PixelProof

Inspiration

Realizing the upcoming content generation era of AI, we sought to build an early solution to detect if videos and/or images are generated by AI.

What it does

Input a video file (or YouTube video link) or image file into our website, and it will print a 'PP-score', the probability that the content is real.

How we built it

We utilize the blockchain to authenticate videos' PP-scores. When a user inputs a video, they get back a PP-score for the video and a PP-key that is uniquely linked to their video. If anyone tries to alter that video, it is no longer linked to the PP-key that is associated with the PP-score of the original video.

Challenges we ran into

Our main challenges include front/back-end communication complications and data collection for model generalization. Surprisingly, we have still managed to create a scalable infrastructure even though we didn't have the time to implement all of our features.

Accomplishments that we're proud of

We managed to train a model on hundreds of thousands of video & image data entries, that outputs our PP-score, in only 24 hours. We also architected a comprehensive encryption method using blockchain to produce our PP-keys. See github readme for more details.

What we learned

We learned and are continuing to learn the sheer importance of the application we are building and the implications it will have very soon when video generation with AI becomes extremely prevalent. We also learned how to optimize teamwork efficiency and pitch our product effectively.