Restoring and improving old photos is now available to everyone.
Google has been actively promoting generative artificial intelligence technologies over the past 18 months, and one of the most notable developments has been the Magic Editor feature . Now the company has introduced a new significant feature for editing photos based on AI — RealFill.
Last week, Google quietly filed a trademark application for a technology called RealFill. The application was filed with the European Intellectual Property Office (EUIPO) and the US Patent Office (USPTO).
The brief description of the trademark refers to "providing non-bouncy software using artificial intelligence for retouching images; providing online non-bouncy software for creating generative models".
RealFill technology was first introduced late last year in an article and on the website of a group of researchers from Google and Cornell University. The article titled "Link-based Generation for Authentic Image Augmentation" describes a method that allows you to more accurately expand and retouch existing images using up to five reference images.
These reference images do not necessarily have to match the target image and can be taken from different angles, under different lighting conditions, apertures, and image styles.
The team first sets up a personalized generative AI model on reference and target images. This process allows the model to learn the lighting, style, and content of the scene in the images.
The results, as seen in the examples above and below, are impressive. The images below also show how RealFill images compare to other solutions, such as Stable Diffusion.
However, the team noted some limitations of RealFill. One of the disadvantages is the need to perform "gradient adjustment" on the input images, which slows down the process. It can also be difficult to recreate the scene in the final image if there is a large difference between the reference images and the target image. In addition, it was found that text can become a problem when using this technique.
Will this feature appear in the Pixel 9 or Google Photos? Proprietary trademarks do not guarantee commercial implementation of RealFill, but there is reason to believe that it may appear in future models of Pixel phones or in Google Photos, if it is planned for commercial release.
This will probably be a cloud-based feature, rather than an on-device photo editing option, given the slow setup process.
Existing solutions for expanding and retouching photos are far from perfect, so using reference images can significantly improve your results. This also means that users can go back to old snapshots in their Google Photos library and create better images.
Either way, this feature is likely to raise even more questions about what a photo is, similar to what has already happened with Google's Magic Editor.
Google has been actively promoting generative artificial intelligence technologies over the past 18 months, and one of the most notable developments has been the Magic Editor feature . Now the company has introduced a new significant feature for editing photos based on AI — RealFill.
Last week, Google quietly filed a trademark application for a technology called RealFill. The application was filed with the European Intellectual Property Office (EUIPO) and the US Patent Office (USPTO).
The brief description of the trademark refers to "providing non-bouncy software using artificial intelligence for retouching images; providing online non-bouncy software for creating generative models".
RealFill technology was first introduced late last year in an article and on the website of a group of researchers from Google and Cornell University. The article titled "Link-based Generation for Authentic Image Augmentation" describes a method that allows you to more accurately expand and retouch existing images using up to five reference images.
These reference images do not necessarily have to match the target image and can be taken from different angles, under different lighting conditions, apertures, and image styles.
The team first sets up a personalized generative AI model on reference and target images. This process allows the model to learn the lighting, style, and content of the scene in the images.
The results, as seen in the examples above and below, are impressive. The images below also show how RealFill images compare to other solutions, such as Stable Diffusion.
However, the team noted some limitations of RealFill. One of the disadvantages is the need to perform "gradient adjustment" on the input images, which slows down the process. It can also be difficult to recreate the scene in the final image if there is a large difference between the reference images and the target image. In addition, it was found that text can become a problem when using this technique.
Will this feature appear in the Pixel 9 or Google Photos? Proprietary trademarks do not guarantee commercial implementation of RealFill, but there is reason to believe that it may appear in future models of Pixel phones or in Google Photos, if it is planned for commercial release.
This will probably be a cloud-based feature, rather than an on-device photo editing option, given the slow setup process.
Existing solutions for expanding and retouching photos are far from perfect, so using reference images can significantly improve your results. This also means that users can go back to old snapshots in their Google Photos library and create better images.
Either way, this feature is likely to raise even more questions about what a photo is, similar to what has already happened with Google's Magic Editor.
