MIT's AI can create images, even edit objects inside them
The folks at MIT have come up with a new AI tool, one that creates images and programmatically edits objects inside them in real time. Normally, the task of creating and editing an image revolves around manually producing digital renders and using tools like Photoshop, which takes a lot of time. However, the new program, dubbed GANPaint Studio, handles almost everything instantly. Here's how.
GANPaint adds, removes objects from images
Available for free, GANPaint Studio generates realistic-looking images from scratch and offers an easy way to add, remove objects inside them. If you want to add an object, say a tree, into a shot, it analyzes the photo and places a tree matching the scene on the chosen spot. Similarly, it removes objects from a photo without spoiling the scene.
AI learned context for placing objects
At the heart of GANPaint Studio lies a Generative adversarial network (GAN), which involves a set of neural networks competing against each other. They use internal reasoning to learn the relationship between objects and uses it to create the images. It ensures that the system places things like clouds in the air and grass on the ground.
Future possibilities with GANPaint
Now, the researchers plan to evolve the GANPaint system and use it for editing videos with more ease. They say the system could prove handy in adding or removing objects - like left out props - into films. Plus, they also think the same tech could also be repurposed to detect fake content in the same way.
You can try GANPaint right away
GANPaint Studio is now available as a free demo for users to try. It will be officially announced at a conference next month, ultimately leading to the full release.
