True Sight

Inspiration

With the explosion of AI-generated content across platforms like TikTok and YouTube, it’s becoming increasingly difficult for users to distinguish between real and AI-created videos. From deepfakes to synthetic voices, this content often spreads without transparency, creating risks around misinformation, trust, and content authenticity.

We wanted to build a tool that helps people instantly identify AI-generated videos—a small but meaningful step toward digital transparency.

What it does

Our app analyzes video and/or audio content using AI detection models to make that call. The interface is clean, minimal, and accessible—users simply input the link/share the link, and we handle the rest.

How we built it

For the frontend, we implemented a clean and minimal interface using HTML, CSS, and JavaScript to ensure simplicity and ease of use. On the backend, we developed a service that downloads videos from URLs (such as TikTok) and processes them using TensorFlow. A pre-trained model performs frame-by-frame analysis to determine whether the content is AI-generated, based on multiple criteria. The system then calculates a confidence percentage and provides a final verdict along with a detailed explanation of the reasoning behind the classification.

Challenges we ran into

-- Expo Android Development: Working with Expo was convenient for rapid development, but posed limitations when handling native Android features like file access and sharing between apps.

-- Connecting Frontend to Backend: Ensuring smooth communication between the frontend and backend, especially for file uploads and handling CORS issues, required careful setup and testing.

-- Legality: We had to consider the ethical and legal implications of downloading and analyzing videos from public platforms, making sure our use stayed within fair use boundaries.

What we learned

-- Using Tensorflow and not relying on AI Wrapper project this time. All jokes aside we learnt a lot more about hosting backend/frontend and how to effectively write code that can be built easily by these tools. A lot of debugging.

What's next for True Sight

-- AI models are forever growing, that being said TrueSight will always need to keep growing and analyzing more of a variety of things to make sure fake information is easily distinguishable.

-- Add further support for apps such as Snapchat, Linkedin, Facebook etc.

-- Making it into a proper app, as of right now TrueSight is in a website view only, (you can add it as a app on Android), but we would like it to be a proper app that can be shared to and that can send notifications.

-- Nicer UI/UX, adding some more animations/pages to make the website feel more interactive.