unfakeable

UNFAKEABLE.PHOTO

Inspiration

Sharing photos on social media is more common than ever, but it is also becoming increasingly unsafe. Deepfake fraud increased by 1,740% in North America in 2022, and more than 10% of companies have reported attempted or successful attacks using deepfake fraud. In February 2023, popular Twitch streamer QTCinderella had her likeness stolen and turned into highly nonconsensual deepfake pornography. A comprehensive report into deepfakes in 2023 determined that deepfake pornography constitutes 98% of all deepfake videos found online. Worse still, 99% of those targeted by deepfake pornography are women. Introduced federal legislation on this issue includes: the DEEPFAKES Accountability Act (introduced Sept. 2023), the Preventing Deepfakes of Intimate Images Act (introduced May 2023), and the NO FAKES Act (introduced Sept. 2024). However, none of this proposed legislation has been passed to protect individuals and their likenesses from being reproduced nonconsensually using AI. Seeing these violations in real-time inspired us to care about this explicit attack on women (the numbers don't like). It also made us realize that if this can happen to celebrities and CEOs with many resources to fight back, then everyday people are very vulnerable and the consequences to their careers could be even more tragic. When asked about her relationship with deepfakes Johansson famously said: "The fact is that trying to protect yourself from the internet and its depravity is basically a lost cause," but with Unfakeable it's not a lost cause at all! That’s why we set out to create an app that protects users from having their photos turned into deepfakes, and that uses the amazing toolbox that is AI for good.

What it does

It only takes seconds to proactively protect yourself from a deepfake attack! Before posting your latest pic, use our tool to add noise. It blocks deepfake attacks while keeping your pic HD! Stay in control and stop attacks before they start! Steps:

Head on over to unfakeable.photo and upload your latest cute pic!
Wait seconds for your protected image to be generated.
Once it loads just right-click, "save image," and now you're all set to post that latest selfie worry-free!

Challenges we ran into

Challenge 1: (adding noise and it not being too blurry) Challenge 2: (moving from getting our app to run locally to it loading on Streamlit) Challenge 3: We had a really hard time redirecting the domain that we picked out to our Streamlit app. Our original approach was to add a CNAME record in Porkbun to point our domain to our Streamlit app's URL directly, but this was unsuccessful because the redirect did not resolve properly when mapped with a CNAME from a custom domain. We also tried ALIAS and HTTP DNS records at one point, which resulted in no success. The solution we found was to switch to GitHub Pages where we are now hosting an intermediary page. We now have AAAA DNS records in Porkbun that point our domain to GitHub's IP addresses which connect to our intermediary page. Our intermediary page has an HTML file that redirects visitors to our Streamlit app's actual URL.

Accomplishments that we're proud of

Successfully preventing deepfakes while preserving image quality.
Creating a powerful and novel machine learning pipeline to solve an active problem.
Successfully navigating a really tricky domain redirection issue after using multiple different methods before finding one that worked and learning all about DNS Records in the process!

4. Creating a working tool that actively empowers and supports women! Unfakeable is more than just a functional tool and our HackViolet project, it's the first free deepfake protection software available online.

What we learned

We learned a lot about web design, the dangers of deepfakes, and modern machine-learning techniques. We also learned about DNS and redirects, version control from our adventures with GitHub, and how to deploy a website using Streamlit.

Our Favorite Parts of the Process

Emmit: Getting crazy results at the start when our models went haywire. lol Lizzie: Jumping up and down and feeling super accomplished when we finally got to access our app when we searched our domain! So awesome!

What's next for Unfakeable

We hope to scale the project, making it more robust against deepfake tools and able to support more users. We also hope to educate people on the emerging dangers brought about by AI, along with how AI can be used for good.