EvoGen-Harness：让图像生成框架学会在哪里、如何进化

View PDF HTML (experimental)

Abstract:Modern text-to-image (T2I) systems can be improved without modifying generator parameters by adapting the external system around frozen generators. However, existing approaches typically optimize a predefined dimension, such as prompts, routing, or workflows, restricting the space in which generation failures can be corrected. Allowing multiple generator-external responsibilities to evolve provides a broader adaptation space, but introduces a new challenge: visual feedback reveals what failed, but not where persistent evolution should occur or how this space should be explored efficiently. We introduce EvoGen-Harness, a generator-agnostic framework for multi-responsibility image-generation harness evolution, together with Trace (Trajectory-Relative Attribution and Coordinated Evolution). Trace aggregates evidence across stochastic executions, uses failure attribution as a search prior to focus candidate updates, and progressively re-attributes residual failures to coordinate evolution across responsibilities, while No-Patch and held-out validation prevent unnecessary or harmful updates. Across GenEval2, T2I-CompBench++, and WISE, EvoGen-Harness improves over the strongest evaluated baselines by +0.2633, +0.0720, and +0.0752, respectively, while achieving 87.9-91.4% attribution recall, 94.8% No-Patch accuracy, and only 1.9% regression. These results demonstrate that attribution-guided multi-responsibility evolution can substantially enhance frozen T2I systems beyond single-dimension adaptation.

Comments:	23 pages, 7 figures
Subjects:	Machine Learning (cs.LG)
Cite as:	arXiv:2610.00383 [cs.LG]
	(or arXiv:2610.00383v1 [cs.LG] for this version)
	https://doi.org/10.48550/arXiv.2610.00383 arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Jiabin Luo [view email]
[v1] Wed, 30 Sep 2026 09:30:52 UTC (3,662 KB)