Inspiration
As an artist and a user of the online art community, Gautham has seen the damage to and outrage of artists caused by the rise of AI art online. While many have worked to create AI tools to help artists fight the spread of AI art, we wanted to create a tool that fans of online artists could use to do their part!
What it does
AIR, true to the acronym, replaces AI images with text that lets the user know that the image that was replaced was AI generated.
How we built it
We used vanilla HTML/CSS/JS to program the browser extension, and used Python flask to build the API that the extension pings to determine whether or not an image is AI generated.
Challenges we ran into
TONS. First, we weren't able to get the model we found working in JS, and spent approximately 10 hours trying to figure out how to convert the model to ONNX to get it working independent of the package we were intending to use. However, all of us are new to ML, so this was a Herculean task. Finally, we came up with the idea to use the Flask python package to run the model on a python server and use HTTP requests to process the images. Then we ran into another huge challenge in that we couldn't read data from online images that were encoded into data URIs. We have yet to fix this issue, but we're working hard on it!
Accomplishments that we're proud of
Managing to create a functional and smooth chrome extension as well as a working Python Flask server was really amazing for us.
What we learned
We learned that AI is a very complex and deep field; We're excited to spend a much longer time looking at it in the future.
What's next for AIR - AI Image Replacer
Next up for AIR is allowing data URIs to be processed through the server and adding the option to hover over images to check if they are artificial instead of deleting them automatically.
Log in or sign up for Devpost to join the conversation.