Rethinking UMM Visual Generation: Masked Modeling for Efficient Image-Only Pre-training

Abstract

Unified Multimodal Models (UMMs) are often constrained by the pre-training of their $\text{visual generation components}$, which typically relies on inefficient paradigms and scarce, high-quality text-image paired data. In this paper, we systematically analyze pre-training recipes for $\text{UMM visual generation}$ and identify these two issues as the major bottlenecks. To address them, we propose $\text{Image-Only Training for UMMs (IOMM)}$, a data-efficient two-stage training framework. The first stage pre-trains the visual generative component $\text{exclusively}$ using abundant unlabeled image-only data, thereby removing the dependency on paired data $\text{for this costly phase}$. The second stage fine-tunes the model using a mixture of unlabeled images and a small curated set of text-image pairs, leading to improved instruction alignment and generative quality. Extensive experiments show that IOMM not only improves training efficiency but also achieves state-of-the-art (SOTA) performance. For example, our IOMM-B (3.6B) model was trained from scratch using only $\sim \text{1050}$ H800 GPU hours (with the vast majority, $\text{1000}$ hours, dedicated to the efficient $\text{image-only pre-training stage}$). It achieves $\text{0.89}$ on GenEval and $\text{0.55}$ on WISE--surpassing strong baselines such as BAGEL-7B (0.82 & 0.55) and BLIP3-o-4B (0.84 & 0.50). Code is available $\text{href}$.