CTRL-ALT-DECEIVE

Inspiration

The accelerating advancements in AI image generation are creating a world where visual deception is becoming alarmingly easy. The once distinct Venn diagram of the physical and digital is rapidly collapsing into a near-perfect circle, making it incredibly difficult to distinguish between authentic media and sophisticated fakes. AI can now generate convincing images of non-existent medical professionals promoting false cures, create deep fake videos of political figures spreading damaging lies, or fabricate compelling visuals of environmental disasters to manipulate policy debates. This project is driven by the urgency to address this growing issue, which threatens to diminish public trust in information.

What it does

CTRL-ALT-DECIEVE is a proactive Google Chrome extension that is designed to rebuild your trust in online visuals. It operates discreetly in the background to perform real time analysis on images leveraging AI to identify potential AI generated content. With CTRL-ALT-DECIEVE, you can feel more confident in the authenticity of the information you consume, leaving no room for doubt.

How we built it

CTRL-ALT-DECIEVE is a combination of a fine tuned Convolutional Neural Network - MobileNetV3Large, with a lightweight JavaScript frontend, seamlessly integrated into a Chrome extension. Using javascripts capabilities, the model runs during page loads, as images are loaded into chrome they get checked by our model for AI content, all in real time, directly in the user’s browser

Challenges we ran into

One of the biggest challenges we faced in executing our project was the scarcity of a comprehensive dataset. There were plenty of AI generated face datasets for deepfake detection, but a dataset with a good range of AI generated images proved to be difficult. And on top of that any detailed datasets, diverse datasets were often 50 to 60 GB which was far too large for us to clean and train in the short time frame of the project. Ultimately we decided to collect data from various sources to create a dataset that met the project requirements.

Accomplishments that we're proud of

Machine learning models are often tough to interpret the unexpected failures in both data sets and models were not a dead end.While the hurdles were disheartening at times our team leveraged them as learning opportunities that lead us down the path to this project. Of course tight deadlines like these raise stress but seeing every step through and sticking to the plan kept us together and we are very proud of the work we have done and the final product we have created!

What's next for CTRL-ALT-DECEIVE

The Google Chrome extension is just the start of CTRL-ALT DECEIVE, our vision extends beyond just a single browser. Our next focus would be to expand availability to iOS and Android mobile devices allowing users to benefit from real time verification of content authenticity across a wider range of environments, from quick news searches to videos on TikTok.