So I just bought a large amount of credits and am looking for tips on ai art generation. Anyone do this much? Also if anyone has an idea for some stellar images let me know and I can generate and post the tagged images here.
I've been putting together an AI generated image pack using NovelAI for Fort of Chains here so I've been generating quite a few images the last few months. Best advice I can give is to treat it as a glorified waifu/pinup generator. It's very unlikely that you'll get anything consistent enough from image to image to make something like a comic or VN.
The more of a person that you try to show at once (i.e. full body), the worse the details will look, especially the face.
Don't rely on anything that requires hands with any detail (holding something up close, playing an instrument etc.), the AI sucks dick at drawing fingers.
In my experience, trying to give any woman small tits causes the AI to try to turn her into a loli which is concerning.
Any kind of action scene is a total crapshoot on whether you get something cool, or an amalgamation of fleshy parts.
I'm 99% sure the AI pulls tags from various boorus, so you can check out images there and see how they are tagged, then apply those tags to your prompt and see how they get reflected.
Here are some images that have been generated that I liked and the prompts I used to get them to give you an idea of what I've found that works well for my purposes.
Do be patient, most of the time you're playing a glorified gacha game with the system and your prompt, just hoping that the AI can cobble your tags together into something halfway coherent. I'd recommend getting the $25/month sub if you can swing it for the unlimited normal image generations, just to test the tags out without chewing through your Anlas. Have fun!
I should probably also point out that checkpoints can contain malicious code. So don't go downloading them from just anywhere. Best to stick to trusted sites and curated guides, at least at first. See also
Cpt. Briggs attempts to throw 25 holla dolla in order to get lewds! Roll d20 for neural net proficiency! Natural 1. BONK! Your lewd prompts have been censored and thrown into the horny jail. Enjoy writing SFW literature for the entire month!
VRAM is the big thing, the ideal is 8G or more. At 6G (1660ti) I have to use a command line switch to reduce resource use (mostly just affects speed AFAIK) and am limited to pretty low resolutions. I think the largest I've generated is 512x768. The general recommended minimum is 4G, though from what I understand it's technically possible with as little as 2G.
From what I understand, the literature side of NAI is uncensored. It's possible to run that (or something similar) locally too, but it actually requires more powerful hardware than image generation. I'm kinda interested in that too, so feel free to give me your login details if you aren't gonna use that subscription...
VRAM is the big thing, the ideal is 8G or more. At 6G (1660ti) I have to use a command line switch to reduce resource use (mostly just affects speed AFAIK) and am limited to pretty low resolutions. I think the largest I've generated is 512x768. The general recommended minimum is 4G, though from what I understand it's technically possible with as little as 2G.
From what I understand, the literature side of NAI is uncensored. It's possible to run that (or something similar) locally too, but it actually requires more powerful hardware than image generation. I'm kinda interested in that too, so feel free to give me your login details if you aren't gonna use that subscription...
I have the 6gb version, so we will see. Low resolutions are actually fine as I am looking for something that can be perceived as view from camera/webcam/cctv vibe etc. I tried downloading few modules into novel AI since some assistance (for game development) with writing (especially dialogues, I like to set the stage and let characters figure it out - it doesnt work much xD ) is much appreciated, buut so far modules seems REALLY focused on their theme. I can't stray even a little from usual theme and cant make more than few paragraphs without devolving into forcing a NSFW theme of module. Although maybe it is because not correctly utilizing all settings with lore, memory etc.
For example I downloaded a module trained on some liteerotica stuff, defined the world as medieval fantasy, created a small group (5ppl) of characters with defined relationships, but after a while NAI started to force weird sentences like using cars, marriages, office bosses...but maybe that can be rid off with proper use of all the settings.
VRAM is the big thing, the ideal is 8G or more. At 6G (1660ti) I have to use a command line switch to reduce resource use (mostly just affects speed AFAIK) and am limited to pretty low resolutions. I think the largest I've generated is 512x768. The general recommended minimum is 4G, though from what I understand it's technically possible with as little as 2G.
From what I understand, the literature side of NAI is uncensored. It's possible to run that (or something similar) locally too, but it actually requires more powerful hardware than image generation. I'm kinda interested in that too, so feel free to give me your login details if you aren't gonna use that subscription...
I just now started playing around with AI generators and don't know much. I think if you wanted to use one to generate the artwork for a game it is going to take a lot of effort. You would basically need to decide on a character then create a Lora for your character to override what it would typically draw. Then include that Lora along with other Loras for the different scenes. Of course once you had it all setup it would be pretty simple.
Attached is an idea...I took a basic prompt to draw a girl and then tried to change her up to make her into gawr gura. The single gura image was me changing the checkpoint to one that drew with a more anime style.
