Welcome to Inkbunny...
Allowed ratings
To view member-only content, create an account. ( Hide )
NaughtyBox

Story AI?

Is there something like novel ai out there for free for stories/text based media specifically
like something you can run entirely on your computer I mean rather than sending it to a server or something?
Viewed: 107 times
Added: 1 year, 11 months ago
 
Tikkity
1 year, 11 months ago
Hey, you might want to look into Kobold AI which allows you to run a model locally. That being said, I do trust NovelAI with my stories and privacy since it's one of their core principles (encrypting stories / no access to user content on their end, etc). NovelAI was founded as a direct response to the fallout of AI Dungeon censoring certain explicit content and the leaks that happened before that.
NaughtyBox
1 year, 11 months ago
kobold ai eh?
alrighty
NaughtyBox
1 year, 11 months ago
Surprisingly got it to work it seems...
Hwelp
Time to figure out how to shove 600k words worth of horny story material from writers who no longer write into it
Tikkity
1 year, 11 months ago
Nice that you got it to work. As for your question: Oof, that's a bit harder than it might seem. Unless they added support for modules/own training material. You're usually kind of limited to what the model was trained on and limited to the context size that your computer can support. I'm not super up-to-date on the latest developments on that front though. Perhaps someone else can find you a solution.
NaughtyBox
1 year, 11 months ago
ohhhh I see
It seems tho what I'm supposed to do is using soft prompts
matthegamer
1 year, 11 months ago
I've seen OneyPlays videos of AI programs. They've proven themselves very funny.
AkaiKitsune
1 year, 11 months ago
I have been using chatgpt myself to make suggestive gay bois story myself. but its been a hassle to work around their prudishness, because they clam shut as soon as things getting a bit too kinky.

still looking for others that can allow me full nsfw stories :D
Reizinho
1 year, 11 months ago
Assuming you are on a Lenovo Ideapad3 running Ubuntu:
0. sudo apt-get install python3-numpy python3-sentencepiece
1. Clone the Llama.cpp git.
2. Enter it's directory from a terminal and run make.
3. Go to WizardLM-7B-Uncensored's page on Huggingface.
4. Get all files (just cloning it does not work).
5. Put the files inside Llama.cpp's models directory (put it inside another directory of your choice and name this directory as 7B).
6. Get into Llama.cpp's directory from terminal.
7. python3 convert.py models/7B
8. ./quantize models/7B/ggml-model-f16.bin
9. ./main --model models/7B/ggml-model-q4_0.bin --prompt "Write a story about two cute rabbits who enjoy having contests and seeing which one is strong. Describe one of their challenges, about who goes the longest time without ejaculating (whoever lasts the longest wins). Describe their growing discomfort as the challenge continues. Write about how badly they need to cum. Eventually, one of them loses it and starts cumming by accident, releasing everything." --interactive --reverse-prompt "You:" --color
Tikkity
1 year, 11 months ago
Not to be a dick, but why would that be something to assume? That's super specific. o.o
Reizinho
1 year, 11 months ago
Because that's the setup I tested it on.x3 Can't ensure it would work on his hardware, since I don't have specifications on it...
LunarFlare
1 year, 11 months ago
I second KoboldAI as a good place to start, just be warned that it can require substantial computer resources to get good speed and run decent models.  But otherwise there are tons of models out there you can run 100% locally if your computer requires it.  The main thing is you need a good graphics card, ideally an Nvidia card with at least 12gb of ram, 24gb is ideal.  Lots of system ram helps, too.

KoboldAI is basically just a client, you'll need to find a model your PC can handle and go from there.  Unfortunately there's tons of models, lots of clients, and this ML technology is advancing at a mile a minute, I'm still trying to make sense of it all.  That being said, my computer upgrade is due in just a few days, once I get that set up I'll work on figuring it out and getting things up and running.  You're welcome to send any questions my way and we can try to figure it out together.  The KoboldAI Wiki is a good place to start.  Note that a bulk of the home LLM resources are dedicated to chatbot style models and programs, which isn't quite as good for story writing.

I also plan on forging a way to share my computer resources with others using the AI Horde, in the future I might be able to lend my computer to the community, if you're interested.  We'll see how it goes, though!  Good luck and have fun 💜
NaughtyBox
1 year, 11 months ago
I think I've managed to get it working for the most part
Right now though I'm trying to figure out how to feed my substantial story porn collection to it.
(not quite sure what I'm doing wrong but... *shrugs*)
LunarFlare
1 year, 11 months ago
Hmm... well if you're trying to feed it a big collection of data, that sounds more like the realm of creating a custom soft prompt, LoRA or other fine-tuning which is definitely beyond my knowledge right now.  I'd probably start here.

Generally while running the LLM, it can only "remember" a pretty small amount of information, definitely not enough for even a single story.  This is represented by token allotments, and usually only contain what's actively happening in the story, it's all the data that influences the next output specifically, not the AI generally.  The token allotment is used for both the "memory" (the lorebook and author's notes in NovelAI, for example) and for holding as much of the story/chat in progress as reasonable.

LunarFlare
1 year, 11 months ago
Also the KoboldAI discord has some good resources by the looks of it, if you want to join
New Comment:
Move reply box to top
Log in or create an account to comment.