The Year in AI Papers
The year/Independent research

Paper 2510.06308

Lumina-DiMOO: An Omni Diffusion Large Language Model for Multi-Modal Generation and Understanding

Yi Xin, Qi Qin, Siqi Luo, Kaiwen Zhu, Juncheng Yan, Yan Tai, Jiayi Lei, Yuewen Cao, Keqi Wang, Yibin Wang, Jinbin Bai, Qian Yu, et al.

Published
Oct 2025
Research lab
Independent
Citations
87
GitHub
1K stars

01 In brief

Summary

Lumina-DiMOO is an open-source, 8B-parameter unified multi-modal model based on a fully discrete diffusion architecture, built on the LLaDA-Base dLLM.

It supports text-to-image generation at arbitrary resolutions, image-to-image tasks (editing, style transfer, subject-driven, controllable, multi-view), and image understanding.

The model uses a unified masked diffusion objective over text and image tokens, with a 16x downsampling aMUSEd-VQ tokenizer and special tokens for structure.

Inference employs parallel sampling with a cosine mask schedule and a training-free ML-Cache for 2x speedup, achieving 32x faster text-to-image generation than Lumina-mGPT 2.0.

Training includes four stages: pre-training (80M pairs), mid-training for diverse tasks, SFT, and a novel Self-GRPO reinforcement learning stage that jointly optimizes generation and understanding.

On GenEval, Lumina-DiMOO scores 0.88 (0.91 with Self-GRPO), surpassing GPT-4o (0.84) and BAGEL (0.82); it also leads on DPG (86.04), UniGenBench (71.12), and OneIG-EN (0.455), and achieves strong understanding results (POPE 87.4, MMB 84.5, MMMU 58.6).

The model enables zero-shot inpainting and interactive retouching.

Code and checkpoints are released.

02 From the paper

Abstract

We introduce Lumina-DiMOO, an open-source foundational model for seamless multi-modal generation and understanding. Lumina-DiMOO sets itself apart from prior unified models by utilizing a fully discrete diffusion modeling to handle inputs and outputs across various modalities. This innovative approach allows Lumina-DiMOO to achieve higher sampling efficiency compared to previous autoregressive (AR) or hybrid AR-Diffusion paradigms and adeptly support a broad spectrum of multi-modal tasks, including text-to-image generation, image-to-image generation (e.g., image editing, subject-driven generation, and image inpainting, etc.), as well as image understanding. Lumina-DiMOO achieves state-of-the-art performance on multiple benchmarks, surpassing existing open-source unified multi-modal models. To foster further advancements in multi-modal and discrete diffusion model research, we release our code and checkpoints to the community. Project Page: https://synbol.github.io/Lumina-DiMOO.