Alibaba launches the world model Happy Oyster, allowing users to give real-time commands to modify scenes being generated by AI.

ME News message. On April 16 (UTC+8), according to Beating Monitoring, Alibaba launched its world model product Happy Oyster. Its positioning is not an AI video generation tool, but a real-time interactive world engine. Unlike the one-time workflow of existing AI video tools—“write prompts, wait for rendering, and compile the final clips”—Happy Oyster continuously receives instructions during generation. The scenes respond in real time and continue to evolve.

The product has two modes. In the “Director” mode, users act as real-time directors: the generated content is a persistent physical world that maintains lighting, gravity, character movement, and scene causality. Users can switch camera angles, direct characters, or change the storyline at any time using text, voice, or images. The world immediately responds and continues unfolding. In the “Roaming” mode, users can freely move from a first-person perspective. It supports keyboard and camera controls. After users leave the initial frame, the world keeps generating and remains coherent.

Technical parameters: Director mode lasts up to 3 minutes at up to 720p; Roaming mode has a limit of 1 minute at 480p. Both modes support multimodal input and output audio synchronously.

For the film and television, as well as gaming industries, Happy Oyster shifts AI video generation from “mass-producing assets” to “real-time virtual production.” During generation, directors can adjust visual styles without needing to fully re-render every time. Game developers can use text to generate walkable level prototype scenes. Alibaba has recently released new models in a concentrated manner and reorganized its organizational structure to focus on AI commercialization. Happy Oyster is among the products most strongly oriented toward content creators and game developers.

(Source: BlockBeats)