LiveByteDance OmniHuman · audio-driven avatar video2 endpoints

OmniHuman API — OmniHuman 1.5 & OmniHuman 2.0 Talking Avatar Video

Use the OmniHuman 1.5 API and OmniHuman 2.0 API on Muapi to turn one portrait and one audio track into a realistic talking or singing video. OmniHuman 2.0 accepts up to 300 seconds of audio with prompt control over scene, motion and camera; OmniHuman 1.5 outputs 720p or 1080p with fast mode. One API key, pay per generation.

Get an OmniHuman API Key Try OmniHuman 2.0 Try OmniHuman 1.5

OmniHuman 2.0

Up to 300 seconds of audio with prompt-controlled scene, motion and camera1 models

Image + Audio to VideoLiveFeatured

OmniHuman 2.0

Animate one portrait with 4–300 second audio into a talking or singing video, with an optional prompt for scene, character motion and camera.

Audio up to 300s

$0.10/sec of audio

Try Model

OmniHuman 1.5

720p or 1080p talking and singing avatar video with fast mode1 models

Image + Audio to VideoLive

OmniHuman 1.5

Turn a portrait and an audio track into a realistic talking or singing video at 720p or 1080p, with optional fast mode and seed control.

720p / 1080p

$0.045–$0.06/sec

Try Model

What is the OmniHuman API?

OmniHuman is ByteDance's audio-driven human animation model family. Give it a single portrait image and an audio track and it generates a video in which the person speaks or sings, with lip sync, facial expression and body motion driven by the audio. Muapi serves both OmniHuman 1.5 and OmniHuman 2.0 behind one API key and one asynchronous request-and-poll workflow.

OmniHuman 2.0 accepts wav or mp3 audio from 4 to 300 seconds and an optional prompt of up to 300 characters describing the scene, character motion and camera movement (Chinese, English, Japanese, Korean, Spanish and Indonesian prompts are supported). It is billed at $0.10 per second of audio, rounded up, from $0.40 (4 seconds) to $30.00 (300 seconds).

OmniHuman 1.5 outputs at 720p or 1080p with an optional fast mode, seed control and a style prompt. It is billed per second of audio at $0.045 (720p) or $0.060 (1080p), for clips from 5 to 60 seconds.

OmniHuman API capabilities

Talking and singing portraits

Animate one portrait image with speech or music and get natural lip sync, expressions and head movement.

Long-form audio up to 300s

OmniHuman 2.0 handles audio from 4 to 300 seconds, enough for explainers, lessons and full songs.

Prompt control

OmniHuman 2.0 takes an optional 300-character prompt for scene, character motion and camera movement.

720p and 1080p output

OmniHuman 1.5 lets you choose 720p or 1080p and enable fast mode for quicker iterations.

OmniHuman API use cases

Digital presenters and spokespeople

Turn one portrait and a voiceover into a presenter video; swap the audio to localize into other languages.

Education and explainers

Convert lesson audio into a talking instructor video up to five minutes long with OmniHuman 2.0.

Music and social clips

Make a character or artwork sing along to a track with natural lip sync for short-form content.

Quick start — OmniHuman API code examples

Submit an OmniHuman task with your Muapi key, receive a request ID, and poll the shared predictions endpoint for the finished video URL.

1. Submit an OmniHuman job

curl -X POST https://api.muapi.ai/api/v1/omnihuman-2-0 \
  -H "x-api-key: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"image_url":"https://example.com/portrait.png","audio_url":"https://example.com/speech.mp3","prompt":"A museum guide explains how starlight lets us look into the past in a friendly, engaging manner."}'

# Response: {"request_id":"REQUEST_ID"}

The example calls OmniHuman 2.0 with image_url, audio_url and an optional prompt. For OmniHuman 1.5, post to omnihuman-1-5 and optionally set output_resolution (720 or 1080) and pe_fast_mode.

2. Poll the result

curl https://api.muapi.ai/api/v1/predictions/REQUEST_ID/result \
  -H "x-api-key: YOUR_API_KEY"

Poll until the task completes, then read the generated video URL from the result response.

OmniHuman API FAQ

Is there an OmniHuman 2.0 API?

Yes. Muapi exposes OmniHuman 2.0 at POST /api/v1/omnihuman-2-0 with image_url, audio_url and an optional prompt, and returns a request ID you poll for the video.

Is there an OmniHuman 1.5 API?

Yes. OmniHuman 1.5 is available at POST /api/v1/omnihuman-1-5 with image_url, audio_url, an optional prompt, output_resolution (720 or 1080), pe_fast_mode and seed.

How much does the OmniHuman API cost?

OmniHuman 2.0 is $0.10 per second of audio (rounded up, minimum 4 seconds, maximum 300 seconds). OmniHuman 1.5 is $0.045 per second at 720p and $0.060 per second at 1080p.

How is OmniHuman 2.0 different from OmniHuman 1.5?

OmniHuman 2.0 takes much longer audio (up to 300 seconds), supports prompt control of scene, motion and camera, and is billed at a flat per-second rate. OmniHuman 1.5 offers 720p/1080p selection, fast mode and seed control for shorter clips.

What inputs does OmniHuman need?

A portrait image URL and an audio URL are required. OmniHuman 2.0 expects wav or mp3 audio of 4 to 300 seconds under 15 MB and images under 30 MB with an aspect ratio between 0.4 and 2.5.

Is OmniHuman available on Fal.ai or Replicate?

OmniHuman 2.0 is not listed on Fal.ai or Replicate. Muapi serves both OmniHuman versions through one unified API.

Compare AI avatar and lipsync APIs →See OmniHuman next to InfiniteTalk, Kling Avatar, Wan 2.2 and more.InfiniteTalk API →Audio-driven image and video lipsync as an alternative.Browse lipsync models →Compare audio-driven video endpoints.Read the API docs →Learn Muapi's authentication, submit and polling flow.

Build with OmniHuman

Add talking and singing avatar video to your product with OmniHuman 1.5 and 2.0 through one Muapi API key.

Get an API Key Try OmniHuman