Alibaba launched Qwen-Image-Edit on August 19, expanding its 20B Qwen-Image model into the field of image editing. The system is designed to deliver both high-level semantic changes and fine-grained visual edits by feeding input images into Qwen2.5-VL for semantic control and a variational autoencoder (VAE) encoder for appearance control.

The company says Qwen-Image-Edit is built on three pillars. First, it enables semantic and appearance editing, handling tasks that range from rotating objects and applying style transfers to removing fine hair strands or inserting objects with accurate reflections. Second, it introduces bilingual text editing in Chinese and English, allowing users to add, delete or modify text while preserving font, size and style. Third, benchmark evaluations indicate state-of-the-art results across public image editing datasets.

Demonstrations showcased a wide range of use cases. Qwen’s capybara mascot was used to generate emoji packs while preserving character consistency across edits. The system rotated objects by 90 and 180 degrees and applied Studio Ghibli-style transfers to portraits, pointing to potential uses in avatar creation and digital art. 

On the appearance side, it changed individual letters’ colours, swapped backgrounds, adjusted clothing, and removed small details without disrupting the rest of the image.

Text editing was highlighted as a standout capability. In one demonstration, the model progressively corrected errors in a generated version of the Chinese calligraphy artwork using chained edits until the text was fully accurate.

One can try it out on Qwen’s chatbot portal or explore more about it on Hugging Face.

With Qwen-Image-Edit, the company is positioning itself in a competitive field that includes Google’s Gemini image editing capabilities. It says that the new system will lower barriers to visual content creation while opening possibilities for design, avatars and creative applications.

The post Alibaba’s Qwen-Image-Edit Brings Precision Text, Visual Editing to AI Images appeared first on Analytics India Magazine.