README.md

metadata

title: Paella
emoji: 🥘
colorFrom: orange
colorTo: yellow
sdk: gradio
sdk_version: 3.9.1
app_file: app.py
pinned: false
models:
  - pcuenq/Paella
license: mit

Paella is a novel text-to-image model that uses a compressed quantized latent space, based on a f8 VQGAN, and a masked training objective to achieve fast generation in ~10 inference steps.