SYNTHOGRAPHY: PATIENCE REQUIRED

As some of you know, I’m currently looking for new employment, so I’ve had a lot of free time on my hands - time that I’ve been using to experiment with various new AI tools. I’ve recently shared my thoughts on AI writing tools, like ChatGPT, so I wanted to give a quick update on my synthographic (AI-generated images) adventure.

Patience most definitely required

The main thing I’ve learned is that using AI tools to create images is actually a very slow process. Well, okay, it can generate images in seconds, but to generate the image you actually want takes a lot of patience. It's all about iteration - slow, gradual steps. It actually reminds me of sculpting more than drawing or painting.

Although art and photography are just a hobby of mine, I do have quite a following on certain platforms, like Flickr for example (yes, I know, Flickr isn’t ‘cool’, but it's far better for networking with photographers and artists than most social media, and usually more civilized than DeviantArt), and I’ve had lots of people ask me for advice. Almost without exception, the problems stem from people approaching AI image tools with the misconception that they offer a quick way of creating what's wanted. Inevitably, this leads to disappointment and frustration.

Sure, technically speaking, AI tools like Stable Diffusion, Midjourney, and Dreambooth, are a ‘quick’ way of creating - but, then again, scribbling a line on a piece of paper is a quick way of creating too. However, getting those tools to create the image you are imagining in your mind can be a long process, involving dozens, or even hundreds of adjustments and iterations. And, yes, I know, the scribbled line isn’t a perfect analogy, because even the first image AI spits out can look visually amazing, unlike the scribble - but, isn’t the point of creating an image to unite intent and visuals? If it’s not then surely everything is just debased to the point of nothingness, and humanity may as well give up.

Fingers and thumbs: AI isn’t all that smart

Don’t get me wrong, these AI tools are incredible on so many levels, but they are still really, really dumb on many levels too. For example, I’ve created hundreds of images of people, and a shocking number of them have an unnatural number of digits on their hands! Seriously? I mean, it can pull off insane tricks, but can’t count fingers and thumbs? Still, if you have the patience, you can normally get it right… eventually! 

Likewise, sometimes the tools struggle with background continuities. Like, for example, if someone is standing in front of a wall, frequently the structure won’t line up properly on either side of their body. Now, I know it's easy for me to point to things like this, and there’s probably some super logical technical explanation, but it just seems odd. Mind you, I’m sure all these minor niggles will soon get addressed as the tools get smarter.

Faces reflecting our nature?

Now for something that AI might be doing that reflects something very fundamental to our nature (or, at least, the nature of most of us). I’ve always created a lot of photography and art centered around beautiful women (yes, I don’t care, I’m a straight male and like female beauty - the horror!), so many of the AI works I’ve been creating have featured pretty women. And, despite usually using my own original photography featuring over seventy different real-life models as base image prompts, the AI tends to gradually consolidate their faces. Really, if I go on long enough with reruns, I can eventually get all seventy-odd models to end up with one of five or six faces. Yes, to get to that point does take more iterations than normally needed, but there’s definitely a trend.

Reflecting on this made me think back to some things I’ve read about beauty. Specifically, studies have found that what most people find ‘beautiful’ and ‘attractive’ is actually the most ‘normal’ or ‘common.’ Of course, this seems odd to many people, because, for much of our lives, we’ve been told that beauty is all about being unique and different - especially in today’s woke climate. But, in those studies I mentioned, they discovered that by merging hundreds of different female faces of the same age, for example, they came up with the ‘ideal beauty’ - so the most ‘normal’ is actually the hottest. I strongly suspect that AI tools are, quite logically, reflecting this. Of course, this doesn’t mean that being unique and different isn’t attractive - far from it. In fact, one of the challenges of using AI image tools is capturing the unique subtle visual traits that make people so fascinating.

The journey continues

Anyway, I’ll leave it there for now, as my synthographic adventure is very much still in progress, and I’ll bring you more updates as it unfolds. But, for now, I’m loving experimenting with AI tools and seeing where the journey goes. But, I’m still standing by my personal policy of always using my own original photography, drawings, or paintings as initial prompt images for every creation. Yes, I’m able to get good results using just text prompts, but at a personal level, I feel using my own images coupled with text prompts, makes the entire process more of a team effort.