Add, Edit and Delete Objects with One Click Using AI

Image editing software has had major advancements in recent years. However, using an AI created by researchers from MIT and IBM, it is now even easier to modify pictures to your heart’s content, with the ability to add, edit and remove objects with just a single click.

This new instrument, named Generative Adversarial Network (GAN) Paint Studio, allows the modification of uploaded images while preserving its other original details with near-perfection. Adding a tree, for example, will cause related objects to be automatically rectified to make the finished image look more realistic.

Adding grass in front of a building using GAN Paint Studio.

Adding grass in front of a building using GAN Paint Studio.

Antonio Torralba, a professor at MIT and member of the MIT-IBM Watson AI lab, stated the software has a “strong personality” and will only perform an action if it deems it would fit the image (and doesn’t if it thinks it won’t):

All drawing apps will follow user instructions, but ours might decide not to draw anything if the user commands to put an object in an impossible location. It’s a drawing tool with a strong personality, and it opens a window that allows us to understand how GANs learn to represent the visual world.
— Antonio Torralba

Firstly, the researchers tested the tool by deleting specific objects and noting how artefacts around the deleted object were affected by it. Next, they used this information to train the AI algorithm to make changes when an object is added/deleted to make the image look realistic.

This was done using a two-part network, which used generators to make samples of realistic images and discriminators to discern the differences between generated and “real” pictures. The discriminator gives feedback which is then utilised to change the generator architecture to improve its ability at making realistic pictures.

The team acknowledges their creation could be used to maliciously manipulate pictures; however, they argue the tool will assist in understanding what kind of errors AI makes in relation to image modification, and hence allowing “deepfakes” to be exposed for the falsification that they are.

Try it out for free here.