Saved in:
Bibliographic Details
Main Authors: Wang, Zeyu, Chen, Zilong, Gou, Chenhui, Li, Feng, Deng, Chaorui, Zhu, Deyao, Li, Kunchang, Yu, Weihao, Tu, Haoqin, Fan, Haoqi, Xie, Cihang
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2510.22946
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908666372816896
author Wang, Zeyu
Chen, Zilong
Gou, Chenhui
Li, Feng
Deng, Chaorui
Zhu, Deyao
Li, Kunchang
Yu, Weihao
Tu, Haoqin
Fan, Haoqi
Xie, Cihang
author_facet Wang, Zeyu
Chen, Zilong
Gou, Chenhui
Li, Feng
Deng, Chaorui
Zhu, Deyao
Li, Kunchang
Yu, Weihao
Tu, Haoqin
Fan, Haoqi
Xie, Cihang
contents Unified multimodal models have recently shown remarkable gains in both capability and versatility, yet most leading systems are still trained from scratch and require substantial computational resources. In this paper, we show that competitive performance can be obtained far more efficiently by strategically fusing publicly available models specialized for either generation or understanding. Our key design is to retain the original blocks while additionally interleaving multimodal self-attention blocks throughout the networks. This double fusion mechanism (1) effectively enables rich multi-modal fusion while largely preserving the original strengths of the base models, and (2) catalyzes synergistic fusion of high-level semantic representations from the understanding encoder with low-level spatial signals from the generation encoder. By training with only ~ 35B tokens, this approach achieves strong results across multiple benchmarks: 0.91 on GenEval for compositional text-to-image generation, 82.16 on DPG-Bench for complex text-to-image generation, 6.06 on GEditBench, and 3.77 on ImgEdit-Bench for image editing. By fully releasing the entire suite of code, model weights, and datasets, we hope to support future research on unified multimodal modeling.
format Preprint
id arxiv_https___arxiv_org_abs_2510_22946
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle LightFusion: A Light-weighted, Double Fusion Framework for Unified Multimodal Understanding and Generation
Wang, Zeyu
Chen, Zilong
Gou, Chenhui
Li, Feng
Deng, Chaorui
Zhu, Deyao
Li, Kunchang
Yu, Weihao
Tu, Haoqin
Fan, Haoqi
Xie, Cihang
Computer Vision and Pattern Recognition
Unified multimodal models have recently shown remarkable gains in both capability and versatility, yet most leading systems are still trained from scratch and require substantial computational resources. In this paper, we show that competitive performance can be obtained far more efficiently by strategically fusing publicly available models specialized for either generation or understanding. Our key design is to retain the original blocks while additionally interleaving multimodal self-attention blocks throughout the networks. This double fusion mechanism (1) effectively enables rich multi-modal fusion while largely preserving the original strengths of the base models, and (2) catalyzes synergistic fusion of high-level semantic representations from the understanding encoder with low-level spatial signals from the generation encoder. By training with only ~ 35B tokens, this approach achieves strong results across multiple benchmarks: 0.91 on GenEval for compositional text-to-image generation, 82.16 on DPG-Bench for complex text-to-image generation, 6.06 on GEditBench, and 3.77 on ImgEdit-Bench for image editing. By fully releasing the entire suite of code, model weights, and datasets, we hope to support future research on unified multimodal modeling.
title LightFusion: A Light-weighted, Double Fusion Framework for Unified Multimodal Understanding and Generation
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2510.22946