Roblox launches AI 4D generation beta for game objects

Roblox has launched the beta release of 4D generation, a new AI system that enables text-driven generation of functional, interactive 3D objects in Roblox experiences.^[1][2]

The launch moves Roblox’s AI tools beyond static assets, enabling objects that can move, respond to physics, and interact with gameplay systems.

Roblox announced the beta on February 4, 2026, alongside updated developer documentation and tooling.

Play GIF

GIF

Context and background

Over the past year, Roblox has introduced AI features aimed at simplifying 3D content creation.

Earlier tools, powered by the company’s Cube foundation model, focused on generating static meshes from text.^[3]

Roblox has also discussed longer-term research into AI “world models,” which aim to generate and modify entire environments in real time.

In a recent video, Roblox CEO David Baszucki outlined how the company is exploring AI world models as a future direction for interactive creation.

Those efforts remain experimental, while 4D generation is positioned as a production-ready system for use in live games.

Key details

4D generation is delivered through a new API, GenerateModelAsync, which Roblox says replaces its earlier Mesh Generation API.

According to Roblox’s developer forum announcement, the older API will be deprecated and shut down on March 18, six weeks after the beta launch.^[2]

The system uses predefined schemas that define how generated geometry is broken into parts.

During the beta, Roblox is limiting 4D generation to two predefined schemas. Schemas describe how AI-generated geometry is organized so gameplay systems can interact with it.

The Car-5 schema produces a multi-part object with five components, while Body-1 generates a single mesh.

Roblox says it plans to expand toward an open schema model in future releases.

After geometry is generated, creators attach what Roblox calls retargetable behaviors.

These scripts adapt movement, physics, and interaction logic to the size and shape of each generated object, allowing the same behavior code to work across many designs.

Roblox says typical generation takes 20 to 40 seconds, with rate limits starting at 10 requests per minute per experience, scaling with concurrent users.

Generated objects replicate across servers so all players in an experience can see and interact with them in real time.

The company has also added optional controls for object size and triangle count to help creators manage performance.

Why this matters

Many AI tools can generate 3D visuals, but turning those visuals into usable game objects often requires significant manual work.

Roblox’s approach focuses on structure, generating objects that already conform to known part layouts so they can be connected directly to gameplay systems.

By integrating AI generation into supported engine workflows, Roblox is positioning AI as a tool for building interactive systems, not just visual content.

It also signals a broader shift in the platform, as AI-generated functional objects move from experiments into officially supported features.

Outlook

Roblox says it plans to introduce custom, open schemas that would allow creators to define their own object structures.

The company has not provided a firm timeline for when those capabilities will be available.

Roblox has also said it intends to add object persistence, enabling player-generated items to persist across play sessions.

Both features are under development and are not part of the current beta release.