Introducing 'Qwen-Image-Edit,' an image editing AI that can redraw characters in different situations while maintaining their original appearance, and also allows for redrawing of characters and 'rotation of subjects.'



The AI development team at Chinese IT company Alibaba has announced

'Qwen-Image-Edit,' which expands image editing tasks based on the image generation AI ' Qwen-Image, ' allowing it to create different scenes while maintaining the original appearance of characters.

Qwen-Image-Edit: Image Editing with Higher Quality and Efficiency | Qwen
https://qwenlm.github.io/blog/qwen-image-edit/





Alibaba's AI development team has produced the large-scale language model 'Qwen' series and the image generation AI ' Qwen VLo .' On August 4, 2025, they announced 'Qwen-Image,' which can accurately draw text, a task that existing image generation AIs such as OpenAI struggle with.

Introducing the image generation AI 'Qwen-Image,' capable of generating high-quality images exceeding those of OpenAI and Flux, and demonstrating phenomenal text drawing performance that can naturally depict 'multiple lines of kanji' - GIGAZINE



'Qwen-Image-Edit' is an extension of 'Qwen-Image' for image editing tasks, and its three main features are 'semantic editing and appearance editing,' 'accurate text editing,' and 'powerful benchmark performance.'

Semantic editing is a feature that allows users to edit a scene without altering the appearance of the input image. An example using Qwen's mascot, the capybara, has been published. The capybara on the far left, under 'Input Image,' is the capybara entered into 'Qwen-Image-Edit.' Various elements have been added to the image via prompts, and the pose has been changed, but the capybara's basic appearance has been maintained. The 'Qwen' lettering on the shirt has also been accurately depicted.



This is a depiction of a capybara using prompts based on the 16 personality types of

the MBTI .



Qwen-Image-Edit can also rotate objects 180 degrees, not just 90 degrees. Looking at the example, you can see that the rotation is possible without any distortion of clothing or hairstyle.



You can also input a portrait and change it into various styles.



Appearance editing allows you to add, remove, and modify specific elements. Below is an example with a sign added.



Remove any trapped hair.



In the example below, only the letter 'n' has been changed to blue, while the rest remain the same.



Background change.



Change clothes.



It is also good at handling text, which other image generation AIs struggle with, and does not break down even if the contents of the string are rewritten.



The poster was edited, not only changing the text but also changing the woman to a man at the end.



Even with vertical calligraphy images, you can specify the areas to be corrected and edit them.



You can also edit the photo by putting it in a frame and displaying it in your room.



The Qwen team commented, 'We hope that Qwen-Image-Edit will further advance the field of image generation, lower the technical barriers to visual content creation, and inspire even more innovative applications.'

The model data is available on Hugging Face.

Qwen/Qwen-Image-Edit · Hugging Face
https://huggingface.co/Qwen/Qwen-Image-Edit

Related Posts:

in Software, Posted by logc_nt