Inspiration
I was inspired by Tom Scott who made a youtube video where he used GPT-2 to generate possible future videos for YouTubers and showed the generations to the YouTuber.
What it does
After you input a youtube channel's ID, it gets all the videos the channel has made and then uses GPT-2 to generate video titles that sound like what the channel would upload.
How I built it
I used Youtube's Data API to get a youtube channel's videos and then used GPT-2 and Google Collaboratory to generate new video titles for the channel.
Challenges I ran into
At first, I planned to use web scraping to get the channel videos but that gave me a lot of problems including YouTube eventually not allowing me to even scrape off of it. That is what made me use YouTube's API. I also had problems with GPT-2 not running locally on my machine so I ended up running it on a Collaboratory notebook.
Accomplishments that I'm proud of
I'm proud of the fact that I found a lot of things about Youtube's API and I have a lot of future projects planned in which uses Youtube's API to help make my life easier. I also plan to use Youtube's API to do a few fun things with my own youtube channel.
What I learned
I learned how to use Youtube's API and how to fine-tune GPT-2 for optimal generation
What's next for GPT-2 Youtube Videos
I plan to extend the knowledge I have gained building this to also use GPT-2 to generate new videos I might click on based on my previously liked videos.
Built With
- colab
- gpt-2
- python
- youtube-data-api
Log in or sign up for Devpost to join the conversation.