The company Alibaba introduced the updated open AI model Qwen-Image version 2512. According to blind tests on the AI Arena platform, this model became the leader among open solutions in its class. Qwen-Image received improvements in reproducing photorealistic faces, as well as in creating text and data without noticeable artifacts.
The model can draw complex diagrams, posters, and presentation slides, accurately reproducing letters. It recognizes detailed queries with many parameters and accurately executes the specified composition. Users can test Qwen-Image for free in the Qwen chat .
To launch the model in the ComfyUI graphical interface, only 14 GB of video memory is now required. Developers have prepared a detailed guide and provided GGUF weights for working with the model. A demo version of the model is available on the HuggingFace platform.
