HiDream.ai Develops Interactive AI World Models
Chinese AI firm HiDream.ai has launched its HiDream-O1-World model, capable of generating interactive environments from text, image, and other inputs. The company focuses on developing AI models for broader applications.
Chinese artificial intelligence company HiDream.ai has introduced its new interactive world model, HiDream-O1-World. The model generates complex, navigable virtual environments based on text, image, and interactive inputs, allowing users to alter scenes as they unfold.
Founded in March 2023, HiDream.ai focuses on creating generative multimodal AI models. CEO Tao Mei, formerly of Microsoft Research Asia and JD.com, brings extensive experience in vision and language processing. The company's core technical team is highly qualified, with over 90% holding advanced degrees.
Initially, HiDream.ai concentrated on image and video generation. The company employed a cost-saving "dual-model" approach, testing concepts with cheaper image models before scaling to video. HiDream.ai has also utilized open-source contributions, releasing its 8-billion-parameter HiDream-O1-Image model under an MIT license.
The move into interactive world models aligns with a broader industry trend as AI shifts from content creation to simulating interactive experiences. While HiDream-O1-World initially ranked highly on the WBench Navi leaderboard, its position has since shifted. HiDream.ai has also adapted its business strategy to better serve enterprise clients, moving beyond raw models to offer integrated software platforms and workflows.