Question about AI "art"

[U.N.Owen]

I share many of the doubts people have about AI. Even at its best, it lacks the spark and wit of real art. But, since I can write and imagine but can't draw, it has some appeal to me as a way to create images based on my ideas... and while a lot of AI "art" is complete crap, some of it is decent, and a small percentage is actually pretty good.

My question is, for anyone that knows: how are people getting those decent or pretty good images? I've tried a number of AI generators, and no matter how careful I am with prompts, I basically never get anything that even slightly resembles what I describe.

Is it a question of which image generator people are using? Is there some better generator (presumably an expensive one) that people are using to get the better results? I've had pretty much the same useless outcomes with every generator I've tried.

 

[Ticklefeather00]

It's almost impossible to get a nice scene from a single AI prompt. At least in my experience. I'll often create a basic scene with bing ai and then use openart to remove clothing, add lingerie, etc. Also to alter faces and expressions, add bondage... cuffs, etc. and then I generate additional characters 1 at a time, usually in open art and use a program called gimp, much like photoshop only free, to layer the additional characters into the scene... It can be quite time consuming, but I, like yourself, can't draw for nothing and I I have many scenes in my head that I'd like to see.. The AI Generators are getting better but not yet to the extent that you can describe a complicated scene and have it turn out a decent result. The first pic below was a request from one of my watchers on devantart. The second and third are ideas of mine... Both took several hours and multiple tools. They're not great but fun to make.

 

[U.N.Owen]

It's almost impossible to get a nice scene from a single AI prompt. At least in my experience. I'll often create a basic scene with bing ai and then use openart to remove clothing, add lingerie, etc. Also to alter faces and expressions, add bondage... cuffs, etc. and then I generate additional characters 1 at a time, usually in open art and use a program called gimp, much like photoshop only free, to layer the additional characters into the scene... It can be quite time consuming, but I, like yourself, can't draw for nothing and I I have many scenes in my head that I'd like to see.. The AI Generators are getting better but not yet to the extent that you can describe a complicated scene and have it turn out a decent result. The pic below was a request from one of my watchers on devantart. The second and third are ideas of mine... Both took several hours and multiple tools. They're not great but fun to make.

That does make intuitive sense.

I do wonder though if the people churning out dozens of images at a time on Deviantart etc are putting in that much effort... even though some of their images are actually pretty decent.

 

[Ticklefeather00]

That does make intuitive sense.

I do wonder though if the people churning out dozens of images at a time on Deviantart etc are putting in that much effort... even though some of their images are actually pretty decent.

AI does a really good job with single characters, backgrounds etc but not so good with getting the characters to interact with each other.

 

[U.N.Owen]

AI does a really good job with single characters, backgrounds etc but not so good with getting the characters to interact with each other.

I haven't even had that much luck with single characters. If you have a character doing something specific in a specific setting, it will ignore or misinterpret parts of your description, it seems.

 

[Toermentor]

This is my opinion on this matter as a user of AI tools: The more creative control you have the better your art will be. AI by itself produces slop... and it will do that until it gains sentience, which by that time AI "art" will be the least of our problems. I make sure I'm hand holding Stable Diffusion every step of the way till it makes exactly what I envisioned. If you want to see my process for reference you can go here since I posted my workflow:

https://www.deviantart.com/toerment...flow-the-good-the-bad-and-the-ugly-1132099666



Despite of what many think this is a very time consuming process, often taking 20-30 h of work per image.

If you want to get into it hit me up, I'll give you some pointers.

 

[U.N.Owen]

This is my opinion on this matter as a user of AI tools: The more creative control you have the better your art will be. AI by itself produces slop... and it will do that until it gains sentience, which by that time AI "art" will be the least of our problems. I make sure I'm hand holding Stable Diffusion every step of the way till it makes exactly what I envisioned. If you want to see my process for reference you can go here since I posted my workflow:

https://www.deviantart.com/toerment...flow-the-good-the-bad-and-the-ugly-1132099666



Despite of what many think this is a very time consuming process, often taking 20-30 h of work per image.

If you want to get into it hit me up, I'll give you some pointers.

Thanks, that's a very interesting insight.

I do wonder how the people on Deviantart and Pixiv who drop literally 50 similar images at once (many of them with extra toes or other anatomical problems) are doing what they're doing... because some of their images are pretty decent, in amidst the ones they should have deleted. Are they putting a lot of time into each individual picture, even the ones that didn't really work?

 

[Kyle 1989]

I used to use bing and would get some decent sole shots and once in a blue moon actual tickling contact, but like most people say 95% of the time its slop, 6 toes, etc. and almost every prompt asking for tickling is blocked. I switched to using Ideogram.ai and never looked back. its not free, i pay like 20 a month for 1000 image generations or something, it blocks almost zero prompts involving feet or tickling, doesnt seem to care much at all, and can do many popular characters very well. you still wind up getting one "perfect" image maybe for every 5 or so generated, but the sucess rate is much higher than bing, and you can generate in multiple styles, illustration, anime, photo realistic, etc....these images were all straight from the ideogram A.I. generator and have not been altered at all....sometimes i still have to "clean up" little things in photoshop but if you want consistent, regular feet and tickling this one is about the best ive tried. the hardest part ive found is when using a feather or tickle tool, to get it to make contact with that item, but once you get the perfect shot, you can remix it and get infinite variations....

 

[U.N.Owen]

Thanks! I tried their limited free option, and got some pretty decent results straight off. Much better than anything else I've tried.

 

[TKLR42]

Freaky good!

 

[U.N.Owen]

I used to use bing and would get some decent sole shots and once in a blue moon actual tickling contact, but like most people say 95% of the time its slop, 6 toes, etc. and almost every prompt asking for tickling is blocked. I switched to using Ideogram.ai and never looked back. its not free, i pay like 20 a month for 1000 image generations or something, it blocks almost zero prompts involving feet or tickling, doesnt seem to care much at all, and can do many popular characters very well. you still wind up getting one "perfect" image maybe for every 5 or so generated, but the sucess rate is much higher than bing, and you can generate in multiple styles, illustration, anime, photo realistic, etc....these images were all straight from the ideogram A.I. generator and have not been altered at all....sometimes i still have to "clean up" little things in photoshop but if you want consistent, regular feet and tickling this one is about the best ive tried. the hardest part ive found is when using a feather or tickle tool, to get it to make contact with that item, but once you get the perfect shot, you can remix it and get infinite variations....

I mean, for a first try (and free), this is a hell of a lot better than anything I've gotten anywhere else.

 

[Luv2Tkl2]

This is my opinion on this matter as a user of AI tools: The more creative control you have the better your art will be. AI by itself produces slop... and it will do that until it gains sentience, which by that time AI "art" will be the least of our problems. I make sure I'm hand holding Stable Diffusion every step of the way till it makes exactly what I envisioned. If you want to see my process for reference you can go here since I posted my workflow:

https://www.deviantart.com/toerment...flow-the-good-the-bad-and-the-ugly-1132099666



Despite of what many think this is a very time consuming process, often taking 20-30 h of work per image.

If you want to get into it hit me up, I'll give you some pointers.

Thanks very much for that. It's enormously helpful and interesting.

 

[Toermentor]

Thanks very much for that. It's enormously helpful and interesting.

You are welcome. It's only in the service of more people trying making cool things so we all can enjoy it!