Automated Videos

Project Links

Key points

Word Generation

Words are generated from themes that I define. An AI generates a list of words and their translations based on the theme.

Word Difficulty

I transfer the AI results to a database managed by NestJS. Each word is manually assigned a difficulty level via a dedicated interface.

Voice Generation

Each word is pronounced by an AI-generated voice running locally. The voice is created using Coqui TTS. I provide a reference audio file for the voice and a text file, and it generates a voice that pronounces the text.

Video Generation

Videos are generated using Remotion. I apply an introduction sentence and a theme, followed by a series of words to guess with increasing difficulty until the end of the video. The whole process is accompanied by a 3D character that reacts to the question and answer. Remotion can generate a video in a few minutes, retrieving information stored in the database through requests to the NestJS server.

Video Upload

YouTube and TikTok offer APIs for uploading videos, but they have access restrictions. So, I developed a NodeJS script to upload videos using Puppeteer, a tool for automating tasks on websites. The entire process—login, upload, filling in details, and scheduling the video—is automated.

Challenges

• The generated voices sometimes fail, requiring manual verification of each result. If the voice is bad or buggy, I have to rerun the AI with a new seed.

• Since the video upload system is automated through a script using a virtual browser, any update to the YouTube or TikTok interface can potentially break the script.

• I initially thought the content would attract a certain audience due to the quality of the videos compared to competing channels. However, the lack of exposure led me to stop publishing new videos.