Midjourney revolutionises photo editing: AI will soon edit even real images

Until now, Midjourney, one of the most popular text-to-photo (text-to-photo) generative Ai, only offered the possibility of starting with an idea, expressed in a textual description, which was then 'translated' by the algorithm to create hyperrealistic photos. Soon, thanks to a new feature announced by CEO David Holtz via a message on the official Discord server, it will be possible to edit existing images, even if not created by an AI. This is a significant innovation, as no other tool had yet offered such a possibility.

Although there are already tools on the market that allow retouching of Ai-generated images, these are mainly limited to works generated by the system itself. Midjourney breaks this pattern, offering the possibility of working on pre-existing images. This is a significant development, not only for the technological aspect - it requires a deep understanding of the image on the part of the system - but also for the impact it could have on image manipulation. In an era where deep fakes are a growing threat, the possibility of altering images of real people without consent raises ethical and legal concerns.

Holtz revealed that the heart of this update is 'depth controlnets', a technology that allows AI to intervene on the shape and texture of an image based on simple textual descriptions. It will be possible to upload any image from the web and customise it: enlarge or shrink elements, adjust angles, and even make complex changes such as replacing materials or surfaces of objects in the photo.

Once the image is uploaded, users will have access to various tools. It will be possible to enlarge or reduce the image, change the perspective and work directly via the Midjourney website, without the need to go through Discord. One of the most popular additions is the digital brush, which allows precise editing at specific points in the image. One of the most innovative functions is 'Retexture', which allows you to change colours and surfaces without altering the structure of the image. Imagine transforming a black leather jacket into a red silk one while retaining its shape. This new ability to understand the image opens the way to a wide range of creative and professional uses. For example, designers, photographers or advertisers could perfect details in their images without resorting to complex software such as Photoshop.

Midjourney is already aware of the ethical implications and has started to implement metadata in AI-generated images to facilitate their identification and prevent misuse. However, it has not yet adhered to the broader C2PA standard, which offers a more comprehensive approach to certifying the origin of images. Holtz stated that access to the new functionality will initially be limited to a small group of users, with human moderation to monitor the results. This 'beta' approach will allow the implications of the technology to be better studied before a large-scale release. "We are not yet sure how best to handle the limitations of this feature," the CEO explained. "We want to gather feedback from users to find the right balance between accessibility and control." A moderate approach, which should serve to further refine security measures against possible abuse. To date, the company already adopts the Digital Source Type standard to mark images created by Ai, but many argue that these measures alone are not enough to stop the spread of manipulated content, such as deepfakes or visual forgeries. Certainly, with more and more companies facing similar issues, Midjourney's decisions could influence not only competition but also the definition of new regulations in the sector.