Using AI

I recently updated my mobile phone and part of the package included a free year of Google’s Gemini Pro. I’ve been using more AI assistance at my workplace to help my research efficiency and improve my work throughput, and thought I would take advantage of the advanced Gemini access to do the same for my own personal research.

I’ve already been using Proton’s Lumo AI for some writing research, and it has been helpful to help me understand such things as how best to do mean and nasty things to people in a short story I was writing, but have since abandoned, and a trial run of asking for a writing prompt with a certain theme have been interesting and potentially helpful for those improvisational writing pieces I sometimes start. Much better than the prompts that you can find elsewhere, largely because I can define the genre and theme, or present it with and example text and say: “What are the weak points in these paragraphs and are their any pinch points?” I have yet to do more than trial it in that way, just experimenting to see where it might be useful.

But Lumo is strictly text and what it can find on the internet, and every once in a while it is nice to do research based on images or do something more complex than Lumo is programmed to handle.

That’s where Gemini has potential. The big question that came to mind was, I wonder if it can clean up and extrapolate images it has been fed to create an improved image without too much uncanny valley.

So I took some old, worn out photos of myself today to experiment with.

Yep, that’s a much more flattering upscale and closer to what I looked like at the time in comparison to that washed-out, blurry photobooth capture (for a silly fan-club membership card). If you notice, I had started migrating my hair from a black Robert Smith ratsnest, letting my natural curl show and interim coloring my roots with a dark brown semi-permanent. I continued to grow it out and cut off all the black and let the dark brown fade to my natural blond. But this was the immediate post-goth look I had where I embraced the Doors, The Wonder Stuff, hippie-folk music and paisley. I also stopped wearing eye make-up. The short version of some of the why? I had the local skinheads looking to do the Doc Martens stomp on me because of some false rumors someone was spreading around meand they were looking for a goth, not a hippie. Besides, it was time to move on with my appearance.

I’m actually pretty damned impressed with what AI did to my photo. It even pickup up the scruff on my neck (visible, but only if you look really hard). The eye tone is about right, if maybe slightly darker than my natural.

I decided to give it a real challenge. Take a pixelated old yearbook B&W photo and extrapolate why I would look like in color and with upscaling. The results were no less impressive when you think about how much guessing had to go into the whole thing:

Well, that’s impressive, if it overcorrects on the skin-tone and gives me more of an ethnic feel than my Irish/Slovak/Finn/Swede self would warrant. Still… Dang. I’m impressed considering how little it had to go on. But, if I’ve learned one thing, you’re often better off handling AI in a more stepwise fashion than what I just threw at it.

So, I tried again, simplifying my request to limit to upscaling and applying enhancements. And then I instructed to colorize the B&W photo. It ends up being much closer to reality.

The hair is about spot on, as is my coloring. The jawline is off slightly, as are a few other features, but dang is it close to the person I once saw in the mirror every morning. Spooky close. And, while I did wear makeup, I only did it when I had time or had a reason to. This is pretty close to me on a day without eye makeup.

Of course, I tried out some other recent photos from yesteryear, all with similarly impressive results.

Okay, Michael, how does this work for the purposes of research?

Well, I have a few photos that I would like to compare against photos of me when it comes to trying to locate my maternal grandfather’s family. I had that DNA hit, but have gotten no response when I sent a previously unknown first half-cousin two separate messages. I take that as being that I am potentially an unexpected relationship that they are not certain how to navigate. However, I have located several pictures of possible relationships, hers included, that I could ask AI to do some comparisons on. I don’t need to pester anyone if I can use AI in that manner.

But comparing images and enhancing them helps do all kinds of forensic type research to better understand any number of things. Here is a neolithic symbol found in “X”, where else have examples of it been found? See the potential?

Anyways, I hope you enjoyed my little journey into trying to create better pictures from bad photos. Let me know what you think.