en.Wedoany.com Reported - Alibaba ATH has launched HappyOyster 1.0, a real-time interactive open-world model product. This product can construct and dynamically evolve digital worlds, allowing users to freely explore and interact within the generated environment, distinguishing it from one-time text-to-video products on the market.
Current AI video generation products share common limitations: after users input a description, the rendered video clip is unchangeable and cannot interact with the visual content. As duration increases, issues such as inconsistent character appearances and sudden changes in object states often arise. HappyOyster 1.0 aims to create a digital world that can be enacted, explored, and interacted with, where users can issue real-time commands after the scene is generated, and the world responds and evolves accordingly.
The product offers two core modes. The Adventure mode is a world exploration mode, allowing users to control characters to move, jump, and attack within the generated world. The model automatically matches playable interaction methods based on the scene content; for example, a carriage appearing in the scene unlocks horse-riding functionality, while a car enables operations such as turning lights on/off and honking the horn.
During exploration, users can take screenshots to capture scenes, save the world, and share a link with one click, allowing others to enter the complete world. The Directing mode is a real-time director mode, where users can input commands to guide the storyline, supporting multimodal references and locking character appearances. This mode also features backtracking and branching capabilities, enabling users to revert to previous nodes and input different commands to steer the plot toward different branches. The official team also provides an experience guide to help users create better worlds.
The technical architecture of HappyOyster 1.0 is based on several core methods. The product employs closed-loop world state modeling technology, compressing the current world state into a latent state summary and recursively passing it through the generation pipeline to support long-range generation consistency.
To address the issue of subject drift, the product uses persistent reference representations that participate in the full attention mechanism, assigning identifiers to characters and objects to maintain identity stability. Its open causal action space places action commands and natural language within the same semantic interface, allowing the model to automatically deduce subsequent coherent action sequences based on commands. In terms of audio-video coordination, audio and video are jointly decoded and generated within the same world state, ensuring synchronized changes in sound and visuals that conform to physical laws.
Currently, in response to the lack of systematic evaluation benchmarks in the world model field, the HappyOyster team is leading efforts to establish relevant benchmarks in collaboration with Nanjing University. The product has been officially launched and can be accessed by registering with a mobile phone number. The official team also plans to open API interfaces in the near future, with potential applications in game creation, short drama generation, entertainment experiences, digital human live streaming, and virtual companionship.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
