AI Image Generation

ChatGPT

So, recently, I was messing around with ChatGPT and got a cute invite to try their image generation features. It gave this sample prompt:

Turn this photo into a magical “Mini Me” world where tiny animated versions of yourself come to life around you. These cute 3D-style mini characters interact with your everyday surroundings — climbing onto your shoulders, sitting on your bag, waving, playing, and copying your poses — creating a playful yet emotional social media story-worthy scene full of personality and story. The original photo remains untouched while the tiny characters bring the image to life with depth, movement, realistic shadows, and a soft aesthetic vibe.

Ok, that sounds like fun. Lets start with this source image:

Here’s what it came up with:

Wow, that’s pretty wild. The 3d model it came up with is pretty close to real life. I like the silly poses and both the integration with the elements in the original photos and the new details that it added (like the bag).

Like all things AI, sometimes things just go off the rails. Please enjoy this fever dream:

My daughter dug up some baby pictures and tried the same prompt and came up with some pretty hilarious results. She also tried the same with some fresh snaps of her siblings that turned out great.

Bonus Round

You can ask ChatGPT to generate the mini-me’s without the original image. Aw, these little guys are so cute.

Gemini

Folks say that Gemini is the jam for image generation at the moment. Lets mess around with that a bit. Lets start that first image again with this prompt:

turn person in picture in a humorous but realistic cartoon character

Out of the gate, it produced the nearly perfect picture below:

It did a few follow up prompts to fix the hair. The original hair was 100% gray, in real life I’m more like 75% gray. It also had a short traditional hair cut, when I wear a buzz cut currently. Surprisingly, it took a few tries to get it to understand what a buzz cut was.

Anyway, I ended up liking that picture enough to make it the hero image for the 2026 version of my homepage.

Ok, lets try one more transformation. Lets start with that same image one more time to make an images to use for a favicon. Here’s the prompt:

make an 8-bit style pixelated portrait

Nice. I was gonna mess around with it some more, but the enemy of perfect is good enough, so I left it at that and ran it through the favicon generator and added the results to my sites templates.

Conclusion

I’m just dipping my toes into AI for coding, general information synthesizing, and messing around with images. The image stuff is way better than it was the last I tried it (back when it got silly things wrong like too many fingers). Code-wise, I get a lot of help with technologies I’m less familiar with, but still find some big problems with hallucinations.

So, AI stuff is cool and getting better everyday, its still best used as a helper rather than an assistant trying to do your work unsupervised.