Instead of making something cerebral I went for ridiculous. Since the Oscars just happened I thought, I wonder what AI would do with the Barbieheimer meme?
It did not disappoint.
Warning! Warning! Warning1
.
.
.
.
.
.
.
.
.
.
If you are a ALL offended by photos of stereotypically "Barbie" looking women, even if they are entirely AI generated, wearing not a whole lot, standing in front of an atomic explosion, look no further.
If you think my mind is in the gutter by creating/posing this (spoiler alert- it is) and think I'm a sexist pig for creating such, well, yeah, I guess I am.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
These are, at worse, PG rated. But they are fantasy. They are not photos of a real person. Just what the AI I'm using thinks Barbie looks like, though I did steer it towards a particular look for a little more consistency.
First try, didn't specify pink pleather but well, Barbie.
But I need an Oppenheimer hat. Looked up online, it was a "porkpie" style hat. Let's see what AI can do with that.
Hmm, that's more of a tophat, and a little dark but close so let it run a few more iterations.
Hmmm. The hat is almost right, but black isn't quite in the Barbie mold. I want a pink hat.
HAHAHA! I got the pink hat all right. I also got a pink mushroom cloud.
Nope, that didn't work. There's so much wrong in this one.
Eh, now the hat is too big again.
That one I think is close to what I had in mind.
That's about right.
More iterations started getting a lot of imperfections.
OK, now on to the real reason for this.
For a few years I've imagined doing some photographic art. Nothing I could actually sell, just ideas in my head for photo shoots. But I'll never be able to actually take those pictures with an actual camera. The issue is time and money. Time, because setting up what are somewhat complicated scenes is not a quick process, and money because I don't have a bunch of pro camera gear to do location shoots like what I have in mind. There are also props. So many props. One of my photo ideas is having a professional dancer (ballroom or ballet) dancing in front of a pile of televisions- in a way that it looks like the televisions are watching her. Another nvolves a group of people in 19th century clothing sitting around a table in the parlor of a ruined house- either a burned-down house with nothing left but rubble and a chimney, or a house that rotted away so that all that's left is a mossy chimney. And double-exposed so they look ghostly. That may be too much for AI but finding the right location is nearly impossible. The only chance I had was about 15 years ago when the local fire department did a training burn of a house in my neighborhood, and what was left matched exactly what I wanted. I should have just taken my camera and done a bunch of establishing shots and then maybe green-screened in the people, but I think seeing that house was what gave me the idea in the first place.
The other issue (that also involves money) is finding people (basically female models) for said photoshoots, being that I'm not an established photographer. I'd have to pay them. One thing I am NOT good at is people skills so I don't know people that would do that sort of project. One woman I worked with at a prior job did express interest with the whole ghost house idea and had friends in the art community that could have possibly made it happen but I never had the time not a location to try and set it up.
So the above photos are really more of a practice run using established themes that an AI understands. The AI works by my feeding it "prompts" and "negative prompts" along with weighting how much freedom the AI has with those prompts. I also have trained it with a certain "face" which gives it a look more consistent. I could actually tell it to use a real person's face, but that's got ethical issues written all over it. But, If I didn't give it a "model" to use, all the photos either look even MORE plastic than these do, they would be wildly different, or they would look like Margot Robbie since that's what the generic, out of the box models would likely use.
The prompt for the barbieheimer pictures above were variations:
negative prompts were:(redactedname:1.5) as barbie walking in the desert wearing a porkpie hat, with a nuclear explosion in the background, in the style of suze randall, Perfect Eyes, (highly detailed skin:1.1), realistic full body photograph, Intricate details, RAW, analog style, sharp focus, 8k, high resolution, dslr
close up, cropped, mutilated hands, too many fingers, text, watermark, caption
This is about 2 week's worth of playing around with it, and absolutely overloading my PC. To get really detailed, high resolution I'd need a better computer. I ran for 2 weeks without a dedicated Graphical Processor, so I bought a used <$100 one off ebay and installed it today which cut the processing time down from 15-30 minutes to 3-8 minutes per image. A better PC with more power (as in a bigger power supply) could handle a real full-blown gaming Graphics card, which would get the processing down to under a minute but cost around $5K.
