Adobe shared its powerful Content-Aware Fill as a sneak peek back in 2010, and the feature has become a powerful tool in the Photoshop arsenal. Now the company is teasing a new technology called Deep Fill, which takes the idea to yet another level using AI techniques.
Basically, Deep Fill tries to understand the content in your photo so that the resulting fill doesn’t only look visually realistic — it should make sense too.
“Deep Fill is a new deep neural network-based image in-painting system that (1) learns to generate image patches that are visually realistic and semantically reasonable, and (2) allows interactive customization of results based on users’ brushing and sketch inputs,” Adobe says.
For example, when removing bandages from a person’s face, Content-Aware Fill won’t hesitate to give the person a third or fourth eyeball. Deep Fill, on the other hand, understands what a face should look like and properly fills in the areas with appropriate fills to keep the face looking like a human face:
For an outdoor scene, Deep Fill can instantly remove people from the scene without messing up the natural landscape in weird ways:
Deep Fill also has a trick up its sleeve: give it a sketch of what you’d like the fill to look like, and it’ll do its best to satisfy your requirements in a realistic way:
Just as with Adobe’s new sneak peeks for Cloak (Content-Aware Fill for video), Scribbler (auto-colorizing black-and-white images and sketches), and Scene Stitch (Content-Aware Fill with stock photos), there’s no word on if or when we’ll be seeing Deep Fill added to Photoshop.