OmniHuman-1 is a ByteDance-developed multimodal human animation system that uses a diffusion-transformer backbone to generate realistic human videos from a single reference image plus a motion signal (audio or driving video). It supports portrait, half-body, and full-body outputs with arbitrary aspect ratios and stylization options (cartoons, animals, objects). The official site highlights real-time lip-sync, gesture realism, and Seedance/Dreamina mentions. The project is documented across the Home page, a Demo Gallery, a Text-to-Video article, use-case pages, a technical review, and a follow-up OmniHuman-1.5 page. Training is described as Omni-Condition training mixing weak and strong conditioning signals to scale data efficiently. Reported training data amounts are around 18.7k hours. Reproducibility appears to rely on omnihuman-lab.github.io materials; ByteDance has not publicly released official weights or download links. Some features (text-to-video) may not be working; pricing is not listed; access seems demo/private. The arXiv paper and lab pages provide architecture and evaluation context; further questions remain about licensing, API access, and dataset transparency.