Qwen-Image-Edit is a powerful new image-editing model released by Alibaba’s Qwen Team in August 2025. Built on a 20B-parameter Qwen-Image foundation, it brings advanced capabilities in both semantic and appearance editing thanks to a dual-encoding architecture: one branch (Qwen2.5-VL) controls high-level visual semantics, while a VAE encoder handles low-level appearance details.
The model excels at precise bilingual text editing (English and Chinese), preserving font, size, and style when adding, modifying, or deleting text in images. It supports high-fidelity semantic transformations like style transfer, object rotation (including novel view synthesis), IP generation, as well as localized appearance tweaks such as removing fine details or changing colors without affecting surrounding areas.
Qwen-Image-Edit is accessible via platforms like Qwen Chat, Hugging Face, ModelScope, and Alibaba Cloud’s API, where developers can integrate image editing into their applications with straightforward HTTP or SDK calls. Its open-source Apache 2.0 license ensures enterprise-grade adoption and potential cost savings compared to proprietary tools.
