ByteDance/bytedance-omnihuman-2-0/Released Oct 2026

OmniHuman 2.0

Give any face a voice: one image and an audio track become a full performance

Commercial use

Video

About OmniHuman 2.0

OmniHuman 2.0 is ByteDance's audio-driven human animation model: it turns a single image and a voice track into a video of the subject speaking or singing, with lip sync, facial expression, and body gesture that follow the rhythm and meaning of the audio.

It works from any framing, from close-up portraits to full-body shots, and on photographs as well as illustrated, 3D-rendered, and non-human characters. Images with several people produce multi-person scenes. An optional text prompt directs camera movement, specific actions, and added elements that are not in the source image; prompts can be written in English, Chinese, Japanese, Korean, Mexican Spanish, or Indonesian.

The video runs as long as the audio, from 4 seconds up to 5 minutes, and is billed per second of output. Output is 720p-class with its aspect ratio following the input image. Audio must be WAV or MP3 under 15 MB; the image must be between 300 and 6000 pixels per side with an aspect ratio between 0.4 and 2.5.

Metric	Value
Parameter Count	Unknown
Mixture of Experts	Unknown
Context Length	4 to 300 seconds of audio
Multilingual	Yes (prompts in 6 languages; speech in any language)
Quantized*	Unknown

*Quantization refers to whether the served weights are reduced-precision.

See what this model can do

A handful of real prompts and their outputs.

Stand-Up

Stand-Up Set

"Full-body performance with comic timing, pacing and hand gestures that land on the punchlines."

Multi-Person

Morning Radio Duo

"One image, two speakers: each host talks on their own lines and listens and reacts on the other's."

Prompted Effects

The Vanishing Dove

"The prompt adds doves that are not in the source image, timed to the count of three."

Emotion

Halftime Speech

"Expression and body language escalate with the voice, and the team reacts in the background."

Japanese

Bonsai Master

"Japanese lip sync while his hands work the shears on the tree."

Spanish

Abuela's Tortillas

"Mexican Spanish speech with hands shaping dough, directed by a prompt written in Spanish."

Sculpture

Marble Orator

"Lip sync and expression on carved stone, keeping the marble texture and veining."

3D Character

Sloth Barista

"A stylized, non-human character with sleepy blinks and slow gestures that match the drawl."

Prompted Effects

Science Fair Volcano

"Speech, a hand action and an eruption all directed by one prompt over a single still."

Ready to build with OmniHuman 2.0?

Try OmniHuman 2.0 in the Workbench to prompt it, compare outputs, and iterate on prompts without writing any code. When you're ready to ship, call the same model from our API and build your own apps on top of it.

curl -sSf -X POST https://hub.oxen.ai/api/ai/videos/generate \
    -H "Content-Type: application/json" \
    -H "Authorization: Bearer $OXEN_API_KEY" \
    -d '{
  "model": "bytedance-omnihuman-2-0",
  "input_image": "https://hub.oxen.ai/api/repos/ox/Oxen-AI-Assets/file/main/images/ox_zoom_out_1926_1076.png",
  "input_audio": "https://hub.oxen.ai/api/repos/ox/Oxen-AI-Assets/file/main/audios/example.mp3"
}'

Request parameters

Every field OmniHuman 2.0 accepts in the request body. See the API reference for response formats.

POSThttps://hub.oxen.ai/api/ai/videos/generate

Field	Type	Required	Default	Description
`prompt`	`string`	No	—	Optional direction for camera movement, actions, and added elements.
`input_image`	`string` uri	Yes	—	The person or character to animate.
`input_audio`	`string` uri	Yes	—	Speech or singing that drives the performance.

Model JSON Video generation reference Async queue Models endpoint