Alibaba has just introduced Qwen-Image-Edit, its image editing tool powered by the Qwen-Image 20B model. This new development incorporates both editing and image variation capabilities, as well as advanced text insertion or modification, providing precise and consistent results.
Another major advantage of Qwen-Image-Edit is that it may be used free of charge through the Qwen Chat platform by selecting the “Image Editing” function.
According to the information shared by Alibaba, “Qwen-Image-Edit successfully extends the unique text rendering capabilities of Qwen-Image to image editing tasks, enabling precise text editing. Furthermore, Qwen-Image-Edit simultaneously inputs the image into Qwen2.5-VL (for visual semantic control) and the VAE encoder (for visual appearance control), thereby enabling both semantic and appearance editing capabilities.”
Visual appearance and visual semantics refer to two different aspects of image and text editing. Editing visual appearance pertains to actions such as adding, removing, or changing elements within the image, whereas editing visual semantics involves alterations like changing styles or rotating objects.
Now that we have covered the technical details, let us move on to what is truly enjoyable: discovering its capabilities!
Until relatively recently, creating variations of an image while maintaining the consistency of characters, objects, or environments remained an unresolved challenge for AI image generation tools. However, the rapid advancement of this technology has managed to overcome this barrier, as demonstrated by models like Midjourney and GPT-4o, and now, Qwen-Image-Edit.
Alibaba’s AI image editing tool exhibits outstanding performance in this regard, as it is capable of generating variations that preserve the desired characteristics while modifying other parts of the image. The company itself shared this example, in which Qwen-Image-Edit created several versions of Capibara, the mascot of the Qwen models, performing different activities.
Alibaba explains: “although most of the pixels in the edited image differ from those in the input image (the leftmost image), the consistency of the Capibara character is perfectly maintained. The powerful semantic editing function of Qwen-Image-Edit allows for the simple and diverse creation of original IP (intellectual property) content.”
One of the major improvements integrated into Qwen-Image-Edit is its text editing capability. This feature allows it to add, remove, and modify text while preserving the original font, size, and style. The only limitation to note at this time is that it currently functions correctly only for text in Chinese and English.
The editing capabilities of this tool also allow it to rotate objects and characters, making it possible to create images from new perspectives, showcasing their content and main subjects in a different light.
Of course, Qwen-Image-Edit is also capable of applying new artistic styles to an image while preserving its content. You simply need to specify the style or texture you would like your new creation to have, and the AI will generate it for you.
Similarly, you may also instruct the tool to modify the appearance of a specific element within the image. For example, I first generated an image of a cartoon-style red panda surfing using the “Image Generation” function, and then applied the “Image Editing” function to transform only the red panda, giving it a plush texture.
Another feature provided by Qwen-Image-Edit is the ability to add or remove objects and characters from images. It achieves this while keeping the rest of the artwork unchanged, which is extremely useful. In addition, this capacity may be applied to more complex scenarios, such as costume changes.
In addition, the tool also allows you to modify environments and change the backgrounds of images, thereby placing the main elements or characters in new settings or contexts.
Photo: Images generated with Qwen-Image-Edit
Your email address will not be published. Required fields are marked *
Δ