Robot Jenny, The Sequel!: The J-1000?

It’s probably a shameful admission for a millennial, I’m not good at taking selfies. “Finding my good side” was simply not a skill of mine.

However, I wasn’t completely satisfied with my last visit to the Remini app where I used the best casual photos I had. I decided to repeat the simulations with more professional photos.

So I naturally did what you are not supposed to do on vacation, wore some casual Friday clothing to hike around Yellowstone National Park, attempting to kill two birds with one stone. I figured I couldn’t go wrong with the epic background. But hey, at least I restrained myself from pulling out the black blazer I had!

I had a specific look in mind for my photos. So, I wondered what the result would be if I uploaded the same picture, 8 times. Would it create a better simulation of the original photo I took?

So first I created 8 copies of the same photo. I downloaded the selected photo to my computer, and made just enough adjustments to each of the variations that Remini would read them as separate photos. I did the same spot corrections to each version and progressively added 5% color and clarity enhancement to each version. They look so different, I know!

Remini allowed for the image to be enhanced (version 9 & 10) but didn’t allow me to delete the photos I had already put in or the previous AI generations.

Apart from the photo that was enhanced above, what was produced with this attempt wasn’t really what I was going for. The photo enhancement only took a few seconds but the AI generations took 3-8 minutes.

I tried several models with some type of outdoor setting...but the results kept getting further away from what I wanted, or even bearing a resemblance to what I look like in real life!

After adjusting some settings, I found where I could select only the images I wanted the program to reference for the AI generation.

I selected the original 8 image and the 2 enhanced ones, with a new outdoor model. Repeating this simulation 3 times, most of the generated photos looked much more like me with this limited pool of photos to base the simulations on.

After looking at these simulations, it seems like the photo being drawn is averaging facial data points across photos it is using for reference and running the photos through their own beautifier algorithm. It would explain why the smile in these simulations is so consistent. I also looks like the main cue for the appearance of the hair comes from the model, who had curly hair in the avatar photo, where in the photos I put in for reference, my hair was straight. The hair color was closer to my own than the model’s though.

The photos from this round of AI revisions that I liked the best were the enhanced versions of my originals.

I suppose I’m a fan of the Remini simulations where their hyper-reality is on the light side!

In the end, I’m not so sure about the Robot Jennies.

I’m fond of the original I guess!