Athleisure Simplicity and the Education Afforded by AI

There is one advantage that Stable Diffusion offers that is somewhat unique to me compared to other photographers. Stable Diffusion allows me to share the before and after as long as I'm willing to reveal my process.

Well, I'm willing to reveal my process.

I realized this when I was pondering the benefits of AI and how I could literally talk to Tony Stark about anything. AI isn't afraid of subject matter (within reason) and is more than willing to go down the rabbit hole with me in my interview. Similiary, my Stable Diffusion "subjects" aren't people so they don't have feelings and aren't going to care if I show the process for creating the final product. Historically there were two problems with showing the unprocessed images. 1. It makes the model look terrible because the original images are usually chockful of problems including but not limited to the model being fat, skin blemishes, wardrobe mistakes, etc. 2. I didn't want people to think that I couldn't get better better models if judging my abilities from the unprocessed image.

Fortunately Stable Diffusion (and all text-to-image generators) resolve both problems. 1. There are no humans involved in the creation so no one's feelings can get hurt (outside of my own) and 2. I'm not using models so you can't judge me for my model choice.

But this isn't what I came here to talk about today.

This image of what seems like a blonde girl is somewhat mundane. It's fully clothed for one, which is a departure from most of the images that I create. Secondly the backdrop is austere and devoid of anything special; it is literally grey (probably) paper.

So why this image?

There's something about simplicity that is beautiful. I suppose this post is akin to the grey paper post from last week. But rather than extolling the virtues of grey paper, I'm now talking about the beauty of simplicity. That's studio portraits (closeups) are some of my favorite pictures. It's actually why I have always been (unbeknownst to many) drawn to faces first and bodies (a distant) second. There's something powerful about a look, a gaze, a pose, an angle, that doesn't rely on gimmicks to work. Sure I chose this pose, this body language, and asked for this kind of a subject with this kind of a look, wearing what is still skin-tight clothing... but what makes this image work isn't that she's overtly sexy or doing anything particular. She just is. And sometimes (not all the time) it's just good enough.

To understand this process however, let me start at the beginning.

I started this journey with this image I'd found on the Internet. I was hoping to make a better version of it. Unfortunately the pose didn't play well with the AI models understanding of overalls and ultimately I coudln't create anything that was more interesting.

I then flipped through my own archive of edited imagegs and settled on this image from my horses series because I liked the pose. No they're not the same girl. I literally used different AI models to make each image:

In ControlNet I ripped the pose with OpenPose. What you don't see are the many iterations of modifying this pose (particularly the hand position) so that it is improved (and different) from the original image. I probably went through 10+ iterations of modifying the pose alone.

After 50+ different images and countless changes to the prompt, I got something a little different. Something more clothed but more interesting.

The problem with this image is that it showcases one big flaw of AI models; they're largely trained on porn and Instagram. The model's chest was pushed up too far and felt like it defied the laws of gravity. The shape of female breasts is of course a very subjective matter but I think most would agree that natural and believable shapes look better. Therefore I took it upon myself to alter the size and shape of the chest so that it fit better with my own ideals. But as I always say, your mileage may vary.

I also took liberties with changing the shape and size of the subject's butt. I could pretty much repeat everything that I said above with the exception of the "defying gravity" part. When I look at an image I look for balance. Some part of that balance is photographic, meaning I shot it too high or too low or too close or too far etc. Some part of the balance is intrinsic, meaning it has to do with the subject's proportions, issues that are not illusions/issues created by photography. In this case I took issue with the fact that the subject wasn't to the proportions of my liking (few are). Again YMMV. Do I need to explain that I grew up on a diet of 90's supermodels again? Maybe. I was sure I had a post addressing this in the past...

Of course the final crop, the B&W treatment, there are a myriad of little details I've skipped in the interest of time. The main objective in writing this post was to marvel at how an image so "contrived" could still capture and retain my attention enough to edit and then write a post about it. And two, to discuss the opportunity that AI imagery affords for me on the education side of things because I don't have to be afraid that the model will be afraid of backlash over the unprocessed images (or worse when and ask me to take down the pictures). Will there be more to come? Time will tell. You guys can make requests here or on Instagram.

Superimposed you can see the changes better