Alibaba claims new Qwen Image 2.1 AI model beats Google Nano Banana 2.0 with minuscule 7B parameter model
Alibaba Group's AI division has released Qwen 2.1 Image, an ultra-lightweight image generation model with just 7B parameters, able to run on local hardware like the RTX 3090.
Alibaba has unveiled a new compact image generation AI model, Qwen Image 2.1, boasting only 7 billion parameters. This lightweight model can run on older consumer graphics cards such as the RTX 3090. Despite its small size, Alibaba claims Qwen Image 2.1 surpasses Google's Nano Banana 2.0 in terms of capability, according to their internal benchmark.
Notable features of Qwen Image 2.1 include native transparency support, enabling the creation of images with transparent backgrounds. This is particularly beneficial for artists working on various projects or for print-on-demand products like stickers. Additionally, the model can handle up to 10 reference images, improving upon the earlier version.
However, a new licensing agreement has raised eyebrows. Unlike previous versions, Qwen Image 2.1 explicitly prohibits commercial resale of the model, requiring users to obtain a separate license from the developer if they intend to use it commercially. This has sparked debate within the AI community.
Despite the licensing change, Qwen Image 2.1 appears to be a strong competitor among open-weight models. It has demonstrated impressive performance on local hardware, with users reporting successful generation of 1MP and 2K resolution images on RTX 3060, 5070, and 5080 graphics cards. The model's compact size and efficient processing make it an attractive option for those seeking rapid, high-quality image generation on their local devices.
Written by urgent.news from Tom's Hardware's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
