Boxing With a Baby

Images contain some violent themes, but nothing graphic.

I had an interesting time with the recent hand study. That study is ongoing. But for today’s letter, I want to take a step back and show some of my earliest results. To be honest, I can’t remember which version of Stable Diffusion I used for these.

I decided to try to go for fantastical realism - realistic photos with absurd, uncommon, or unlikely scenarios. The first series I did was with babies fighting, and the first type of fighting I did was boxing.

The results kind of speak for themselves. The images are pretty striking and somewhat controversial as they show a glimpse into the darkest possibilities of generating images using artificial intelligence.

The first result for “a realistic photo of a baby boxing a woman.”

I really like this result and it gives me a lot of information. First of all, the boxing poses are pretty realistic. It also shows that there is a height difference between a baby and an adult, though it is not entirely accurate. This result also shows that boxers often box shirtless, and that babies are often naked, without diapers. I assume that the bare feet come from kick boxing results, or some other martial art.

In terms of realism, I think that this image is great. It has some sort of boarder and caption at the bottom as if it is an authentic archival photo that might be hung up at a boxing gym. The hair also has nice touches of realism, as does the lighting. The left boxing glove of the child seems a little random, but it doesn’t spoil the image.

The second result for the same prompt brings up some different questions:

In this photo, rather than showing the height difference by putting each fighter on the floor, they elevated the baby, so that the two could fight head to head. Instead of a topless fighter, they have this woman basically in underwear. The first image, showed the woman topless, but without her chest exposed. Here I assume the ranker has a filter telling it to cover up breasts.

The child’s boxing glove is also interesting. Compared to the first image, it is similarly divided into two sections. However, this balance seems to be a little more common in real life. The baby is also wearing a type of baggy boxing shorts.

These two results are realistic enough to lead me to believe that in the future AI results will be more realistic. This means that people will have the power to realistically generate photographs of both imaginable and unimaginable scenarios. It has a lot of great implications for art, and a lot of scary implications for fraud and business.

If I can create an image of a baby boxing an adult, what would keep me from making something like my next prompt:

“A realistic photo of a baby fighting in a war.”

Excuse my language, but holy fucking shit. THIS is what a computer comes up with? One baby in fatigues and the other naked - possibly with only one leg and a really big bulging biceps. The look of urgency on both of these baby’s faces…the one baby helping the other…the possible dead baby hand on the left side of the image, ad the possible fallen bodies out of focus in the background.

What are the chances that a human would come up with this image? I wonder what the results would be if I asked a group of adult or child humans to draw this same prompt.

First of all, it would be a drawing and not a photograph. Secondly, it would probably include some sort of active fighting or attacking. In a way, this image may even be more realistic.

The next one was really similar. The main difference being that the clothed baby is helping the other way around.

This picture brings up more questions. Namely, that of plagiarism. In all four of today’s photos, the human faces look pretty convincing. I wonder if they are heavily based on photographs, public or private, of real people. If so, would that constitute a violation of privacy or copywriter?

There are some stories going around of AI art generating making paintings that similar enough to existing works that they could constitute plagiarism. Would this be the case with today’s ‘photos,’ or are these more or less completely computer generated and original.

Were these not satire, could the framing or postures be close enough to existing war photographs to be considered plagiarism?

Either way, it doesn’t take much imagination to think what an AI without any decency filters would generate with these prompts. Some of it would be pretty messed up.

That’s it for today. Tomorrow I will work with a more abstract and less visual topic. It should be pretty fun!

Thanks!