Qwen-Image-PE: Learning Visual Design for Text-to-Image Prompt Enhancement

What happened: Qwen researchers introduced Qwen-Image-PE, a two-stage training method for text-to-image prompt enhancement that uses Caption-Based Reverse Distillation and a Groupwise Design Reward. Key numbers: Qwen-Image-PE-397B improves over original requests by 7.81 points on Qwen-Image-Bench with Qwen-Image-2.1. Why it matters: The authors report gains in visual design, aesthetics, image quality, and human preference, and say the method generalizes to four additional T2I backbones, surpassing native prompt enhancers on HunyuanImage-2.1 and FLUX 3 Image Klein.

Qwen researchers introduced Qwen-Image-PE, a two-stage training method for text-to-image prompt enhancement that uses Caption-Based Reverse Distillation and a Groupwise Design Reward.

The authors report gains in visual design, aesthetics, image quality, and human preference, and say the method generalizes to four additional T2I backbones, surpassing native prompt enhancers on HunyuanImage-2.1 and FLUX 3 Image Klein.