Model | OmniHuman 1.5 (`omnihuman-1.5`) |
| Required image | One JPG/JPEG, PNG, or WebP image; up to 10 MB on SeedVideo AI |
| Required audio | One MP3 or WAV file; up to 35 seconds and 20 MB on SeedVideo AI |
| Prompt | Optional performance, emotion, gesture, or camera direction; up to 2,500 characters |
| Prompt languages | Chinese, English, Japanese, Korean, Mexican Spanish, and Indonesian are listed by the upstream API |
| Optional controls | Fast mode, Auto mask, Subject check, and a non-negative integer Seed |
| Processing | Asynchronous EvoLink task; completed provider output is copied to durable SeedVideo AI storage |
Official OmniHuman 1.5 research demos
See how the published research system handles general performance, emotion, and rhythmic motion.
General expressive performance
Official research model demo showing coordinated facial expression, speech timing, body gesture, and camera behavior.
Emotion-led performance
Official research model demo focused on how emotional audio cues influence facial and upper-body expression.
Rhythmic movement
Official research model demo illustrating motion aligned to stronger rhythm and performance timing.
These clips come from the official OmniHuman 1.5 research page. They were not generated on SeedVideo AI, and results from this EvoLink-backed API integration may differ.
Official OmniHuman 1.5 projectHow to make an OmniHuman 1.5 avatar video
A short workflow from portrait and voice track to an asynchronous generated result.
Choose one clear portrait
Use a sharp, unobstructed human subject with enough visible face and body detail for the performance you want.
Add MP3 or WAV audio
Upload a clean track no longer than 35 seconds. The verified server duration determines credit cost.
Add optional direction
Describe delivery, emotion, gesture, or camera intent, then enable only the checks or masking controls you need.
Generate and follow progress
SeedVideo AI securely stores the inputs, submits an asynchronous EvoLink task, and opens the generation status view.
Where image-and-audio avatars are useful
Use OmniHuman when the spoken performance should lead the face, body, and camera behavior.
Product spokesperson clips
Pair a presenter image with a short product script for demos, release notes, and landing-page explainers.
Course and training intros
Create instructor-led openings or module summaries without recording a new camera take for every revision.
Localized creator messages
Test alternate voice tracks and language versions while keeping a consistent on-screen identity.
Social storytelling
Turn character art, portraits, or campaign talent into short voice-led vertical-video concepts.
Music and rhythm studies
Explore movement that follows stronger timing cues when a performance track carries rhythm.
Internal communications
Produce concise update or onboarding videos from approved employee or brand-presenter media.
Input tips for a more usable result
Good source media gives the model clearer identity, timing, and motion signals.
Keep the subject readable
Avoid heavy occlusion, extreme blur, tiny faces, or a crowded image with several equally prominent people.
Use clean, final audio
Reduce background noise and long silence. The model follows timing and emotion from the uploaded track.
Match framing to motion
Use a closer portrait for facial delivery and a wider human image when you expect visible hand or body movement.
Write direction, not a transcript
The audio already supplies speech. Use the optional prompt for delivery, emotion, gesture, and camera guidance.
OmniHuman 1.5 vs Seedance 2.0
Both produce AI video, but they solve different creative jobs and expect different inputs.
| Decision point | OmniHuman 1.5 | Seedance 2.0 |
|---|---|---|
| Primary job | Audio-driven talking avatar and human performance | General text-, image-, and reference-driven video creation |
| Required inputs | One human image + one MP3/WAV track | Prompt, image, or multimodal references depending on mode |
| Audio role | Required performance driver | Generated sound or a multimodal reference depending on workflow |
| Best fit | Spokespeople, lessons, localization, character performance | Ads, scenes, product shots, cinematic concepts, video iteration |
| Site duration limit | Uploaded audio up to 35 seconds | Varies by generation and reference mode |
OmniHuman 1.5 credit pricing
Credits follow the verified audio duration, rounded up to the next started second.
- 10 base credits per started audio second
- Duration is measured on the server and rounded up
- Lite, Pro, and Premium plan credit discounts apply automatically
- Examples: 5.0 seconds = 50 base credits; 5.1 seconds = 60 base credits
OmniHuman 1.5 FAQ
Direct answers about inputs, duration, prompts, pricing, research demos, and usage rights.
Primary sources and integration disclosure
Review the official research and the upstream API contract behind the factual claims on this page.
OmniHuman 1.5 is research from ByteDance Intelligent Creation. SeedVideo AI is an independent service and is not affiliated with ByteDance; this product integration accesses the model through EvoLink.
You are responsible for having the necessary permission and rights for every image and audio file. Do not use generated avatars to impersonate, defraud, harass, or mislead people.
Turn one portrait and one voice track into a performance
Upload your permitted media above and let the audio drive the avatar’s delivery.
Go to the generator