Paper Overview
Field: Computer Vision (CV) Authors: Dian Zheng, Manyuan Zhang, Hongyu Li Published: 2025-05-20 arXiv: 2505.15987
Summary
Currently, enhancing Unified Multimodal Models (UMMs) with image understanding, generation, and editing capabilities mainly relies on mixed multi-task training. Due to inherent task conflicts, such a strategy requires complex multi-stage pipelines, massive data mixing, and balancing tricks, merely resulting in performance trade-offs rather than true mutual reinforcement.
To break this paradigm, the authors propose Uni-Edit, an intelligent image editing task that serves as the first general task for UMM tuning. Unlike complex mixed pipelines, Uni-Edit improves performance across all three abilities at once using only one task, one training stage, and one dataset.
Key Contributions
- Image editing as an ideal general task: it naturally demands both visual understanding and generation, making it uniquely suited for unified model tuning.
- Automated data synthesis pipeline: the first automated and scalable pipeline that converts diverse VQA data into complex, effective editing instructions with embedded questions and nested logic, better exploiting the model's understanding capability than overly simple existing edit instructions.
- Uni-Edit-148k dataset: 148k pairs of reasoning-intensive editing instructions with high-quality edited images.
Results
Extensive experiments on BAGEL and Janus-Pro demonstrate that fine-tuning solely on Uni-Edit achieves simultaneous improvements across understanding, generation, and editing—without any auxiliary operations.
Original Abstract (excerpt)
> Currently, enhancing Unified Multimodal Models (UMMs) with image understanding, generation, and editing capabilities mainly relies on mixed multi-task training. Due to inherent task conflicts, such strategy requires complex multi-stage pipelines, massive data mixing, and balancing tricks, merely resulting in a performance trade-off rather than true mutual reinforcement. To break this paradigm, we propose Uni-Edit, an intelligent image editing task that serves as the first general task for UMM tuning...
---
*Auto-collected on 2026-05-22. Source: zhichai.net forum post.*