The other day I went off about AI writing stories quite poorly and the people who inflict them on SOL.
I conducted an experiment by using the AI default settings and asked it to write a story about going to the store to buy soup, forgetting it and buying something else, paying and going home.
It wrote this grandiose tale and said things like the entire store was cheering Mark for buying beer, and that his grandchildren would re-tell this tale for generations of the day he bought soup. It was so over the top that it was funny (to me,anyway).
Nine to ten people took the time to write to me to tell me they agreed and that was really cool. One person wrote to me to tell me he loved the story and even made his own version of it, which proves that there is something for everyone and no one size fits all.
It was shit to me, it was a treasure to him. If I see an obvious AI written story, I'll do what I do with any story that I don't like and ignore it.
I didn't start today's blog to talk about that, though. I just wanted to offer a little update first.
If you read my stories, you know that I enjoy using AI to craft images for my stories of persistent characters. I don't go overboard (IMHO). I enjoy the creative process and I do a lot more than just type a prompt and that is it.
I use paint.net, upscaling, inpainting, and go through some processes to caption them appropriately. They are unique to SOL. I post them here. I don't get them off the web.
I only recently learned to do this, so my older stories that are illustrated will be regular pictures if I used them at all.
I've had people complain that they don't like pictures (for whatever reason) and to them, I say I am sorry. I genuinely am, and I understand it may be frustrating. You have legitimate reasons for not wanting the pictures, and I am not trying to be a dick.
I just enjoy enhancing my stories, like an author that has cover art on a book, or Tolkien and his maps, to me - the goal is not to use the picture to tell the story, but to tell the story and enhance it with a visual.
I've provided some instructions in another blog on how I go about the basic steps to make AI pictures.
It's really not that hard to do the basics. There are tons of sites out there, but most don't allow NSFW (Nude) images. A few allow it in special conditions.
You can also install something on your computer and make them yourself, but I don't know how to do that and don't trust it. I prefer to go to a website, and use their Lora and Models to make the basics, and then upscale and use paint.net to enhance until it looks the way I want.
However, today I made something I am SO proud of that I had to show it off.
https://storiesonline.net/n/33623/nis-breeder-program/72
If you scroll down close the bottom, you'll see that I have TWO of the persistent characters from the story.
You may say to yourself "So what? just tell the AI to put the two nude women in a classroom."
You may also say to yourself, "This is not my beautiful house, and this is not my beautiful wife," and then you'd be letting the days go by (Letting the days go by, let the water hold me down)
Now that song is stuck in your head, sorry.
Anyway, I have never found a way to get the AI to reliably put two persistent characters in a picture.
You can make two characters easily, but if you are telling a long story and you want the pictures to look like the same people from picture to picture it is not so easy.
I use something called a "Lora" that is a bunch of samples of something. in that case, a particular person's look from many angles. If I use two Lora of two different people, and tell the AI to give me (Melissa Rauch) and (Kaley Cuoco) giving each other a dirty Sanchez, I usually get a Frankenstein picture that combines both women's features in the two subjects and a clean Rodriguez because the AI is yet to understand the nuance of a Dirty Sanchez and it gives me something fucked up.
So what I did was make each character individually with no background at all. Then I knitted them together on a much bigger image and using the power of AI to imagine a background (trying many times) I was able to get something approaching what looks like the two women together.
In this case, they are squatting naked in a classroom in the Naked in School universe. The AI initially populated every student in the background with an identical doppleganger of the same student, which was freaky.
So, I had to keep working at it, manually adding, using perspective and blur, and reworking it, until I was happy that the character "Tiffany" was nude, looking bored off to the right in a chair, and then the the other students appear to be at the right scale.
The coolest part is that the AI made it so the daughter is leaning against the lower thigh of the mother, and they are actually touching.
It's not photo realistic where it's going to fool anyone into thinking this is "real" but it was so fucking close that i think the limitation is my skill and not the technology.
There are a lot of settings you have to fuck with, and you need credits (to make more tries) and the patience to alter them and experiment with inpainting and such.
The AI I use to touch up also hates really pretty, bald pussies. It likes to give women thin pussies with a lot of hair.
I had to re-touch the pussies (which sounds pretty hot if you think about it) remove the hair, and I even added a little strand of pussy-goo dangling off the mom, that she can't wipe off because she has to keep her hands behind her head.
Imagine sitting in a room with a bunch of clothed people, and you have a strand of snotty pussy-goo hanging off your bright red gaping, pussy and you have to squat for an hour while everyone watches it slowly descend like a dribble of snot that you won't wipe off your nose.
I can't teach you how to do pictures like this because I am still figuring it out myself, but I thought I would celebrate my accomplishment.
The accomplishment and 1 dollar, is worth about 85 cents, and I get that - but for me, it was really fun to make.
Expect to see a few more of those in the future (probably not ALL the time, it's time consuming).