Description as a Tweet:

Interactive Deepfake (IDF) is a digital character that talks and acts like a human. It intelligently answers questions, asks questions and captures their corresponding responses. It can represent your business, produce marketing videos, conduct tutorials and much more.

Inspiration:

There are about 590M daily users that interact with video-based content on the internet. So, it goes without saying that if you're a business looking to reach a large audience, you NEED to be making videos on the internet.
But, the reality is video creation is expensive and time-consuming.
Our solution to this problem is a web-based tool that enables your businesses to create video in a few clicks.

What it does:

Our website link is: matherium.netlify.app
Our website automates video creation. You simply select an avatar and feed in a script and our AI outputs a video of your avatar speaking your script. Due to the large amount of time it takes to train a GAN neural network, it was impossible to get our AI up and running with accurate results within the few hours we had so at this point, our project is a preliminary demonstration of what the experience will look like with a simple local video fetch.
But, over the weekend we built neural network architecture for the project so we should be able to get AI up and running within a day.

How we built it:

For the neural network arch, we created a generative neural network using tensorflow. We used Wavenet to handle Text-to-speech.
We created the website using simple html, css and js and a flask server.

Technologies we used:

• HTML/CSS
• Javascript
• Python
• Flask
• AI/Machine Learning

Challenges we ran into:

Since a GAN is notorious for being hard to train, we didn't have enough time to get our neural arch trained for accurate results within the span of time we had.
We had trouble setting up our own Text--to-speech network without using licensed papers. We decided to use Wavenet in the end.
We had some issues designing the website the way we wanted it to look to create a simple experience highlighting the demo. We had some trouble fetching a video from a flask server to play on the website.

Accomplishments we're proud of:

We're proud of having created a great website experience despite our initial challenges and bringing our exact vision to reality.
We're proud of the neural architecture we have in place that should be able to produce great results given some training and tuning time.
We're proud of how we were able to think quickly and pivot to a project that shows our vision for this product within the small time frame.

What we've learned:

We learnt a lot about TTS neural architecture and about cutting edge text-to-video generative neural architectures. We also learnt a bunch of CSS+JS tricks to make our website look nice.

What's next:

Deploying our AI to create a working and near-accurate product.
Developing a Knowledge base network that enables the deepfake to listen to your user and answer questions regarding a topic for which it was provided a knowledge base document.

Built with:

Python, Flask, HTML/CSS, JS, Tensorflow

Prizes we're going for:

• Best Software Hack
• Best Web Hack
• Best AI/ML Hack
• Best Venture Pitch

Best Web Hack

Ananth Preetham
Siddharth Preetham
Rohan Adla

Table 46 (TBL)