HermesFlow: Seamlessly Closing the Gap in Multimodal Understanding and Generation
This addresses a common bottleneck in MLLMs for improving multimodal AI applications, though it is incremental as it builds on existing autoregressive paradigms.
The paper tackles the gap between understanding and generation capabilities in Multimodal Large Language Models (MLLMs) by proposing HermesFlow, a framework that uses homologous preference data and iterative optimization to align these capabilities, resulting in significant superiority over prior methods in narrowing this gap.
The remarkable success of the autoregressive paradigm has made significant advancement in Multimodal Large Language Models (MLLMs), with powerful models like Show-o, Transfusion and Emu3 achieving notable progress in unified image understanding and generation. For the first time, we uncover a common phenomenon: the understanding capabilities of MLLMs are typically stronger than their generative capabilities, with a significant gap between the two. Building on this insight, we propose HermesFlow, a simple yet general framework designed to seamlessly bridge the gap between understanding and generation in MLLMs. Specifically, we take the homologous data as input to curate homologous preference data of both understanding and generation. Through Pair-DPO and self-play iterative optimization, HermesFlow effectively aligns multimodal understanding and generation using homologous preference data. Extensive experiments demonstrate the significant superiority of our approach over prior methods, particularly in narrowing the gap between multimodal understanding and generation. These findings highlight the potential of HermesFlow as a general alignment framework for next-generation multimodal foundation models. Code: https://github.com/Gen-Verse/HermesFlow