HiDream Unveils HiDream-O1-Video-1.0, a Native Omnimodal Video Model Built for Physical Consistency

HiDream.ai has introduced its latest AI model, HiDream-O1-Video-1.0, designed for native omnimodal video generation. This model supports inputs from text, images and video, and produces high-quality 1080p videos lasting between 5 to 20 seconds with synchronized audio.

HiDream V1 is built around a deeper understanding of creative intent and physical consistency. It excels in generating high-fidelity videos, maintaining narrative coherence, consistent character portrayal, physical plausibility, and audio-visual synchronization. In benchmark tests by Artificial Analysis and Arena.ai, HiDream V1 secured the 4th and 8th positions respectively, positioning it among the top video generation models globally.

The model's innovative architecture involves understanding user intent, planning the narrative and character states, and generating while aligning with the initial intent. It also accounts for physical reasoning involving factors like gravity, inertia, collisions, and spatial continuity, making the generated videos more reflective of the real world.

Unlike other models that require a fixed video duration, HiDream V1 determines the video length based on the unfolding event and the content's pacing requirements. Thus, the model can generate more natural pacing and complete sequences without unnecessary pauses. HiDream V1 represents a step forward in video generation, moving beyond mere visual sharpness and smooth motion to a deeper understanding of instructions, coherent performances, and aligned sound and visuals.