ByteDance launches OmniHuman 2.0, generating videos up to 5 minutes from a single image

What happened

ByteDance, the parent company of TikTok, has launched OmniHuman 2.0, the new version of its AI-powered video generation model. It is available through BytePlus, the company's enterprise AI platform.

What it does

OmniHuman 2.0 generates videos ranging from 30 seconds to 300 seconds, or 5 minutes, from a single image and an audio recording. Rather than focusing only on lip sync, the system analyzes the rhythm, tone and meaning of the audio to shape the character's gestures and emotional responses. Users can direct on-screen motion, object generation and camera behavior with text prompts.

Who and what it covers

The model is not limited to human images; animals, 3D characters and characters in different visual styles can also be animated. Multi-character scenes are supported, and some elements not present in the reference image can be added to the scene via prompts.

Pricing and access

According to BytePlus, motion quality improved in the new version and cost fell by 40%. Pricing is listed at $0.07 per second. The model is also being opened to developers through an API, so companies can integrate it into their own content production tools, marketing solutions or digital character applications.

Use cases

BytePlus says the model can be used for advertising and brand content, short videos, film and television production, AI music videos and user-generated content. Rhythm-driven motion and performances that shift with the emotion of the music stand out among the music video scenarios.