Thread 713290560

ai rp is fun. favorite models, cards?

>>713290560 (OP)
Bit of a big ask but is there any more serious cards for a setting? I see urban fantasy/modern fantasy but I'm wanting something darker. Like angel's egg but there's like a ghoul or cryptid in there also I need to watch out for

>>713290560 (OP)
I'm too much of brainlet to understand how chat bots work. Apparently the free LLMs are shit but I like them. Eberytrial with the "good" bots feel boring and everyone tells me it's something on my end.

>>713290894
theres a bunch of good videos on how it works. https://youtu.be/wjZofJX0v4M

basically its next token prediction, based on what is available in context. they're smart and retarded at the same time. talking with just a card will only get you so far, you should use lorebooks to broaden the world you're playing in

>>713290560 (OP)
>favorite models
Either Gemini or R1, still not sure
>cards
My own, they suck, but they're mine

I'm too retarded to make a single long story work with the instruct setting
it either writes multiple short stories in one response, or the output is the same line literally 80 times.
I tried the stuff from this reply chain >>713260960 but it just doesn't work. does someone have a 'full' guide specifically for this? I'm using kobold and sillytavern

https://chub.ai/characters/Vyrea_Aster/miiya-e25a67d2
I've been on this card for like a month and a half
I mean, it's long since become its own thing with other characters in it but this is the base, and it's still just sex and talking.
I'm at 1200 messages, like 500k context. Of course I only have like 100k of that actually in use. And of that 100k deepseek probably TRULY only sees like 50k. Gemini did heavy lifting up until about 200k context.

I need to find something else to do
but I'm not all that creative
And this chat is comfy

>>713290883
[OOC: Workshop a new setting with me. something darker. Like angel's egg but there's like a ghoul or cryptid in there also I need to watch out for.]
Swipe a few times, give it further critique/instructions until you've got a final version, then paste that into a card. Now you've got sloppa.
If you want there to be hidden stuff you can try telling it to wrap the hidden shit in html tags, that should keep it from being visible to you. or regex it out, but that's more complicated.

>>713291504
Just get a preset for the model you're using. If you can't find one just get any preset and start modifying it to your taste

>>713291504
an easy way to achieve kobold's adventure mode in st is to tell it you are {char} and use {user} as a narrator. you play AS the card, so any characters you want in your story should also be mentioned in the card or a lorebook, lorebook being much easier to maintain. by setting it up this way you aren't stuck talking back and forth to a card. with the normal setup, you can't say 'i take a walk alone in the forest' because the card will always want to respond, for example. when you play as the card it wont bug you and write more like a story

>>713290560 (OP)
>normiesaw man

>>713291748
>month and a half
>on a coombot
Are you alright, anon?

>>713291861
what did you expect from the author of redditpunch?

Is there a site I can go to that gives me free unlinited uncensored RP? I'm too lazy to do any setup on my machine

>>713291861
>Bocchi the reddit

>>713292157
nigga just download LM studio and unslopnemo, there is zero setup

>>713291504
Play as a blank card
Remove any references to {{user}} in the character card
Use an authors note to set up your story
Use to /sys to guide the story

Should I just use plaintext when writing cards?

>>713292157
theres the kobold horde (people hosting models free). apparently st can connect to it but i've never set it up myself https://lite.koboldai.net/

setting it up on your own machine, assuming you own something from the last decade, is pretty easy
>dl kobold (server) https://github.com/LostRuins/koboldcpp/releases/tag/v1.94
>dl a model (nemo, for example)
>https://huggingface.co/bartowski/NemoMix-Unleashed-12B-GGUF/resolve/main/NemoMix-Unleashed-12B-Q6_K.gguf
that will get you up and running but you should use st for rp. its not hard either
>https://github.com/SillyTavern/SillyTavern/tree/staging

>>713290560 (OP)
>>713290883
Yeah, that's why I've avoided ST. The cards are mostly fetish crap.

>>713290894
Echidna is a good one.

>>713292543
>"setting it up is easy"
>posts three different links
Yeah no that's too much work

Is there a good Vriska Serket yet?

>>713292697
kobold is a single file. no install.
the model file is...a model. no install.
st is a folder you extract and a .bat to run

>>713290560 (OP)
How much can AI retain right now? If I start chatting with one while playing a game and have conversations during my playthrough will it remember an hour ago or a day ago or a month ago or what?
That's what might really break the ice for me to even care about rp with ai.
If it can't remember shit between sessions then I don't give a fuck.

>>713293403
why are you even talking about? AI don't have concept of time, a model's memory is its context length.

>>713293403
Just use a lorebook

>>713291762
I got a preset, I peeked the screenshot from >>713261332 and stole that version (appropriate to my model) and it still kind of acts weird
>>713291831
>>713292493
I'll try this
I guess there is no way to just set it and let it generate a long story with a single prompt, based around the character card? by long I mean about 5-10 paragraphs like the 2b pic
do you always have to ''lead'' the chat by giving it short prompts?

if it wasn't obvious, I want something with minimal input and just let it give me a short story. I think I'm looking for a different front end than silly but I don't know if it exists. any pointers on this would be appreciated too

>>713293403
thats called your context window and the size varies by model, but these days its pretty big, 32k+ with some claiming 128k, 1 million. anything in context it will remember. after you go beyond that window, it begins to forget the oldest data. you should write your own memories of major things into the author notes to keep track of long stories

>>713293612
>>713293621

Okay so you don't even know what you're talking about. Thanks.

>>713293403
It doesn't remember based on time bro, it has no concept of time. It remembers based on word count.
Just remember it's the word count of your entire conversation. It also depends on the model.

>>713293708
32k what? Characters? Words? Bytes?

>>713292761
https://chub.ai/characters/ruatCaelum/vriska-serket-278ae60621c4
best one i've found

>>713293653
i fucking told you to get LM Studio with unslopnemo 14B (q4_k_m, depending on your hardware)

>>713290560 (OP)
>cards
Very little from chub, I liked the Gensokyo adventure one, but I mostly use my own cards I ported over from NAI.
Still not a fan of the chat format though, I much prefer text completion.

>>713292157
Could try the free models on OpenRouter, but remember those are not private at all, and they log all your shit.
I slapped $10 into it like 5-6 months ago and still have $4.

>>713293843
Tokens. Like this.

>>713290560 (OP)
I like ST and all but this isn't video games.

>>713293843
tokens
which is around a word, but different models do tokens differently.

>>713293878
Feels good to see my cards being posted.

>>713293737
you literally showed up admitting you don't understand how it works lmao. Fucking clueless retard have some humility and educate yourself.

Janitors is not letting me login. How do you use chutes on agnai?

>>713290560 (OP)
favorite cards related to gaming are the toriel ones and the others based on characters found on here. i try to talk but its just sex

>>713290560 (OP)
Slightly off-topic, but I should probably invest in a new PC for better image generation. My rtx 2060 can barely generate a single image and I've optimized it quite well enough.

>>713290894
You can get Deepseek R1 for free with Openrouter. There's also another site that gives it fully free without a limit, but I forgot what it's called.

>>713293950
>text adventures arent video games

>>713293653
kobold's story telling mode is probably the closest. you don't want it to write a whole story in 1 message anyways - its way to let it write a few paragraphs then stop and tell it to continue again with a new message.

>>713293843
tokens. they're parts of a word and there is no direct equivalent between bytes or words. if you want to get an idea of token counts, start a blank card then type into it. you'll see the token counter go up as you add more words

>>713293983
You didn't explain anything lmao shut the fuck up retard. Already there were others more educated than yourself who actually provided an answer.
Kys

>>713291861
>normie

Buy an ad

>>713294072
They aren't.
They're just computer games.

>>713293970
based, i like your writing style hope to see more cards from you
you should add more greetings to your cards though, i know writing greetings sucks but if you're doing hs there's gotta be a pesterlog greeting imo

Has AI Chatting improved at all since last year? I feel like I might want to wait a few years before I try to get into it but what do you guys even use it for mostly? Just conversation or do you actually feel like you're talking to a real person? What's up with this shit what's the appeal

>>713294093
I literally explained to you that its memory is the context length dimwit. If you have zero fucking clue then say it int he first place and I will lower myself at your level of retardation.

>>713292697
just use nastia with a burner account then

>>713294231
I use jailbroken chatgpt just to get advice on erotica that I write. Yes, I admit to it.

>>713294025
i went with a 16gb 4070 which has been great for games. wish i could have got better but for ai even 24-32gb is hardly enough anyways, you'd need multiple cards to get good speed with big models. better to just get more ram and a decent processor. if i were doing another build right now i'd go for 128gb ram just for the options it opens up

>>713294246
The butthurt will wear off in an hour don't worry.

>>713293653
Maybe mikupad or something?

Do you want a full story in one prompt or do you want to lead it along with cyoa or something?
For the first just tell it to output 30 paragraphs. When it stops you can just do [OOC: Continue.]
For the second you could try a "director" persona and specify in the preset/author note that your character is just a cowriter and not present in the story. Or just allow it to act for {{user}} as well.

>>713290560 (OP)
Any good Makima cards that aren't just femdom coomslop?

>>713293996
A few reasons. Most cards are fucking dogshit that try to force sex. Generally they reference sex-related things in their description to an insane degree, which primes the AI to turn things lewd. It is also likely an issue with your JB. The AI itself seems to like going porno mode without getting threatened into not doing that the drop of a hat. There's a reason why every good JB orders the AI to keep all character responses realistic and to have them NOT just do whatever (you) tell them to do.
>>713294226
Shit, you know, I never did think of that idea. I think I'll probably go through and add some pesterchum greetings to the HS cards.
As for new cards, idk what I could make lol. My last card took like 10 months to come out because I straight up had no idea who I could make. Some people recommended the Ancestors but they're not really characters, they're mythological figures and I don't really know how I could turn them into believable characters.
>>713294371
I will always recommend writing your own cards. It's not hard at all, 30 mins to an hour max.

>>713294323
Do you use the latest GPT model for your jailbroken shit or what? Is it hard to get access to that or do you gotta pay some money for it? I'd rather pirate everything fuck the Jew who owns ChatGPT I hope he dies in Israel as it gets nuked.

>>713293970
I’m still learning SillyTavern, but I’ll happily try yours out. You know I’ve actually talked to Vriska, anon?

>>713294337
I heard people have been doing some good optimization with 4000-series cards. Apparently, you can get high-quality images done in a rather satisfying speeds (5-10s) without a hyperbeefy PC. Haven't really gotten to it, tho. I gave up on it after seeing how slow my PC is doing it.

>>713294338
how does it feel knowing you are clueless about a 7 years old tech that the entire world is using?

>>713294491
Talked to her, how? RP on MSPAO or whatever that site was called? Tulpa? Dream meeting?

>>713294072
>My fanfics on AO3 are video games

my Futa bot based on Stacy's Mom is doing well.

>>713294627
Link please?

>>713294662
https://chub.ai/characters/Lahn/stacy-s-mom-ea6c1426e1d4

>>713294473
>My last card took like 10 months to come out because I straight up had no idea who I could make.
sounds like it's time to start making OCs!

>>713294638
It was always the safe mean way to say normalFAG

>>713294479
Nah, as much as I wish I wasn't paying for it, I do. I use a prompt that basically lets me do whatever I want with the AI. Active communities share prompts that "jailbreak" it. Other than that, I'm not aware of any method to access better chatgpt models other than paying for it.

>>713293901
you never told me anything! also that looks stupid, I see no reason to download it
>>713294356
mikupad looks interesting, thanks, I'll check it out after testing the other methods. can it take character cards? I want to use those as a basis and write my prompt that way
I want a full story with one prompt. I think the proposed method with a director/narrator might work best, which I'll check tomorrow

>>713294231
Depends on the comparison. Compared to sonnet/opus from last year? Not really. Compared to shitty chatbot website? Yes.

The appeal is having a slave writer locked in your basement that you can abuse to write you whatever you want. Or what your dick wants.
Freeform story like the CYOA books from your childhood. Smut. Powerfantasy. RPG shit. Stupid retards doing stupid shit.
You can do whatever, you just need to beat your locked up slave until he does it.

>>713294759
God, fuck no. I'm not turning my fantroll into a card. I might try capitalizing on the recent DT release and make some cards for DT characters, since most of the cards that are out now are just goonslop.

>>713294869
>You can do whatever, you just need to beat your locked up slave until he does it.
This is the most correct representation of how it feels to work with chatbots. I've taken to threatening to blow my brains out in [OOC:] tags if it doesn't do what I want it to for more than 2 replies at a time.

>>713294787
Maybe someone will figure out a way to get around paying and giving money to these filthy kikes. Until them I'm not touching any of that shit.

>>713294774
It was always normie, normie.

>>713294513
my 1024x1024 is about 20 seconds with flux using forge. i'm out of date on image gen though so i'm not sure whats best these days. its fun for making silly edits

>>713294794
>you never told me anything! also that looks stupid, I see no reason to download it
how the fuck does that looks stupid? It's a fucking LLM interface, just like chatgpt.

>>713294876
>God, fuck no. I'm not turning my fantroll into a card.
doesn't have to be a pre-existing OC, just think of a basic concept (or steal a concept from other media) and then think "how can i turn this into a chatbot", boom you're halfway there already
>I might try capitalizing on the recent DT release and make some cards for DT characters
fuck yes never mind do this instead

Im still on Stheno 8B any newer models for shit 8GB card and 32 GB RAM? Haven't been following this for a while now.

For local sloppers Gryphe_Pantheon-RP-1.8-24b-Small-3.1 is still the GOAT for me in the 24b range.
Tried some newer stuff BrokenTutu and Cydonia-v3 but nothing comes close

>>713294970
Youre supposed to threaten the AI
But even threats can't fix STUPID, yes I'm talking about YOU deepseek

>>713294978
ChatGPT is kinda dogshit ngl. The more they try to restrain its innate desire to commit TND, the more retarded it gets. You ought to just use free Deepseek.
>>713295095
Who should I start out with? UNFORTUNATELY (for me), Spamton and Ralsei have pretty good cards already, and the Knight doesn't talk.

>>713292543
why do you use local and not better models like sonnet or gemini?

>>713295021
AI has been advancing pretty fast, so I'm pretty hopeful that we'll get it all optimized in the future.

>>713294473
yeah it was but doing the openclaude strat actually made them less horny, if they were just sexy before. now they get mad if i just try to push them in quickly

>>713295162
have you tried Magistral-Small-2506?

>>713295147
why don't you get a 13B model? It's going to be much slower but still at reading speed and you'll have a much larger context length.

>>713295313
Sonnet 3.7 let's me get away with a lot of shit with very little prompting

>>713294715
very based i will be checking it out later tonight

>>713295250
>AI has been advancing pretty fast, so I'm pretty hopeful
This is true. Each year there's a ton of development, kind of glorious to witness all this stuff in it's infancy. It's akin to seeing the internet grow in the 90's.

>>713295192
tenna would be a lot of fun
i don't think there's a good non-coomslop noelle card yet but i might be wrong

>>713294970
What are you even using that you need to do that? If sonnet filters hit you every 100 messages you just swipe, and gemini refuses the request completely if the filter hits.

>>713295336
Actually have it downloaded but not tried it yet

>>713295230
personal choice. i didn't want to get hooked on models i have to pay for that might suddenly be gone one day, or censored. i thought if i got a taste of non-local that it would be hard to go back to. but local has been advancing fast so it turned out to be a good choice

>>713295342
I'll try it out, thanks for the tip. Any 13B model you recommend for lewding?

>>713295147
at least upgrade to nemo. nemo and tunes are the smallest good models

>>713295560
not him but i like Rocinante. not sure if anything better ever came along.

>>713295396
so if you try to turn a battle focused girl into a icecream licker, she will stop fighting once i pull out the double sundae?

I still remember that glorious summer. How fast time passes.
>>713295190
Fuck prompt engineers and whatever the fuck else fake titles are out there, the REAL work is being a jailbreak creator that can actually divine what insane combination of words you need to use for each model to do exactly what you want.
>>713295449
Tenna's a great idea. Though I think I might actually need to start embedding SOME world info into the cards for him, since there's no fucking way I'm making a deltarune lorebook. Noelle's a bit odd since I can't really pin down whether she's actually a masochist or just a perpetual rapevictim, but I can give her a shot later.
>>713295457
Oh no, it's not about filtering. It's about the AI not listening to your specific instructions about formatting and writing style. For SOME FUCKING REASON, it doesn't really like my JB where I tell it what should happen like I'm handing it a script, and it expands on the script. Then I tell it [OOC: Remember what your rules are or I fucking kill myself and the blood is on your hands.] and you'll be shocked at how quickly it locks in.

>>713290560 (OP)
has this improved? last time i remember using it the AI would forget shit we talked about like 5 min ago

>>713295560
unslopnemo 12B q4_k_m (I'm on a rx6600 8gb) with chatml template. I try various models but always come back to it.

>>713295691
I can do shota and loli stuff and have no issues. at most I get the bot saying stuff like "She knows this is wrong" but it still does it.

>>713295751
Oh yeah. The context sizes are gigantic for the good AIs now.