I have been working to create a tool that leverages machine learning to generate Instagram posts. This project was inspired by my desire to explore how to build with AI, and my frustration from having to continually feed the social media beast.
There must be a better way!
To test the abilities of AI-generated social content, I launched an Instagram account, @AI_Motivator, that features motivational quotes.
Version 1 of the AI generator uses sentiment analysis and auto summarization of long form input texts to generate an image. The background and font type for the image are generated based on how positive or negative the quote is. You can see that in action below.
Version 1 can accept a text file as long as a full book and output hundreds of images in only a few minutes. The output is approximately 50% useable, with the rest of the quotes being gibberish or not very interesting.
For Version 2, the goal is to reduce the amount of gibberish. My coder and I are using supervised training to train a neural network to generate better quotes. For supervised training, a human will review the image output and mark it as acceptable or unacceptable. This classification will teach the neural net what constitutes an acceptable quote versus an unacceptable one.
Version 3 of the AI generator will get trained on each Instagram post’s performance to create the full feedback loop. More likes and comments means that the quote is acceptable.
From there we should be able to unleash unsupervised generation, where no input text is required and all of the output is acceptable.
Developing solutions that leverage artificial intelligence is quickly becoming a critical skill. While the role of social content creator isn’t going away any time soon, it (like most jobs) has the potential to radically transform (and improve), powered by AI.