Qwen-Image
lastmod 2026-02-12created 2025-08-053 edits
Image generation model with native text rendering
About
- Qwen-Image is a 20B-parameter MMDiT image generation model with strengths in native text rendering and precise image editing
- Achieves high accuracy and visual quality in rendering complex characters in various languages, including alphabetic scripts and Chinese characters
- Achieves best-in-class performance on various public benchmarks (GenEval, DPG, OneIG-Bench, etc.), with excellent text generation capability as well
- In real demos, accurately renders complex layouts and diverse styles such as multilingual text, posters, PPT slides, and illustrations
- Supports editing features such as style transfer, adding/removing objects, detail enhancement, and pose changes, and aims to expand the open-source ecosystem
See also
- Qwen
- Qwen-Image - Image generation model with native text rendering
- Qwen-Image-Edit - Release of a dedicated image editing model