*I built michelangelo.best — a 100% free, no-login text-to-image playground.*

### Why another T2I site? - I kept getting locked behind paywalls or signup walls when showing AI art to friends. - I wanted something dead-simple: open the page, type a prompt, get a PNG.

### What’s different? 1. *Fast & lightweight*: running a quantized model on a bare-metal 4060 (8 GB) with queueing + auto-cleanup. 2. *Glitch bonus*: the model is not SOTA. Rough edges produce occasional “cursed” or surreal results — the community started calling it free glitch-art mode. I embraced it: there’s a toggle to “allow wild mutations”. 3. *Public REST endpoint*: `POST /generate` with `prompt`, `steps`, etc. (docs in footer).

### Known issues - Faces can melt; hands can go Lovecraftian. - Only English/CN prompts are tested. - Heavy traffic will queue (I show ETA).

### Roadmap - Switch to SDXL Turbo when VRAM budget allows. - Add seed & negative-prompt fields. - Open-source the FastAPI backend if there’s interest.

Would love your feedback!

Demo images (normal + glitchy): https://michelangelo.best/demo/normal.png | [link2] | [link3]

Thanks for checking it out — happy to answer anything.