Overworld is a local, browser-playable AI world model demo for Pinokio. Pick a seed image, wait for the engine to initialize, then move through the generated world with normal game controls.

This launcher uses Overworld's Biome server components with a minimal web game UI. The default profile is the lowest VRAM option and is prepared during install so the first launch can reuse local model and compile caches.

License

• Biome: https://github.com/Overworldai/Biome
• World Engine: https://github.com/Overworldai/world_engine

The Overworld world engine and Biome server are GPL licensed.

Screenshots

How To Use

1. Click Install in Pinokio.
2. Click Start.
3. Click Play when Pinokio opens the local web UI.
4. Choose an example seed image, or drag in your own image.
5. Use WASD to move, point the mouse to look, Shift to move faster, and Space for the jump/action control.

The default install prepares Low VRAM 360p INT8. Other profiles can be selected inside the app; the first run of a new profile may take longer while the model loads and Torch/Triton caches are built.

Profiles

Profile	Model	Quantization	Estimated VRAM
Low VRAM 360p INT8	Overworld/Waypoint-1.5-1B-360P	intw8a8	8-10 GB
360p BF16	Overworld/Waypoint-1.5-1B-360P	none	12-14 GB
720p INT8	Overworld/Waypoint-1.5-1B	intw8a8	14-16 GB
720p BF16	Overworld/Waypoint-1.5-1B	none	18-20 GB

The default 360p model download is about 3.7 GB. Actual disk usage is higher after Python packages, CUDA/PyTorch dependencies, Hugging Face cache files, and compiled kernel caches are created.

Controls

• W/A/S/D: move
• Mouse: look around
• Shift: faster movement
• Space: jump/action
• Lock icon: capture mouse pointer
• Reset icon: reset the current world state
• Menu icon: open profile, seed, upload, and log controls

API

The local server exposes HTTP endpoints and a WebSocket stream at the URL shown by Pinokio.

JavaScript

const ws = new WebSocket("ws://127.0.0.1:PORT/ws")
ws.binaryType = "arraybuffer"

ws.onopen = () => {
  ws.send(JSON.stringify({
    type: "init",
    req_id: "1",
    model: "Overworld/Waypoint-1.5-1B-360P",
    quant: "intw8a8",
    seed_image_data: "<base64 jpeg>",
    seed_filename: "default.jpg",
    scene_authoring: false,
    action_logging: false,
    video_recording: false,
    cap_inference_fps: true
  }))
}

ws.send(JSON.stringify({
  type: "control",
  buttons: ["W"],
  mouse_dx: 0,
  mouse_dy: 0,
  ts: performance.now()
}))

Python

import asyncio
import json
import websockets

async def main():
    async with websockets.connect("ws://127.0.0.1:PORT/ws") as ws:
        await ws.send(json.dumps({
            "type": "control",
            "buttons": ["W"],
            "mouse_dx": 0,
            "mouse_dy": 0,
            "ts": 0,
        }))

asyncio.run(main())

Curl

curl http://127.0.0.1:PORT/health
curl http://127.0.0.1:PORT/api/model-info/Overworld/Waypoint-1.5-1B-360P

Use a WebSocket client for /ws; plain curl cannot drive the live frame stream.