by hyapet » Fri Mar 14, 2025 7:14 am
You're going to have to do a bit of your own research, paulyshoresituation. I'm not going to explain how the whole thing works for you here. It's either I spend fifteen minutes telling you about step one (which any of these sites will explain to you), or you use Google and ask some real basic questions. I'm not going to do Google's job.
Most pay sites aren't too "sketchy." They tell you what's up. If you don't want to share your models or your creations - there's a pretty big button in the options portion of your account that states whether or not your works will be shared. There's so much competition in that space - no one can really afford to start dicking around their users/customers. They also understand that a lot of folks will be nudifying people for whom they don't have the implicit agreement to do such a thing with. If they start sharing pictures that people would (rightfully) want to keep private and to themselves without their knowledge - all hell's going to break out in seconds, never mind days. So, yeah, the sites are pretty straightforward, clear, and honest in that regard.
And yeah - the photoshop required to make really good AI models even better isn't super complicated. It's a process, but I've never used liquify before. I mean, I messed around with it once a long time ago, but I never actually required it. Some things might be a touch more complex, and some things are pretty simple and straightforward.
What really helps is developing an artist's eye for proportions - knowing how big the head is supposed to be - how long the arms are - the basic shape of the human body. From all angles. Then, even when the AI gives you something that "could have been great" - you can reshape and reposition it. It's all about saving time, really. For instance ...
AI is great at filling in two things between two completed points. So, let's say I have a wrist at one end of a selection of space, and a shoulder at the other end of that selection. It's almost a 98% guarantee that the AI will put in the absolute perfect arm within that selection space between those two points.
Now let's say you're asking the AI to make something from essentially nothing. Like ... fill in a body in this space. You might get the absolute perfect body doing exactly what you asked it to do (5% chance), or you could get anything from a body morphing into a truck to a body twisting itself upside down and into itself. When you provide an "open ended problem" (one only end defined), the AI can literally imagine anything.
And that becomes the problems with hands and feet. They're the ends of the body. There is no in-between on them. That's not the only reason they don't get produced well, but knowing how to find hand samples from other images that you want, and then to successfully graft them on to your AI model, seems like it would take a lot more time, but once you become good with it, it actually takes a lot less than hoping the AI might get it right on the two hundredth try.
I know how to put hands and feet together from pretty much nothing - and how to create and color eyes as well. Things that AI is particularly weak at. Also - with skin tones - if you have a very flat and non-defined skin tone - the rest of the skin tone that the AI will make will be based off of that. So knowing how to graft on the skin "type" you want (shiny - freckly - wet/greasy - dry but defined) is also super important.
AI is great at filling in the space between two well defined points. It can also hit a surprising home run when you ask it to make something out of nothing. Actually, it can do that quite often. Just don't sit down and actually expect it to happen. You can be doing trials for an hour and get absolutely nothing.
But even within that nothing, you will find things you like. Let's say you're asking for a chubby girl to be made, but the AI keeps insisting on an hourglass figure. Just make the figure it gives you much larger in the image, and then ask the AI to repair "the already existing chubby girl." The AI will look, be like, "Ah, she is indeed chubby," and then add some actual real flab rolls or chunk.
Trial and error. But, the more you use it, the better you understand it, and likewise, the better you become. I'm no photoshop pro or anything, but I am at the point where if I need to do something I've never done before, I can ask Google what I would need to do it, look up a YouTube video on it, and then proceed to recreate it relatively quickly. Like, I just "get it" now - whereas before - there was a lot of trial and error, and a lot of times I was struggling really hard with something for hours, not realizing there was a better five minute method of doing all of that so easily.
I often go back to my older models that I worked on and am like, "Oh man ... what even," and then start sprucing them up, and I'm like, "Damn ..." at the end. "This is what I wanted it to be all along." Only for me to come back six months later, and say/do that all over again.
Really, with AI by itself, you're being handed a fish. With having some pretty basic photoshop knowledge (and a touch of skill - alongside artistic sensibility), you're being given a fishing rod.