Azazin

Newbie
Oct 23, 2021
45
60
You can find Lora on Sabia on the Civitai website. There you can also see examples of works by users of the site using this Lora. It still falls short of the original, but the result is good.

P.S. If anything, pardon my language - I write through the translator DeepL.
wooow, thank you man
 
  • Like
Reactions: StoryMan

ronadan

Member
Jul 2, 2017
167
360
I haven't mastered ControlNet yet, so what I got in Stable Diffusion without ControlNet is this:
Try a different model. there are a ton of hentai models on civitai.com with better outputs, especially if you go for SDXL ones. there are also LORAs to improve hands, positions and even expressions. ControlNet is not necessarily required to get good enough art for this game. it's rather a tool to get a precise picture based on what you want, while I personally tend to give SD a little bit of freedom to surprise me.

even with Stable Diffusion you need talent and imagination. From the developers we wait 8-10 months each for updates for the game.

All we have to do is wait for Lora on Neve.
mimicking Nomo's artstyle is actually pretty easy with SD. you just need at least 10 (up to whatever amount you can get, maybe 100 is a better number) pictures and then train the ai to learn what it means to draw like him. Depending on how many pictures you use and some other technical details, in a few hours (or less) you can create a LORA for his style. people did the same with the artstyle of Warlock and boobs and now we have a pretty decent LORA for that game.

1699096015199.png 1699096046609.png 1699096066511.png

(look at their hands. they give away it's ai work and not real art from the artist, tho problems like this can be mitigated with a little refining)

The same can be done for Neve. Give SD enough pictures of her till the ai learns what you mean by "Neve", and then it can generate based on that. I know nothing about coding, but I can guarantee you guys that generating art to continue making content for this game is the simple part.
 
Last edited:
  • Like
Reactions: StoryMan

ronadan

Member
Jul 2, 2017
167
360
here's a few more refined ones from LORA's original creator, Enigmata:

1699099628031.png 1699099647354.png 1699099668189.png 1699099722093.png



man's a legendary character LORA creator. I highly recommend his work to anyone trying to create NSFW content with SD, especially if you want fantasy-themed pictures
 
Last edited:
Jun 13, 2017
65
200
The year is 2027.

The last biyearly update for Kingdom of Deception has been postponed again, Nomo is recovering from a dreadful tail bone fracture and the coding has become so hard that they're rewriting it in COBOL. This will presumably speed up the next updates and the skillful use of re-used assets from Noxian Nights will give the developers more leeway to properly implement the next batch of updates that will speed up things.



Meanwhile the Kingdom of Deception "F95 Thread" version with AI art and AI writing has managed to reach its third sequel, Sabia is now fighting Orcs in Space. People still furiously debate if the AI writing or Sierra Lee is best, but considering that the model was trained on her writing the discussion feels moot.


It still falls short of the original, but the result is good.
Shitposting aside, the first couple are scary close to Nomo's style, I'd say even superior to his "simplified" output. Worrisome.
 

Retromancer

Member
Aug 14, 2018
404
588
Wow, this AI art stuff is moving fast.

This does generate some serious ethical questions about what exactly art is etc.

But for this game I don't give a shit. Please, please someone AI the shit out of this and put it out of its misery. You'd probably have to put in more work than the devs have though. Shouldn't take too much time.
 
  • Haha
Reactions: Sneed's

ronadan

Member
Jul 2, 2017
167
360
This does generate some serious ethical questions about what exactly art is etc.
I must remind you that at least for now, ai doesn't make anything original. it can mimic an existing art style, not create from scratch. so you still need an artist at first to start the train and then let ai take it from there.

Please, please someone AI the shit out of this and put it out of its misery. You'd probably have to put in more work than the devs have though. Shouldn't take too much time.
the problem is, it actually takes more time than you think. generating art is not hard. I can provide anyone with enough resources, guides and even prompts. The stable Diffusion community is as horny as we are, so we have access to tons of kinky LORAs to generate a wide range of scenarios. while it might take some time to refine, I believe that on a lazy weekend, a person can generate at least 10-20 pictures.

BUT the technical parts still remain. my knowledge of writing and coding is comparable to a goldfish. I'm busy af, but considering my real job is related to content creation, I can offer my help with pics if we have someone to take over those other parts.
 

StoryMan

New Member
Sep 18, 2018
10
47
I made the artwork literally in 15-20 minutes, without ControlNet - each time I changed the base parameters a little bit and put different samplers.

Model - Divine AnimeMix

VAE - vae ft mse 84000 ema pruned (with this VAE this model gives the original graphic style of LORA).

Interestingly, the graphic style of the abs and boobs exactly copies the original, but the face.... I think I heard that this can be fixed with ControlNet, but I haven't mastered it yet, although I have good guides in my native language.
 
Oct 8, 2018
131
124
Can't wait before the devs suddenly start using (often nightmarish) AI art to 'speed up' production.
When in actuality, it'll just be slower and lower quality, with even less effort put in.
 

ronadan

Member
Jul 2, 2017
167
360
Interestingly, the graphic style of the abs and boobs exactly copies the original, but the face....
it could be because of your prompt. try to include descriptions for face and eyes. there are also LORAs to improve them specifically. as I said before, a hentai model could work much better for this artstyle, like this one:



also get a detail controller. you can get better results by including it. just don't overdo it.



I'm happy to help you learn more. just message me. things like this are off-topic. others may not approve of us two talking about technicalities here.

Can't wait before the devs suddenly start using (often nightmarish) AI art to 'speed up' production.
I don't think they can go lower than this, but based on what we've seen so far, I guess digging a new low is always an option for them.
 

KGARrokh

Member
Oct 18, 2018
145
461
I made the artwork literally in 15-20 minutes, without ControlNet - each time I changed the base parameters a little bit and put different samplers.

Model - Divine AnimeMix

VAE - vae ft mse 84000 ema pruned (with this VAE this model gives the original graphic style of LORA).

Interestingly, the graphic style of the abs and boobs exactly copies the original, but the face.... I think I heard that this can be fixed with ControlNet, but I haven't mastered it yet, although I have good guides in my native language.
Terrible eyes!
 

Enigmanic

Active Member
Nov 4, 2019
569
726
I must remind you that at least for now, ai doesn't make anything original. it can mimic an existing art style, not create from scratch. so you still need an artist at first to start the train and then let ai take it from there.



the problem is, it actually takes more time than you think. generating art is not hard. I can provide anyone with enough resources, guides and even prompts. The stable Diffusion community is as horny as we are, so we have access to tons of kinky LORAs to generate a wide range of scenarios. while it might take some time to refine, I believe that on a lazy weekend, a person can generate at least 10-20 pictures.

BUT the technical parts still remain. my knowledge of writing and coding is comparable to a goldfish. I'm busy af, but considering my real job is related to content creation, I can offer my help with pics if we have someone to take over those other parts.
I would be more than happy to write new content for this project. If you're going to use AI to generate the art then we can rebrand the "actors" into new characters and start an original story with a similar theme. We just need someone to do the coding.
 
  • Like
Reactions: scheriel
3.90 star(s) 77 Votes