OmniHuman 2.0

OmniHuman 2.0 is ByteDance's avatar video model. Give it one picture and an audio track: the person, cartoon character or animal in the picture speaks, sings or performs in time with the sound, with matching lips, facial expressions, gestures and body movement. The video is as long as the audio, from 4 seconds up to 5 minutes, in 720p.

Endpoint id	Model page
flymyai/bytedance-omnihuman-2_0	https://app.flymy.ai/models/flymyai/bytedance-omnihuman-2_0

Make a talking video

A video takes a few minutes, so use the async route: submit, then poll the result. Inputs are form fields: upload the files, or pass public URLs.

curl -X POST https://api.flymy.ai/api/v1/flymyai/bytedance-omnihuman-2_0/predict/async/ \
  -H "X-API-KEY: fly-***" \
  -F 'image=@presenter.jpg' \
  -F 'audio=@voice.mp3' \
  -F 'prompt=He talks warmly to the camera, smiles and gestures with his hands'
# {"prediction_id": "02ecd473..."}

curl "https://api.flymy.ai/api/v1/flymyai/bytedance-omnihuman-2_0/predict/async/result/?request_id=02ecd473..." \
  -H "X-API-KEY: fly-***"

The result route answers 425 while the video is being made. When it is ready, it returns:

{
  "inference_responses": [
    {
      "response": {
        "video": [{"url": "https://storage.googleapis.com/...", "content_type": "video/mp4"}],
        "calculated_price": "0.700000"
      },
      "infer_details": {"status": 200, "request_id": "02ecd473-..."}
    }
  ]
}

url is a signed link valid for 24 hours; download the file to keep it. calculated_price is what the run cost.

Inputs

Field	Required	What it takes
image	yes	One picture of a person, character or animal: JPEG, PNG, WEBP, BMP, TIFF, GIF, HEIC or HEIF, under 30 MB, each side 300-6000 px, width to height between 0.4 and 2.5
audio	yes	The voice or song that drives the video: WAV or MP3, 4 to 300 seconds, under 15 MB
prompt	no	Directs the scene, the movements and the camera, in Chinese, English, Japanese, Korean, Spanish or Indonesian

The model does not write or voice the speech: bring the audio, recorded or made with a text-to-speech model. The result has no watermark.

An audio track that is not WAV or MP3, or is shorter than 4 seconds or longer than 300 seconds, is refused before the video is started, so it costs nothing. So is a picture or sound the provider's content check does not accept.

From an AI assistant

An assistant connected to FlyMy.AI finds the model with recommend_model, quotes it with get_pricing and makes the video with run_model, passing image and audio as public URLs. The result is a link to the video file.

Pricing

$0.07 per second of video, counted in whole seconds: a 10-second clip costs $0.70, a 30-second clip $2.10. A refused or failed run costs nothing.