This is a thread exploring and investigating the different prompts and quirks of AI image generation.
First off, I don't "know" anything about AI image generation, I'm still new at it as well. But this will be an opportunity for both of us to learn the different effects and what they do. What this thread hopes to do is build a more detailed explanation and demonstration of what different prompts do so I can easily reference them.
You may have noticed I did a NSFW thread first. Why? Because poses are harder to do and require more detail and setup to figure out. Once I figured out how the workflow works I feel I can now cover some of the more simple and unique stuff about image generation. This thread won't be about prompt recipes only, we will cover poses, camera angles, techniques, environment, camera effects, a lot of things that people may not know about even though they seem to be obvious. AI has its own preferences and biased that have to be explained. This thread isn't really a "how to" but more of a "why" and "what". Hopefully you'll learn along with me what AI can provide.
One more thing, even though this is the SFW thread because we're dealing with art you should be warned that there my be things that some people don't seem to be "SFW". We will not be talking about hardcore nudity or porn or violence here, that's for the other thread. Basically whatever you see in an art book or a museum will be the level of SFW applied here. If you don't like girls in bikinis, that's your own problem and I won't be hiding them just for this thread. :ass:
A page going over tag experiments. While it is great and detailed unfortunately I think it is a little too straightforward. Hopefully I'll be able to cover in more detail how some of these tags work.
The main thing to understand is that these are not the only tags used in AI art. From what I've read (may be wrong) people looked at various sites and took image data like metadata, descriptions, and tags. It wasn't just the boorus although they were convenient because of their tag system. Because of this, this is why you see stuff like "8k 4k ultra detailed unity 3d detailed dynamic pose artstation" stuff in the AI art prompts. There are more tags to find out there. For example "boudoir photography" gives you pictures of anime girls in lingerie, even though anime art tends to not discuss "boudoir" and there's barely any mention of it in danbooru. The AI has some knowledge of it and can apply it to generation.
(Theory, don't listen to this) this leads me to believe there may be some weird "Justin Bailey" type of tags that are just pure coincidence that work well for a given prompt setup. Be sure to test things you can think of and see what it generates.
The sites I use
I don't have my own hardware to do AI art, so I have to rely on sites.
Primarily Im going to be using NovelAI: https://novelai.net/
Personally I really like it. The opus tier gives unlimited generations (under a certain resolution) so I've been abusing it for my AI generation tests.
The free stuff: https://pixai.art/
-i used to use this but because of the loli crackdown I've been avoiding it.because of the crackdown stuff like "1girl" and "hololive" got blocked. They've since fixed that problem but I'm not gonna really invest my time in this if even basic stuff that isn't loli gets blocked. On the other hand this site has Lora and model support. It's how I generated the AI Pleroma-tan images.
https://waifus.nemusona.com/
Free generator. Bare bones but it does have the ability to import tags from a danbooru page.
Other stuff: https://yodayo.com/
I can't personally vouch for this site because I've never used it but it does have Lora support (they're called "spells" there) and their own AI bot stuff too.
How this thread will work
I will be studying a certain technique or tag and laying out what they do with AI. Now most tags are self explanatory but I haven't tried a lot of them, so the best way to figure them out is to just use them.
Unlike the NSFW thread we won't primarily cover poses and tag combinations. But if I do I will post a stable combination of tags and give an accuracy rating of that combinations ability to produce the object, pose, or effect. When I test combinations I generate 10 images and see how well they generate the wanted effect. I will then make an image demonstrating the effect in action with my own AI generated characters (if possible). I recommend you do your own testing as well and see if you can improve anything too.
The forbidden arts (funny name) :blobwizard:
Because of the nature of AI being an easy tool to make art there is a lot of controversy about the moral and ethical implications of using it. We will not be discussing it here.
However we will be going over subjects that will probably upset many people. So, I've given these subjects a trendy cool name "the forbidden arts" that will be for AI subjects that will be crossing the line and will upset people. These taboo subjects will be about work shortcuts, copyrighted works, and other possible cheats. Basically anything that will piss off people. :backfromgab:
As a profound man said:
"Al tools are divine tools passed down directly to us by the LORD as punishment for artists being really annoying on twitter."
If you get upset over this or AI this isn't the thread for you. There is an important reason why this stuff needs to be covered and if you ignore it you may be left behind and in trouble. And I'll tell you why right now.
Why do you need to learn how to make AI art?
Unlike the snake oil that is chatGPT and hype from dubious "investors" that are retarded with money AI art has very real and practical use RIGHT NOW and the technology is improving very quickly. If you do not learn this now you will lack a practical skill that will be useful in the future.
You may be an artist who thinks they "can already draw" and don't need it. You'll probably get pats on the back for artists who are anti AI art as well. But secretly, behind your back artists are already using AI art to increase their workflow (or for cheating) and are building up their skills with it. Art jobs will most likely ask if you have knowledge on how to make AI art just so they can make things quicker. I can probably bet that there are even artists that don't know how to make AI art padding their resumes saying they do to get jobs. You're gonna have to learn it.
"Oh I'm not an artist and don't work in art so I don't need this!" Wrong, there are other possibilities for AI art too. I can imagine a scenario where you average Joe or Jane are working at a company and need images for some presentations or print outs or something. Then you average Joe or Jane are tasked to make the image because your company won't spend 100 dollars for a graphic artist. So you boot up some program to make AI stock style corporate art images that the company pays $10000 dollars a year to license and use. These you make some weird diverse image of a black guy with a black woman with a Mexican guy all eating salad at a table for a print out that people throw away the next day. Point is, if you know how to make that you don't have to waste hours making that and wasting your time.
This is one of the rare times you can get ahead and learn a tool that may become very common in the future, it wouldn't hurt to learn how AI art works.
Final thoughts before the thread
This is not supposed to be a serious thread. This is a thread to explore AI art. I will primarily using NovelAI for these AI images with probably a few from PixAI (exploring Loras, real women generation, and their animation feature).
Here are my unwanted tags for NovelAI
Unwanted content tags:
lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, very displeasing, displeasing, jpeg artifacts, signature, watermark, username, blurry,
Should work for most things, although I'll be covering some lower quality stuff and some weird "magic" tags that are available. Those will be rare though, we will be looking at higher quality art.
