HermesFlow: Seamlessly Closing the Gap in Multimodal Understanding and Generation

TL;DR

HermesFlow is a framework that effectively aligns multimodal understanding and generation in large language models, significantly reducing their existing capability gap through homologous preference data and iterative optimization.

Contribution

We introduce HermesFlow, a novel framework that bridges the understanding-generation gap in multimodal models using homologous data and self-play optimization.

Findings

HermesFlow outperforms prior methods in aligning understanding and generation.

The approach narrows the capability gap in multimodal models.

Extensive experiments validate the effectiveness of HermesFlow.

Abstract

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