HiDream-O1-Video-1.0

HiDream-O1-Video-1.0 (also known as HiDream V1) is a native omnimodal video generation model developed by HiDream.ai, introduced on September 17, 2026. Designed around a deeper understanding of real-world physics and creative intent, the model accepts multi-modal inputs including text prompts, reference images, and video sequences to produce consistent output clips. It targets physical plausibility and temporal consistency to reduce common generative flaws like object drifting, unnatural motion, or scene resets.

The model generates 1080p high-fidelity video clips ranging from 5 to 20 seconds, featuring content-adaptive duration planning to better pace movements and events. In addition to visual rendering, HiDream-O1-Video-1.0 incorporates native audiovisual joint modeling, allowing it to output clips with synchronized sound effects and dialogue aligned directly with scene dynamics. On industry benchmarks, the model has demonstrated competitive performance on image-to-video evaluation leaderboards.