Qwen-Image: Crafting with Native Text Rendering
qwenlm.github.io
GITHUB HUGGING FACE MODELSCOPE DEMO DISCORD We are thrilled to release Qwen-Image, a 20B MMDiT image foundation model that achieves significant advances in complex text rendering and precise image editing. To try the latest model, feel free to visit Qwen Chat and choose “Image Generation”. The key features include: Superior Text Rendering: Qwen-Image excels at complex text rendering, including multi-line layouts, paragraph-level semantics, and fine-grained details. It supports both alphabetic languages (e.

This allows for less distorted text rendering that doesn’t have as many typos or oddities. It’s not the highest quality image generator itself, but it handles text far better than any other model.

Chinese text:

English/combined text: