Rock, Paper, d1

Liquid AI's d1-3B, a 3B decision model, reads your hand from the webcam and calls the round. It runs on your GPU with WebGPU: your camera never leaves this page.

You

Camera off

d1's opponent

Picks at random before the countdown, face down. d1 only referees.

0you

Best of 5

0bot

Model: LiquidAI/d1-3B (LFM Open License v1.0), converted to ONNX and run with Transformers.js. Each read is one forward pass over one frame and one typed question (rock, paper, scissors or none); the answer is a softmax over the options, as in d1's own system_one. Desktop Chrome with WebGPU; about 3.8 GB on first load, cached after.