README.md

metadata

title: Rosetta Multimodal
emoji: 🪨
colorFrom: indigo
colorTo: blue
sdk: gradio
sdk_version: 5.49.1
app_file: app.py
pinned: false
short_description: Unified text-to-image generation and image understanding
python_version: '3.10'
startup_duration_timeout: 2h

Rosetta — Composable Native Multimodal

A Gradio demo for tencent/Rosetta-inference (Rosetta-3.8B-A1B), a unified multimodal model from Tencent Hunyuan × HKUST that does both text-to-image generation and image understanding with a single backbone (Qwen3-0.6B LLM + Qwen3-VL vision encoder + FLUX.2 VAE).

Modes

• Text → Image — flow-matching image generation from a text prompt.
• Image Understanding — ask questions about an uploaded image (VLM).

Model code is adapted from the upstream repo Lxiangyue/Rosetta. Weights and shared assets (VAE / ViT / tokenizer) are fetched from the model repo on first launch.