Alibaba’s new artificial intelligence simplifies the exact editing of images + Photos

QWen Alibaba’s artificial intelligence team from a new open source artificial intelligence model called QWen-Mimage Edit Unveiled. This powerful tool is capable of performing a range of complex edits that were previously monopolized by professional software such as Photoshop by receiving simple text commands. This model has been published in the open.

The power of QWen-Mimage Edit is rooted in his mother’s model, QWen-Mill, which was known for its unique ability to render complex texts in Chinese and English. This new model uses a clever “Dual Encoding” architecture. This means the input image is sent simultaneously to two different parts of artificial intelligence: one to understand the general meaning of the scene and the other to maintain the visual details and the original image style. This approach allows the model to perform two completely different types of editing with stunning precision.

The first type, Semantic edits That is where the general concept of the scene changes, but the main identity of objects is preserved. For example, users have been able to rebuild a photo of Manhattan in the style of Lego toys or take a picture in the style of studio studio. The second type, Appearance edits That focuses on accurate and local changes. In this case, the bulk of the image remains intact and only specific details change. Examples include removing a strand of hair from a portrait, adding a promotional panel with precise reflection in the water, or changing the color of a letter in a textual image.

Alibaba’s artificial intelligence for professional editing of images

One of the most prominent features of this model is the ability to edit Chinese and English text inside images. Users can easily add new texts to posters or boards, delete or change existing texts, while maintaining the main font and style. This feature has been proven especially in editing calligraphy works that require high accuracy.

This model has been released in an open source and was licensed by Apache 4.0. This means that companies and developers can download it for free and install it on their hardware, which can save a lot of software costs. In addition, the QWen-Midage Edit has been publicly available through various platforms such as QWen Chat and Hugging Face or Alibaba Cloud API (at $ 1.5 per image).