REST API

The VoxelAI Studio API turns a text prompt into a clean, game-ready 3D voxel model. It is built around an agentic LLM pipeline: rather than inflating a picture into a mesh, the model reasons about your prompt and operates the editor the way you would — creating layers, defining materials, and placing real voxels with primitives and boolean operations. The result is a structured, layered, fully editable model, not a baked blob.

Public API access is on the way

These endpoints already power the Voxel web app and the Roblox Studio plugin, and the reference here is accurate to what ships today. What is still rolling out is self-serve access for third-party developers — programmatically issuing your own tokens and calling the API from your own app. For now, the only externally-issued credential is the vxlrbx_ Personal Access Token minted through the Roblox Studio device-link flow (see Authentication). Want early access for an integration? Reach out.

How generation works

Generation is synchronous — you send a prompt and get the finished model back. There are no jobs to submit and poll. Two endpoints deliver the same agentic build differently:

• POST /api/generate/model — streams the build over Server-Sent Events, one operation at a time, so you can render it assembling live. This is what the in-app generator uses.
• POST /api/generate/roblox — runs the same build and returns the finished model as a single JSON document (a flat voxel list plus the material palette). Recommended for any non-streaming client or agent.

See Endpoints for full request and response shapes, and Authentication for the required Bearer token.

The image-to-3D pipeline is deprecated

Earlier versions exposed a two-step generate images → voxelize pipeline that ran on RunPod GPU workers. That pipeline is deprecated and no longer used. If you find references to /api/generate/images, /api/generate/voxelize, or /api/generate/status, ignore them for new work — they’re documented only as a historical note.

Tokens

An agentic model build costs 3 tokens, deducted up front when the request is submitted and automatically refunded if it fails, is empty, or is aborted — so a failed or abandoned generation never costs you anything.

The balance is server-authoritative — read it any time with GET /api/tokens. New free accounts get a one-time grant of 6 tokens, then refill monthly (free 4 / builder 320 / pro 720 / creator 2000). See the pricing page for plans and token packs.

Note on prompting

You do not need to over-engineer your prompt. A simple noun phrase like “a rustic wooden treasure chest” is enough — the agent handles the voxel styling and structure for you.