Reflection: Generated Generative AI News
This is the first ever generated image for the YouTube shorts project detailed here
Last year I started a newsletter dedicated around what is going on in the Generative AI space. Trying to select from hundreds of articles, threads, videos and repositories every week the most important topics.
Natuarlly this newsletter also had a News section and I thought, hey why not publish these as YouTube Shorts along side the newsletter.
The idea behind this was to create an alternative consumption way for these news bullets that I would write myself.
But I didn’t want to create these videos by hand as it would 3-5 news bullets each week, I had to find a way to automate them. So I wrote a script that extracts the newsbullets from my newsletter and then first open up each link to a screenshot of the news item. Next I would create an audio file using text-to-speech to generate the main audio track for the news piece.
The length of the generated audio file would serve as the base for the amount of images that I had to create. The idea is to have a new image up every 5 seconds. So the calcuation was easy audio track length divided by 5 minus 1 (the screenshot) rounded up.
Ok so now that we know how many images we need, we can promt an LLM to generate image generation prompts and than use these to call an image generation model to have the images.
Finally I’d use the Python package moviepy to edit everything together, add some keywords that were extracted from the image prompts. Have a title and description generated for me and then upload the finished shorts to YouTube.
If you haven’t seen any, I’d recommend you check it out as I collected them all in a playlist on my YouTube channel.
And the numbers seem to show that this contet works as since the launch of these videos 4 out of the 5 most viewed pieces of content on my YouTube channel are these videos:
Top 5 most viewed content pieces on my channel between the releases off the first short and the release of this article
I have also seen big swings in terms of ups and downs in the number of viewers per short.
But I am now stopping creating these shorts for two main reasons:
I’m myself am not a big fan of short form content. I think it provides little to no value and invites unhealthy behvaior like doom scrolling and I do not want to be part of that.
There has been a rise of so called AI slop that is starting to take over the internet and while I always made sure to properly tag my content as AI generated and later used the YouTube feature for this, I still do not want to contribute to it.
So here I am 127 YouTube shorts later. I think this project was a success based on the learning I had from creating the script to generated the videos. The project also forced me to further reflect on the type of content that I want to be known for and publish on the internet and short form content is not it. I will leave the shorts up on my channel as they are so time sensitive in their nature that they get hardly any views after their first week of release anyways.
All in this project cost me around $60 in terms of API usage, which comes out to around 50 cent per video and with prices dropping I think there is a clear trendline that will move the cost to produce this type of content towards zero - but hopefully we will value our time more than watching AI generated slop.
