Qwen releases open model for text-in-image generation

The Qwen team at Alibaba has released Qwen-Image, a new open-source model for generating images from text prompts. Released under the permissive Apache 2.0 license, the model aims to solve one of the most common frustrations with AI image generation: rendering legible text.

While many popular diffusion models struggle to create coherent letters and words, Qwen-Image is specifically trained to produce readable text within its creations. The model demonstrates a strong capability for rendering characters in both English and Chinese, a significant challenge given the complexity of the scripts.

This focus on typography is a practical step forward for generative AI. The ability to reliably create images with accurate text opens up new possibilities for designers, marketers, and developers building tools for ad copy mockups, product designs, or social media content where text is a critical component.

The model is based on a standard diffusion architecture, pairing a text encoder with an image generator to interpret prompts. The full model weights and usage instructions are available for download on the Qwen team's Hugging Face repository.