I created a NovelAI Euterpe module lately for fat fetish stories. I trained it on around 3.5 MB of stories with 100% coverage with 4320 steps. Final training loss was 2.76. I primarily focused on my favorite stories, so the data is biased towards stories with themes like immobility, mobility struggle, and mild health issues. Please let me know what you think about it.

https://files.catbox.moe/depp95.zip

>>4359 (OP)
is the site not working or is it just me?

>>4396
nvm it was just me

>>4359 (OP)
So, uh, how does one use this?

>>4401
if you're on the novelai site, drag the module file onto the site page, and it should upload

when it's uploaded, you should be able to use it

>>4359 (OP)
This is great, I dragged it to a new story and it's having an immediate effect, keep up the good work. Can I ask you how long it took you to train it? Been thinking about creating a module of my own to help keep my stories on track.

>>4475
The training is done on the NovelAI website and only takes a few minutes. However it took me like a month to clean all of the data so it was the right format since it was from a bunch of sources, and fetish writers aren't the best when it comes to spelling and grammar. I would highly recommend learning stuff like Python and regex to automate this process, if you are training on other people's stories. If you are training on a single author or your own stories, then you can probably do this manually. You need a minimum of like 1000 steps (so a bit less than 1 MB of text) for a module and you get diminishing returns past 3000 steps, according to the NovelAI devs. Read this guide for more info.

https://naidb.miraheze.org/wiki/Datasetting_for_AI_Modules

>>4359 (OP)
Came back after a couple days to say thank you. I've been using it for days and generated some decent shit. It has the same usual pitfalls as AI but erotic content never had to be good to be usable.

Great job

>>5200
Glad to hear it. If you generate something you really like, I would love to read it.

This might be a decent thread to ask this (and also bump OP's hard work):

I'm about to fork out the $15 for a NovelAI subscription. I distinctly remember some services/programs/whatever that used the NovelAI API to generate specific stuff. Like roleplaying and chatbots.
I think one's name was something like "KobaltAI"? I can't remember the other one.

Does anyone know what I'm talking about?

>>9741 The post was likely deleted by the mods. My memory's exceptional. It's unlikely anybody remembers since I don't. I'm guessing you've already tried Google, so you should browse through 4chan. I hear those guys really got a kick out of the AI, supposedly.

>>9741
These are called frontends. Something to use to display text that comes from the api that these frontends call from, like novelai or koboldai. There are more of course, like poe.com
You'll need to get your authorization cookie or whatever it's called from novelai, then put that into either tavernAI, which is simple, and has the least feature but the easiest to use. Or SillyTavern, which has loads of features but takes a little while to get used to. Has a lot more customization than tavernAI. TavernAI has a database of characters built in, whereas ST needs you to grab them from chub.ai
I recommend ST more, since it has a lot more features and generally more competent developers. There is a discord server if you need it too. Once you have either ST or taverai installed, you'll need to go to settings and find the selection for novelai api, then click on the instructions or "?" to continue to find the key for your novelai account. Once this is done for either frontend, then you can find good presets for Clio from Novelai's discord server. I recommend this because normal Clio isn't that great, so higher rep pen and other such settings makes Clio a lot better. The main advantage to keep in mind for Novelai, is that you get a lot for the money, and you should never have your access revoked or anything regarding NSFW use. Unlike CAI where mentioning a dick is filtered lol. Clio also keeps up with larger local models (7B 4Bit), meaning for no effort you can have access to something that would need 8GB of VRAM and a lot of fiddling around to get working sufficiently. If you need help, just find the relevant discord servers and ask there, and it should be solved hopefully easily.

>>5200
Would love to see it if you decide on posting it

I'm surprised there's not more modules for Novel that are based around this. How has the new model Clio treating everyone? Its been alright, but I hope to get modules on there soon

>>5777
Hey have you made any similar modules for newer models? Not sure if they're all compatible and I'm being a dumbass but a newer module of this quality on something like Kayra would go crazy