Live page · Day archive

UniEvo-VL: An On-policy Self-Distillation Training Recipe for Multimodal Model Self-improvement

Read the original at HF Daily Papers

Summary

Researchers introduce UniEvo-VL, a self-evolving framework that improves image generation scores on Qwen-image-2512 from 0.747 to 0.808 on GenEval.

Carried by: Hacker News. First seen: .