HappyOyster 1.0, Alibaba's real-time world model, is open to everyone.
Alibaba's ATH branch opens HappyOyster 1.0, an interactive real-time world model generating 720p environments from text, images, and sound.
HappyOyster is launching version 1.0 and opening to the public, following an initial early access phase. Developed by Alibaba Token Hub (ATH), the branch the group established to consolidate its AI initiatives, this world model doesn't produce a static video: it generates an interactive world that evolves in real time based on instructions, integrating image and sound.
Two modes define the experience. In Wandering mode, users generate a universe from text or an image, then navigate it like a game, using the keyboard (WASD keys), in first or third person, from underwater environments to dreamlike landscapes. In Directing mode, the user takes on the role of director: they direct the camera, characters, and plot via voice, text, or image, can pause a scene and rewrite its storyline, for sequences the company claims can be up to three minutes long in 720p.
HappyOyster builds upon HappyHorse, ATH's previous video generation model, and positions Alibaba in the race for world models, competing with Tencent and World Labs. Access is available via the website, with free credits offered daily for one month.