A ReactJS application made for NUS Orbital 2021. Here is the link for the app deployed on Vercel: AnyTutor.
In the project directory, you can run:
Runs the app in the development mode.
Open http://localhost:3000 to view it in the browser.
The page will reload if you make edits.
You will also see any lint errors in the console.
Since the Covid-19 pandemic started, many schools and tuition centers have shifted their lessons online. However, such online lessons are notorious for being less engaging than face-to-face classes, and coupled with many student’s tendencies to get distracted easily, these lessons usually end up being less productive than they should be. Moreover, there is an evident lack of teachers to provide each student sufficient attention during lessons.
As such, we aim to reduce the burden on teachers in this changing educational landscape, and at the same time hopefully also make the learning experience for students more interesting, engaging, and thus more productive. Our proposed solution is to utilize Deep Fake technology to produce video lessons conveniently and efficiently. This would ultimately save teachers a significant amount of time they can then spend on planning more engaging lessons and doing more meaningful work.
As of Milestone 2, we have successfully implemented our core feature: a teacher is now able to upload a lesson video (without the teacher’s face) and a short sample video containing their face (or anyone’s face for that matter), and receive as output a new lesson video with the teacher’s face overlaid on the original video in a corner.
Last updated: 28 May
- Generate a deepfake video from user audio and image
- Generate a lip-sync video from your own video and audio
- Generate a tutorial video with an avatar from the input lesson video and the tutor's sample video
- Receive your generated video through email!
- For the features to work, our AWS resources have to be switched on.
- When testing, only submit ONE request at a time. We have limited resources.
- When choosing video content to upload, the recommended duration is 30-60 seconds to get results quickly
- Videos of the tutor chosen must contain a human face in all of the video frames, but the duration may be short (even 5 to 10 seconds will work)
- To obtain the best results, ensure that the tutor in the video varies their facial expressions and includes gestures as if emphasising certain points. To achieve this, it might be better to record a short video keeping these pointers in mind.
- Please reach out if you face any issues or have any suggestions for improvements!
- Provide the user with sample videos to generate the deepfake video by inputting just raw text
- Translation to other languages (e.g. Chinese, Malay) to make education more accessible
- Give the user the option to format the layout of the overlay and output video