Far too much fun with AI

There was a weekend 2 years ago when I fell down a rabbit hole learning about Generative AI. As a software engineer who has long passed the days of being able to build an AI or complex algorithm from scratch, I have spent most of my career integrating things and moving data around. Here suddenly was a new technology unlike anything I’d ever played with before. I’d built simple style transfer and computer vision programs before, but nothing like this.

replicate.com has become a favorite tool to try new models

It wasn’t long before I was using GenAI generators for cover photos of this blog as well as in my home role-playing games to generate concept visuals. Thus far I’ve relied on my OpenAI subscription and mostly throttled use of DAL-E 3 to make images. I’ve stayed away from many of the web toys that are out there, both out of data privacy and, honestly, because I wasn’t ready. I want to have AI be a tool for me, and I don’t consider the generations true art.

playing with DAL-E 3 in early 2024

With a new generation of graphics cards coming, I am getting excited about running open-source generation models locally. I am drooling while watching ComfyUI and ControlNet tutorials to build pipelines for still and video generation. This weekend I went back down the rabbit hole learning music gen and image2video tools. This is a mock trailer for our Cyberpunk game made with https://hailuoai.video and https://www.riffusion.com/ .

I’ve got a lot to learn, and I’ve found learning is the most interesting part. The most advanced thing that’s become easy to do recently is fine-tuning a LoRA model using reference photos of a character. I’m looking forward to doing this for more than just my likeness once I have some local hardware to play with.

a Flux LoRA of my face prompted into a scene and then upscaled and composited in a photo editor

This is an example of an image from Star Wars with a matte mask to accomplish a generative fill

All in all, lots of fun. I think it’s only a matter of time before the novelty wears off, I just think we are in a moment where generative AI has advanced far more rapidly than our culture and mainstream grasp of technology can keep up. Frankly, as a society, we are still struggling with social media and cell phones. Prompting is fun … but I wouldn’t choose to consume something that didn’t have the point of view of a real author behind it. I’m looking forward to finding more ways to use it as a tool in the future.