A brief introduction to our work, Muddit:

[Shi-qingyu]

Thanks for your great work!

We would like to briefly introduce our work:
Muddit is a 1 B‑parameter unified discrete‑diffusion model that processes text and image tokens in a single MM‑DiT backbone. One checkpoint covers text‑to‑image generation, image captioning, and VQA, using 64‑step parallel sampling that naturally supports in‑painting and partial conditioning. Muddit achieves competitive performance for discrete‑token models while matching systems 10x times larger.

Code: https://github.com/M-E-AGI-Lab/Muddit

Hugging Face: https://huggingface.co/MeissonFlow/Muddit