SCRAMBLE

Inspiration

Independent, online creators lack protection from data scraping systems that steal and use their content for AI training. Malicious users may use online content to create deceptive or sexual Deepfakes, violating user privacy and consent.

What it does

Our goal is to develop a Glaze/Nightshade based system that applies real time adversarial artifacts to videos and live streams. This prevents AI training on personal data while keeping content visually unchanged for human viewers.

How we built it

We built an streaming plugin that identifies facial landmarks with mediapipe, before hashing the face locations with Glaze/Nightshade, allowing user content to be uploaded online without risk of AI training.

Challenges we ran into

Learning how to prototype our project with Figma. Our computers struggled to run Figma but we built a full fledge prototype that demonstrates the image manipulation.

Accomplishments that we're proud of

Created a solution for online creators to protect their online content in a convenient, nonintrusive way.

What we learned

How to ethically balance concerns between different stakeholders. We wanted users to be protected without disrupting content online.

What's next for SCRAMBLE

Talking with regulatory bodies/ big tech companies to integrate this into online content.

Built With

• figma
• glaze
• mediapipe
• nightshade