Adobe has detailed a research project in which it has developed an artificial intelligence capable of generating realistic colored images from a sketch. This isn’t the first time we’ve seen an AI able to generate realistic images, but they generally lack an element of control that lets the user dictate the final results. Adobe’s Scribbler is different.
The technology was recently detailed by Adobe Research, where researchers pointed out that existing deep convolutional networks have been used to create imagery, but that the existing methods ‘are of limited utility’ due to the user’s inability to control whatever the AI ultimately produces. Adobe’s new project is different.
The goal here is to create realistic images, ones where the user can mostly control what the final image will look like. To achieve this while having the AI do most of the work, Adobe’s project depends on an existing sketch of the eventual final image. Users are able to make the sketch detailed, then add small bits of color to different components.
Using the combination of sketched lines and inserted colors, the AI is able to generate realistic final images and colorize the image automatically while taking cues from the original partially colorized sketch. Users are able to edit the image and see the results of those edits in real time, the researchers explain.
Given that this is a research project, you’re not going to have access to the technology yourself…at least not yet. It is possible that we could see a feature that utilizes this type of tech in a future Adobe product, such as Photoshop, though no such plans were stated. You can view the entire paper for free here.
SOURCE: Adobe Research