What Is Happy Oyster?

Happy Oyster is Alibaba's 3D world model, launched on April 16, 2026. It represents a fundamental shift from the passive generation approach of traditional text-to-video models to active simulation of three-dimensional worlds. Rather than producing a flat video you watch, Happy Oyster creates interactive 3D environments you can direct, explore, and modify in real time.

How Happy Oyster works

At its core, Happy Oyster uses what Alibaba calls a native multimodal architecture. This means the model deeply integrates multimodal understanding with audio-video co-generation in a single unified system. The technical goal is to simulate and restore the physical and spatial attributes of real-world environments, creating a foundational layer for 3D content generation.

The model performs what Alibaba describes as "world evolution modeling over long time spans." Instead of generating a short clip and stopping, Happy Oyster continuously simulates how a world changes and develops. This makes it fundamentally different from tools like Sora, Runway, or even its sibling model Happy Horse, which produce time-limited video outputs.

Happy Oyster adapts scenes in real time based on user interaction. If you change something in the environment, the model responds by recalculating how the rest of the world should look and behave. This real-time adaptation is the key technical differentiator.

The two core modes

Happy Oyster ships with two distinct interaction paradigms.

Directing mode gives you creative control over a physical 3D world. You can build scenes, adjust lighting, modify storylines, and tweak compositions as events unfold. Think of it as a real-time director's chair for AI-generated worlds. The model responds to your adjustments instantly, maintaining physical coherence while incorporating your creative choices.

Wandering mode takes a different approach. From a single text prompt, the model generates an expanding 3D environment that you explore in first person. As you move through the space, the world continues to grow and evolve around you. There is no fixed boundary; the environment expands as you explore it.

Both modes represent Alibaba's vision of shifting content production from "passive generation" to "active simulation of world evolution."

Who Happy Oyster is built for

Alibaba explicitly targets three audiences with Happy Oyster. Game developers can use it to prototype and generate interactive 3D environments rapidly. Film production teams can use Directing mode to compose scenes and explore visual possibilities before committing to traditional production pipelines. Interactive content creators can build explorable experiences that respond to audience participation.

The model also has broader applications in robot training and environmental perception for autonomous driving, since the same world simulation technology that creates game environments can also create training environments for physical AI systems.

The technical family tree

Happy Oyster comes from Alibaba's ATH Innovation Division, which is part of the newly formed Alibaba Token Hub business group. The division is led by Zhang Di, former VP of Kuaishou and former head of Kling AI technology. The same team launched Happy Horse one week earlier, a 2D video generation model that achieved 1389 Elo points on the Artificial Analysis leaderboard, topping models like Seedance 2.0 by 115 points.

Happy Horse uses a Transfusion architecture integrating autoregressive text prediction with diffusion-based visual generation. Happy Oyster builds on similar multimodal foundations but extends the output from 2D video to interactive 3D worlds.

What to know before going further

Happy Oyster is currently in limited early access only. There is no public API, no general signup, and no confirmed pricing. If you want to understand how to use it when access becomes available, see How to Use Happy Oyster. For prompt strategies, visit Happy Oyster Prompts. For working alternatives available today, check Happy Oyster Alternatives.