Gradio Open-Sources 4-Step Text-to-Image Mini Model Trained for Under $20

Gradio has open-sourced a 260M-parameter text-to-image model that generates in just 4 steps under an MIT license. Distilled with HuggingChat's help, it cut inference steps from 32 to 4 at a training cost under $20 per night.

2026-10-01 ~ 2026-10-01 · 2 related posts

• Gradio releases a 260M, 4-step text-to-image model that runs in your browser — Gradio · 2026-10-01
• One prompt, under $20, one night: how Gradio distilled a 4-step text-to-image model — Gradio · 2026-10-01