Fardeen Bablu | November 7, 2022
Imagine this: You’re out one-night having fun with friends. You take a few pictures, and they look great! The next morning, however, you realize that they don’t look quite right. What can you do?
Fortunately, a Google research group has developed a tool, called Imagic, which allows users to easily edit images. By inputting a single image and text describing the change you want to see, the tool can then output a changed image. The best part? No weird, unintended changes occur to the image. It just follows the given text prompt.
Initial Stages
The research group, which includes people from Technion and Weizmann Institute of Science, was investigating ways that current AI-Art tools could be innovated upon.
The growth of AI and image modification has been staggering in recent years. From industrial AI-Art from OpenAI, to widespread resources from the likes of labs like MidJouney, AI-Art has never been more accessible than right now. However, many tools we know today struggle with text modification, that is, changing an image through AI with text inputs.
The group decided to simplify the process by taking one input image and text. This simplification allows Imagic to meaningfully change the output image with much distortion.
From: https://arxiv.org/pdf/2210.09276.pdf
Societal Impact
Outside of industrial or personal use, the research group also highlighted some societal concerns with Imagic. Already, AI-Generated images have been used to create false social media accounts, resulting in an increase in misinformation. The group strives to create an identification system for Imagic images to help prevent this.
Considering these concerns, it will take time to release Imagic in the commercial space. Hopefully, someday, we’ll all be able to artificially edit our less-than-stellar images to our liking.
Interested in learning more? Check out the research below: