Qwen-Image-PE learns visual design for text-to-image prompt enhancement
Original title: Qwen-Image-PE: Learning Visual Design for Text-to-Image Prompt Enhancement
The released 9B model beats evaluated prompt-enhancement baselines, while the 397B variant gains 7.81 points on Qwen-Image-Bench.
Qwen-Image-PE trains prompt enhancers to specify lighting, composition, layout and detail rather than only preserving intent and general appeal. Supervised fine-tuning uses Caption-Based Reverse Distillation from dense captions of well-designed images, then reinforcement learning applies a Groupwise Design Reward across candidate images for the same request. On Qwen-Image-Bench with Qwen-Image-2.1, Qwen-Image-PE-397B improves 7.81 points over original requests, and Qwen-Image-PE-9B outperforms all evaluated baselines. The method also transfers to four other text-to-image backbones, including HunyuanImage-2.1 and FLUX 3 Image Klein. Qwen-Image-PE-9B is released.
Unconfirmed: Benchmark gains and baseline comparisons are reported by the paper’s authors.
- Authors say effective enhancement needs enough information density, then scales with data and model size.
- Evaluations also report gains in aesthetics, image quality and human preference.
- It surpasses native prompt enhancers on HunyuanImage-2.1 and FLUX 3 Image Klein.
- Only Qwen-Image-PE-9B is stated as released.
- Vendors
- Qwen
From the original article
1 min read at huggingface.co
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.
By Zekai Zhang, Yanran Zhang, Shengming Yin · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.