---
title: Omni-Diffusion
emoji: 🌀
colorFrom: yellow
colorTo: purple
sdk: gradio
sdk_version: 5.49.1
app_file: app.py
short_description: Any-to-any multimodal masked discrete diffusion model
python_version: "3.10"
startup_duration_timeout: 1h
pinned: false
---

# Omni-Diffusion

Interactive demo for **Omni-Diffusion: Unified Multimodal Understanding and
Generation with Masked Discrete Diffusion**
([paper](https://huggingface.co/papers/2603.06577) ·
[model](https://huggingface.co/lijiang/Omni-Diffusion) ·
[code](https://github.com/VITA-MLLM/Omni-Diffusion)).

Omni-Diffusion is an any-to-any multimodal language model built on a mask-based
discrete diffusion model. It jointly models discrete tokens of text and images.

This Space exposes three of the model's capabilities:

- **Text → Image**: generate an image from a text prompt.
- **Image → Text (VQA)**: answer questions / describe an uploaded image.
- **Text → Text**: instruction following / chat.

Image (de)tokenization uses [`showlab/magvitv2`](https://huggingface.co/showlab/magvitv2).