Fire should not be hair.

Update: Since I wrote this post, I've been thinking about the ethical implications of AI-generated art. Lately, I've trained models on digital art I created myself and incorporating their generations into this blog. Most recently, I started using Tess for artist-friendly AI images.

Since this blog will chronicle my ongoing experiments with and research into AI, I decided to make a logo using ChatGPT 4, the model that includes DALL-E and is only available with a Plus account.

I thought, "Getting a usable result should just take a prompt or two, right?"

Wrong. Here is the original prompt and result:

A pretty good start. But refining this initial concept proved challenging. And sometimes made me giggle.

Like this:

And this:

Ultimately, after about 15-20 rounds of iteration, we arrived at this:

I asked ChatGPT to remove the background while keeping the image exactly the same, which didn't work. After several tries, I removed the background in Adobe Express and voila, the logo, sans background:

Was using ChatGPT easier than creating my own design? Cheaper and faster than hiring help on Fiverr? Absolutely. But it was not seamless and required a bit of persistence to make it work.

As part of my day job, I search for ways that AI can be genuinely useful. And at home, I'm using it to entertain my kids. I will post here to share my experiences, successes, and, of course, the failures in gory detail.

Stay tuned.