Qwen-Image-PE improves text-to-image prompts for better visual design
Qwen-Image-PE: Learning Visual Design for Text-to-Image Prompt Enhancement
Computer Vision and Pattern Recognition
Summary
Text-to-image generators turn words into pictures, but simple prompts often miss important design details. The authors created Qwen-Image-PE, a method that teaches a model to enhance prompts by learning good visual design from existing images and captions. It improves aspects like lighting, layout, and detail, making generated images look better and more coherent. The approach works well across different text-to-image engines and scales with more data and model size.
What this means in practice
- •For graphic designers: Enhance user prompts automatically to create more visually appealing images with better design elements in design software.
- •For game developers: Improve text inputs for procedural image generation to produce higher-quality and more cohesive game asset visuals.
- •For advertising agencies: Generate enhanced, well-designed image concepts from client text descriptions faster when creating marketing materials.$Commercial implications: Enables a commercial product that refines client prompts to produce superior marketing visuals, streamlining creative workflows.
Authors
Zekai Zhang, Yanran Zhang, Shengming Yin, Yan Shu, Zhening Liu, Bingxuan Zhao, Chenyang Zhu, Donghao Zhou, Jiahao Li, Jiahao Wang, Jie Zhang, Kaiyuan Gao, Kun Yan, Lihan Jiang, Ningyuan Tang, Qihang Ai, Rui Song, Shuo Cao, Xiao Xu, Xiaoyue Chen, Yuxiang Chen, Zhendong Wang, Zhouyang Li, Zihao Liu, Zikai Zhou, Bo Zheng, Huishuai Zhang, Dongyan Zhao, Chenfei Wu
Abstract
Learning visual design is essential for prompt enhancement (PE) to turn underspecified user requests into coherent visual specifications for text-to-image (T2I) generation. Existing methods preserve user intent and improve overall visual appeal, with less explicit emphasis on fine-grained design dimensions. To address this gap, we introduce Qwen-Image-PE, which learns visual design through two-stage training. Supervised fine-tuning with Caption-Based Reverse Distillation (CRD) uses dense captions of well-designed images as design supervision. Reinforcement learning with Groupwise Design Reward (GDR) jointly compares all candidate images for the same request to refine lighting, composition, layout, and detail. On Qwen-Image-Bench with Qwen-Image-2.1, Qwen-Image-PE-397B improves over original requests by 7.81 points, while Qwen-Image-PE-9B outperforms all evaluated PE baselines. Further evaluations show gains in visual design, aesthetics, image quality, and human preference. We find that PE scales along three axes: training data, model size, and information density, with sufficient information density as a prerequisite for effective enhancement. Moreover, we demonstrate that Qwen-Image-PE generalizes to four additional T2I backbones, even surpassing native prompt enhancers on HunyuanImage-2.1 and FLUX 3 Image Klein. We release Qwen-Image-PE-9B at https://huggingface.co/Qwen/Qwen-Image-2.1-PE-T2I.