HiDream ai releases HiDream O1 World interactive artificial intelligence model

Beijing, China - August 24, 2026 - HiDream.ai has unveiled HiDream-O1-World, a cutting-edge interactive artificial intelligence model capable of generating complete, three-dimensional worlds from text prompts, images, or simple controls. This innovative system combines roaming, editing, and interaction capabilities, powered by HiDream.ai's self-developed UiT (Unified Transformer) architecture.

Unlike existing models, HiDream-O1-World excels in spatiotemporal and physical consistency, ensuring objects remain stable and realistic as the scene changes. When cameras pan, zoom, or track, the model maintains accurate scene geometry, preventing objects from vanishing or deforming. Collisions, occlusions, and gravitational responses follow real-world causal logic, resulting in a more believable and immersive experience.

Yao Ting, HiDream.ai's Chief Technology Officer, explained, "This is a systematic reconstruction of how the physical world works." By understanding depth, texture, inertia, and light and shadow, AI can begin to genuinely comprehend the world around us. HiDream-O1-World represents a significant step forward in this journey, creating a bridge between AI understanding and a fully realized, interactive world.

Upon its debut, HiDream-O1-World quickly topped industry benchmarks, including WBench, an interactive world model evaluation benchmark developed by Meituan's LongCat team and Fudan University. The model scored an impressive 80.9 on the core Navi sub-leaderboard, leading the Physical dimension at 73.3 and achieving the best overall performance of 88.0 on Consistency. These results surpassed established competitors like Tencent Hunyuan 1.5, setting a new standard for interactive world models.

With just a single click, users can generate a structurally complete and stylistically diverse interactive world from a short description, a single image, or simple controls. For example, uploading a photo of a room allows the model to rapidly construct a high-precision digital twin, filling in the full panorama with accurate proportions and fine-grained detail. Users can explore these worlds in first-person or third-person mode, driving characters and adjusting viewpoints with ease.

The model's capabilities extend beyond mere roving. Users can perform real-time editing, directing characters to perform actions like grabbing, running, crouching, or jumping, or triggering environmental events such as rainfall. These changes maintain globally unified coherence across geometry, lighting, materials, and physical logic, ensuring a seamless and self-consistent experience.

HiDream-O1-World supports a wide range of scenes and styles, from real city streets and natural terrain to anime-style cartoons and AAA-game-grade rendering. The model's two core breakthroughs, spatiotemporal and physical consistency, address long-standing challenges in the field. Long-horizon spatiotemporal consistency allows the model to remember explored structures across viewpoint switches, eliminating drift and scene resets.

Globally stable physical consistency ensures realistic object interactions, with the model performing 13.6% better on visual-plausibility evaluation and 12.7% on causal-fidelity evaluation compared to the industry average.