Abstract
We propose UniDFlow, a unified discrete flow-matching framework for multimodal understanding, generation, and editing. It decouples understanding and generation via task-specific low-rank adapters, avoiding objective interference and representation entanglement, while a novel reference-based multimodal preference alignment optimizes relative outcomes under identical conditioning, improving faithfulness and controllability without large-scale retraining. UniDFlpw achieves SOTA performance across eight benchmarks and exhibits strong zero-shot generalization to tasks including inpainting, in-context image generation, reference-based editing, and compositional generation, despite no explicit task-specific training.
Abstract (translated)
我们提出了一种名为UniDFlow的统一离散流匹配框架，用于多模态理解、生成和编辑。该框架通过特定任务的低秩适配器分离了理解和生成过程，从而避免了目标干扰和表示纠缠。此外，一种基于参考的多模态偏好对齐技术在相同的条件下优化相对结果，提高了准确性和可控性，并且无需大规模重新训练即可实现这些改进。尽管没有进行特定任务的专门训练，UniDFlow在八个基准测试中取得了最先进的性能，并展示了强大的零样本泛化能力，适用于包括补全、上下文图像生成、基于参考的编辑和组合生成等任务。
URL
https://arxiv.org/abs/2602.12221
