README.md

metadata

title: UniAR
emoji: 🎨
colorFrom: gray
colorTo: red
sdk: gradio
sdk_version: 5.47.1
app_file: app.py
short_description: Unified AR model for image understanding & generation
python_version: '3.12'
startup_duration_timeout: 1h

UniAR Demo

Unified Multimodal Autoregressive Modeling with Shared Context-Visual Tokenizer.

This Space demonstrates UniAR (ShareLab-SII/UniAR-RL), a single Transformer that handles both:

• Image Understanding — visual question answering about an input image
• Image Generation — text-to-image synthesis via autoregressive visual token prediction

The model uses a shared discrete visual tokenizer (Binary Spherical Quantization) as the bridge between understanding and generation, enabling a shared context where the model can directly interpret its own generated visual tokens.

Paper: arXiv:2606.18249 · Model: ShareLab-SII/UniAR-RL · Code: github.com/ShareLab-SII/UniAR