In the new Photoshop, AI adds and removes objects like never before.
Adobe has launched a series of new features for Photoshop that allow you to automate and simplify many of the most repetitive tasks in digital image editing, from adding photorealistic elements to removing objects and people , to enlarging low-resolution images without losing quality.
The new features, available on desktop, web, and mobile versions of Photoshop, are the result of analyzing feedback gathered from the creative community over the past few months. "We've identified the key bottlenecks in the creative process," Adobe explains, "those moments when inspiration gets lost amidst the more laborious technical operations."
The new features' target audience is broad, justifying the multiplatform approach. They range from designers developing assets for advertising campaigns to photographers busy retouching details, to content creators producing social media materials directly from their smartphones. The stated goal is to reduce the time spent on repetitive tasks and leave more room for creativity, but readers and a large portion of the professional public remain concerned about those who have built their careers around the ability to perform these tasks.
Harmonizing with AIThe most interesting and surprising feature introduced by Adobe is called Harmonize : it allows you to integrate one or more external elements into an image simply by overlaying the desired cutout. The Adobe Firefly generative model intervenes by correcting part of the perspective and the color gradations of the cutout to realistically fit it into the composition. The user simply chooses the best result from those proposed by the Artificial Intelligence.
In the examples provided by Adobe, the system is used, for example, to insert a flamingo floating in a swimming pool , or a piece of furniture into the image of a room.
The potential of such a feature for marketers and advertising agencies is enormous, as it would dramatically reduce the time it takes to produce promotional images for catalogs and campaigns, for example. The feature is currently available in beta (but seems to already work very well) on desktop Photoshop and can be tested in Early Access in the iPhone app.
Non-forced removalThe tool that does the exact opposite of Harmonize— removing objects, people, or background elements from images—has also been improved. With the latest update, the system works much better: the model generates more realistic content that integrates seamlessly into the context of the composition.
With the Photoshop update, Adobe also introduced the first beta version of an upscale tool, which allows you to enlarge images without losing quality . It's one of the most requested features by the community and among the most popular AI-based features provided by competitors like Topaz Labs. Photoshop's new upscale feature can improve the quality of images up to 8 megapixels in resolution . It can be used in the new Photoshop beta, but you can also try it on the web for free by uploading a photo directly to a mini-site set up specifically by Adobe .
In this case, the target audience is creatives who need higher-resolution images, perhaps for print. This feature, like the others, is based on Adobe's generative models and therefore could create some problems for some types of photographers, particularly in journalism and agency environments.
Because models, no matter how advanced, can “hallucinate” by inserting small, non-existent yet invisible elements , many photojournalism organizations do not accept photos “enhanced” with these tools.
The World Press Photo, the world's most famous photojournalistic award, for example, categorically excludes the use of these tools, specifically citing Adobe and Topaz Labs among the suppliers.
“Tools that violate the contest rules and are not eligible include any AI-based generative image enlargement systems, such as Adobe Super Resolution or Topaz Photo AI,” the rules read . “These tools rely on generative models that introduce new information to enlarge or sharpen images .”
While generatively upscaled images appear identical, but only improved, the models tend to insert so-called "foreign pixels" and portions of the photo that are generated based on the training. They are not a physical, software-based improvement of the pixels already existing in the image, as is the case, for example, with the upscaling systems of some cameras ( such as the Canon EOS R5 and R1 ) that are based on deep learning.
Projects and model selectionAdobe has also introduced a new Projects feature in beta, designed to better organize creative assets, especially for those working collaboratively with other professionals. Finally, the ability to select the generative model to use for AI functions has also been added. The new "Gen AI model picker" lets you choose between Firefly Image 1 and Firefly Image 3 for generative fill and generative image expansion. These two tools allow you to enlarge or fill portions of photos with additional elements "designed" by artificial intelligence based on the original image.
repubblica
