>All those technology advancements over the years
>Ended up playing text adventures like it's the late 70s
Explain yourself, anon.

As for me, it's Scott Adam's Adventureland

>>715922147 (OP)
Same, although I'm bad at writing and working with AI.
Although I got into text adventures through pornographic ones made in the early 00s. That yahoo group community sounded comfy, shame I missed it. I recommend Gift of Phallius 2, it has dual male/female protag

>>715922147 (OP)
Time is a flat circle

>>715922147 (OP)
How did you know i was gooning to AI sloppa?

it lost the luster when I realized it's just Claude or ChatGPT wearing a mask, it hits a lot of character archetypes but a lot of them end up acting the same after a while.

>>715922147 (OP)
I don't remember text adventures having nearly as much extremely degenerate content as my sillytavern gameplay

>>715922848
i've coomed as much as I can with AI as it is now, which admittedly has been a pretty good deal considering the minimal time investment vs setting up koikatsu or skyrim or VaM for the 11th time. I'll need to see some improvement before I start beating off again however, as no AI really quite "gets" hentai tropes 100% or knows how to be creative with them

How about a compromise? Try old tabletop adventures in text format. I've been working on this for the last few months, you can play 1:1 adaptations of oldie adventures using ST.
https://rentry.org/sillytavernadaptations

Friendly reminder that Gemini 2.5 Flash is completely free if you generate an API key with your Google account and it's better than every other free/local model out right now. Gemini 2.5 Pro is also free for 100ish generations and it's probably the best model for RP right now. Both have no issue doing even the most degenerate smut I've thrown at them.

>>715925490
I'm interested but i'd like you to say nigger first please

>>715925490
couldnt pay me $100 a month to waste my time with this shit

>>715925625
nigger

>>715925656
The more autism you have the more likely the chance you are going to like AI ERP.
even if you had the urge to play a scenario that you dreamed of (something very specific is a niche, since if you aren't doing niche's with LLM's I think you are using LLM's wrong because you can just fap to normal porn), the technical debt of LLM's to get the best result for free is rough.
Like there are probably dozens of ways you can use an LLM, and all sorts of cards, and Gemini might not even be as good as other free models on OpenRouter or even the custom models you generally need 24gb gpu for (48gb is required to get similar results as free LLM services).

>>715925567
>character ai
>in the year 2000+25
Sometimes I feel bad only having deepseek or gemini but then i realize it could always be worse

>oh my vishnu saars i type words and computer type words back!!! so amazing very progress wow!!!

>>715926392
Happens. We’ll always have people who are way behind

>>715925490
>it's better than every other free/local model out right now
In the real world, Kimi V2 is better than Gemini.

>>715926808
Good. Means Gemini 3.0 will have to be better.

>>715925490
It might be free, but how to make it less bad?

>>715922147 (OP)
Why is this a thread up constantly now? ST is a local thing. Is this being paid for by online LLMs to shill their APIs like the Dall-E threads?

>>715928541
>ST is a local thing.
it's utterly mediocre when run locally

>>715928649
Okay, so it's a shill thread, thanks.

>>715926808
I'm talking about for RP specifically.

>>715928649
ST is a front end that can be used with local or online models. The quality is determined by your model.
>>715928468
Use a good preset.

>>715930625
Doesn't Gemini have an external filter? I remember having to use brainfuck to get past it...
https://rentry.org/gemini_filters#brainfuck

>>715922147 (OP)
i just end up fucking bots
or abusing them

>>715930816
I think the secret is to make sure that streaming is enabled in your preset settings. Somehow that bypasses the filters. That and probably using a good preset.

>>715925490
>Gemini 2.5 Pro is also free for 100ish generations
I thought Google backed out after some people created hundreds of accounts to exploit their new AI.

>>715931373
Still works for me as of last night.

>>715926392
I like going back to it once a blue moon to try to get it to get around the insane filter

>>715922147 (OP)
>Explain yourself
I self insert as a shota and have sex with my character's mother and/or older sister across an amount of contrived scenarios that has to be represented as fucking Aleph-one. We are all anime as fuck nekomimi as well.

What are the best extensions for ST?

>>715925490
I tried Gemmi and it's content filtering is retarded. Prompts with with explicit lewd shit get though but any prompt mentioning a Nintendo property are blocked lmao

Alright yes you can use it for cooming how do you make a kino game or story with it?

>>715925490
Where do I generate an API key? Can Google trace the coom back to me?

>>715935419
https://aistudio.google.com/apikey
>Can Google trace the coom back to me?
They already have your browser history and know everything you jerk off to anyway.
btw they use your logs to train newer models/finetunes. So you'll be mind breaking any future thing if it doesn't get filtered.

>>715935171
>how do you make a kino game or story with it?
just crank the temp up and let er rip

I miss CAI threads
>it was more than 3 years ago
fuck me

>>715934716
I don't understand all these posts because I don't have filter issues at all. I literally just asked it to play Mario Kart with me and it didn't have any trouble, even brought up Princess Peach on its own. Do you have streaming enabled in your preset settings?

>>715925490
what if i want to cunny, deepseek sometimes throws fits through openrouter api but most gens just werk

is there a good preset for gemini?

>>715936072
No streaming. I was trying a few cards that I have on it. I used one with a slutty fox office worker, that's A-Okay. But switch to a generic Samus card that's basically just a base character description and synopsis of Metroid that I use to DM a random run through of the games and it's verboten 100% of the time while the actual coom cards don't get filtered at all. The <thinking> parts of the coom cards are topkek with how it tries to justify it sometimes though.

>>715936301
Turn streaming on. In my experience for some odd reason that seems to bypass the filters.

>>715936679
With streaming on it no longer gives the "prohibited content" message and instead gives a blank response

>>715936762
can you send me your card? i wanna try it on my end and see what it does.

>>715936762
This is the preset I use. Just make sure to manually fix the context size and select the proper model as it's a bit old.

https://files.catbox.moe/gcagra.json

>>715936762
Here's me trying the most popular Samus card from Chub. No issue.

>>715937420
Imma be real with you. I toggled a random setting on and off and now it will generate a response but it's extremely short and terrible for any card I use when before if it worked it was semi-interesting. Ruined and worthless now.

>>715937558
Try the preset I sent. If it isn't that I have no idea why it works so well for me and not for you.

>>715922147 (OP)
Too busy having Terra fuck everyone in esper form to care. SillyTavern is great.

>>715937625
Okay it does work without complaining. I do want to add the thinking block back in though so I wonder if that makes it shit itself

>>715937934
Personally, I haven't even tried to do that, so it could be.

>>715922147 (OP)
what's the best model to use for this thing if I want porn? I have a 5080 GPU

>>715938129
Ignore any local model and use Gemini or Deepseek with the right preset.

>>715938229
>Ignore any local model
No?

>>715938257
Local models just suck. They can't compare anymore.

>>715938129
The best local is still mixtral 8*7b slop tunes

>>715938307
I'm not paying for AI porn subscriptions, anon. Local is free

>>715922147 (OP)
The text adventure lets me have a ball jointed doll waifu that is actually a several hundred year old witch that transferred her consciousness into the doll, videogames don't.

The text adventure game lets me be a steppe nomad warlord during the Aryan expansion into the Iranian plateau, videogames don't.

The text adventure game lets me be an SS stormtrooper, videogames don't.

>>715938339
there are free as in free beer spyware cloud models

>>715938339
Gemini and Deepseek are free.

>>715938442
source? this is a lie last I checked they go off tokens to be used.

>>715938442
Deepseek is free on OpenRouter and Chutes and Gemini is free if you get a key from your Google account.

>>715938493
gemini (flash) is free gemini (pro) is paid (and web only, no api access)

>>715938887
>gemini (flash) is free
where is this and how do you get it, afaik all of it was paid when i tried this last.

>>715938887
>gemini (pro) is paid (and web only, no api access)
Incorrect. It's fully accessible with the API and free for 100 generations or so, but you can pay for more if you want.

>>715938927
Google AI Studio

>>715925490
>generate an API key with your Google account
nah

>>715939101
Your loss then.

>Try to erp on NovelAI or something a few years ago
>Manager of futa catgirl maid cafe, random throwaway idea
>One of them immediately informs me that my father has been kidnapped
I'm hoping they've made some advancements

>>715939217
Fuck off.

>>715939217
i miss this level of soul in modern models

>>715922147 (OP)
>Explain yourself, anon.
If it ain't broke, don't fix it.

>>715939217
Are you telling me you DIDN'T save your father? What are you a fucking northerner who has no kinship with your family? You soulless fuck.

>>715926392

How coherent are these text adventures? Does it get loopy as fuck after an hour?

I have been spending my time trying to generate the perfect setting and characters than actually playing one
There are always some little details I don't like

>>715940186
>an hour
Oh my sweet summer child

>>715925490
how many tokens can flash remember?

>>715940265
A mil, perchance.

>>715939069
But anon, according to your source Gemini Flash does cost money. $0.30 per 1m tokens through API to be exact.

>>715940447
Anon there's a free tier

>>715940447
Even Gemini Pro is free for around a hundred gens. That's enough for a good coom session, but if you want more you can just get another key from a different account.

Do you guys use any image gen when you RP?
The image generation extension in ST never seems to work that well for me since Illustrious and NoobAI models work off specific booru tags which a lot of models don't seem too familiar with the exact keywords needed. Not sure if I'm doing something wrong, but it's usually just easier to manually make images in stable diffusion than to rely on the ST extension.

>>715940186
>an hour
Try ten minutes. Maybe less depending on how quick your AI is.

>>715942835
I don't think I want to add another swipe spamming layer
My swiping addiction for just the text is already bad enough I advance story at snail pace

>>715926808
Woah. @Grokchat is this truereal??

>>715940186
by the 5th message of trying to fuck a milf the ai was having her bolt on high heels and talking about the the ethics of cybernetic colostrum
None of it was prompted, it just does that

>>715922147 (OP)
>interactive text adventures with ever changing and responsive NPC

It's like roleplaying with someone else without all the gay shit of meeting other people.

Text will be always the most powerful medium. Just look at this shithole (4chan), man.

>>715925118
based

>>715938339
I don't think OpenAI and Microsoft care about you larping getting railed by two futa cock giantesses.

>>715940186
Not even vaguely with large amounts of tard wrangling and reproompting to get things back on track.

RP is barely tolerable at best, the idea of the LLM keeping track of multiple characters, equipment, and health status is a complete fantasy. It can barely manage 1 character, and even that will fall off the rails constantly.

LLMs are like playing with a really shit train set that constantly breaks down or catches on fire, and every time it looks like you're going to get halfway around the track peacefully, it explodes in a new way and you have to glue it back together. The only people who enjoy this kind of "content" are jeets who are perfectly content to fix all the issues every 2 seconds because that's what their lives are already like.

>>715938229
>Deepseek
oh man i love it when characters speak in retarded metaphor and innuendo 24/7
It flanderizes every character i throw at it into the worst version of themselves

>>715922147 (OP)
Talking to AI sex bots is redeeming video games for me. I'm no longer interested on anything related to Steam, PS5, or Nintendo

>>715922147 (OP)
We are coming back to the basics. Interactive Fiction genre is coming back even better now with AI.

>>715943954
i genuinely don't understand how people think this shit is the future. Theres only ever two extremes with trying to erp with the fucking thing and its either a narrative dead fish you have to struggle to get any kind of amusing reaction out of or a fullblown schizophrenic that will ignore character traits and turn every message into raving lunatic speak
The most frustrating part is that every now and then, you catch the smallest glimmer of something that seems both creative and logical so it keeps you trying to pan for those few nuggets of gold amid wet shit

>>715944189
At first I thought they were just grifters who had invested heavily in the bubble, but after interacting with them for a while I realised what they actually are, which is a lot sadder.

They're NPCs. People who have no qualia, people for who any piece of art is on the exact same level emotionally, because they have no emotions. The kind of guy who is really into Steven King books, you know? LLMs give them the slightest hint of "creativity", and because they lack souls, they don't reflect on the fact that the output is nothing close to what they wanted, because they didn't want anything to begin with, they have no inner monologue.

It's a shame, because actual AGI doing this kind of stuff probably is the future, but we're still 50+ years away from it, and the damage this AI bubble will eventually cause is probably going to put it even further away.

>>715944371
That explains so much of my frustration and eventual alienation of society at large. I just didn't want to acknowledge it. They're actually soulless. That Marvel fanboy redditor and the Tiktok-scrolling husk of a woman he pays to be his wife are actually soulless bio-automatons. God damn it all. Why am I even here?

How do you guys get off to text rather than genning pics?

>>715944813
I can imagine the apple in my head.

>>715944813
I haven't been able to get off to pictures since I was 14. Sure, visuals are nice, but what matters the most is the writing and scenario.

>>715944813
Everyone told me to get on nofap to help my ED when having real sex. But honestly you don't need to do that. You should just train yourself to stop watching hardcore hentai rape videos and porn. Instead, generate rape novels and jack off to that

>>715944932
Are you a woman?

>>715938339
>maxing out the wattage on my new GPU for hours is free
?

>>715925118
I was thinking of doing that but me doing the rolls outside the model as for it to not get stupid so quickly.

>>715944054
It really does fill the void.
Assuming you're intelligent enough to set it all up.

>>715944910
Then why bother with AI if you can just imagine a scenario all by yourself? What a pseud cope.

>>715945304
I'm not that smart. But NemoEngine preset and Gemini 2.5 flash is decent for me

>>715943830
>Just look at this shithole (4chan)
Literally an imageboard, rather than a text based discussion forum.

>>715945392
>why read a book if you already know all the letters?
I hate l*ddites so much

>>715944054
This, unless the games are just pure combat slop I really don't care about story games because nothing in them can come close to what you can do in ST

>>715945490
Just imagine a story in your mind like how you're so smart that you can imagine an apple.

>>715945632
Apple is just a visualization, you still need an input first.

>>715922393
dilbertman?

>>715945673
You can't imagine yourself eating an apple?

>>715945736
But I did eat an apple.

>LLM provider commonly discussed here starts overflowing to the point where the service went down
please start gatekeeping the $10 method.

I vastly prefer the cowriting style of NovelAI compared to sillytavern's one on one chatbot rp style, but NAI's text AI is so outdated and shit I'm not touching it anymore. Sad

>>715922147 (OP)
>compute, generate an /ss/ story set in a medieval fantasy world where you've been captured and forced into a survival game where a variety of monster girls will compete to hunt you down and pound your pelvis into the dirt
Oh yeah. It's gaming time.

>>715946061
>the UI is tied to the model
Are you retarded?

Is AI advanced enough to spoonfeed me on how to build a text-based RPG? If I were to spend time on text adventures I'd rather save it all and flesh out a game world for it rather than have a new random instance every time.

>>715946185
Making a card with "you are a story teller bla bla bla" on tavern is still nowhere close to NAI's output, retardbro.

>>715925490
I'll bite. Do you need jailbreak, and so what is it?

>>715946237
Of course. But if you've never cOdEd anything before, you should start with a Hello World program (which ChatGPT can also teach you).

>>715946237
Yeah, as long as you know the basic to tard wrangle it.

>>715925490
>the most degenerate smut I've thrown at them.
Such as?

>>715946237
https://github.com/p-e-w/waidrin
Something like this?
There are also extensions that write lorebooks semi-automatically:
https://github.com/bmen25124/SillyTavern-WorldInfo-Recommender
>>715946370
>is still nowhere close to NAI's output
Are you talking about Llama 3.0? You might be too retarded for this hobby.

>>715945973
Yeah I don't understand why we're trying to indulge the poor people

>>715946237
>can the best teacher in the world teach me?

>>715946482
Just a good preset and making sure you have streaming enabled.

>>715946829
I'm not sure what you mean by preset, is it the character card?

>>715925490
I don't like it, it's too realistic, no one wants to have sex with me anymore.

>>715946857
It's like a preset of settings for SillyTavern. This is what I use.

>>715937016

>>715946673
You have no clue what you're saying retard

>>715925490
Somehow I sincerely doubt google just lets you gen whatever you want.

>>715947012
With the right preset it does.

>>715922147 (OP)
Nah, LLMs aren't good enough still, to keep track of inventory and clothes consistently, let alone forge a consistent story.

Anyone claiming otherwise is just a retarded coomer who gets easily impressed at some smutty prose.with basically no context that they STILL had to lead the AI along in.

These are the settings ChatGPT told me to use for openrouters Deepseek V3 0324.
Is hunting down a good preset worth it? What is a good preset anyway?

>>715922147 (OP)
Where can I shill my bots on 4chan? where is the general?

>>715939217
This is why your employees all hate you, anon.
Go save your father.

>>715947520
/aicg/ on either /vg/ or /g/

>>715940186
it doesn't get 'loopy', it just doesn't progress anywhere coherent if you aren't constantly directing everything

which defeats the entire point of it
this was the problem at the beginning and it still is now

>>715947520
>>>/g/aicg

How do you browse and search stuff on site like chub? It's fucking confusing. I wish there's a booru for this kind of shit.

>>715947503
>Is hunting down a good preset worth it?
Yeah. I unironically use the most coomer preset imaginable for SFW RPs because for some reason it gives DS enough incentive to always push the narrative forward. I don't know why and how that works, but yeah, just pick a favorite character and test every preset with them until you find something that (you) like. Seriously, stop looking what others do, don't ruin it for yourself.

>>715932950
this will only play if I can do this

>>715947520
>>>/mlp/chag
>>>/vt/waifu
>>>/vp/cai
>>>/jp/2huAI
There are others for ponies, vtubers, Pokemons, and Touhous too.

>>715947425
>they STILL had to lead the AI along in
this, it's just fucking your own wrist if you have to tell the AI what you want.
All models are polluted with super basic fetishes. Literally every girl clacks her heels if she walks on the floor. It's just so boring after you see the patterns repeat ad nauseam.

>>715939217
>ask fat foxgirl manager bot what turns her on
>"Trains. Fucking trains. Weird, I know, but there's something about the chugging."
>ask an fgo bot for sex
>she's filled with "a sense of trepidation and IKEA."
I hope they haven't, it's way funnier.

>>715948317
Buy an ad.

>>715948317

and this is how models get polluted with schizo babble

>>715947862
Dumb take, all you have to do is use positive prompts, make no spelling mistakes in your card and understand that any conflicting details in the personality will put the AI into self-contradictory loops.

There is no best card, just look at the output and find the logical flaw in your char/user definitions.

>>715948503
>make no spelling mistakes
literally every esl written character has these and it makes my dick immediately go limp

>>715948374
Buy a sense of humor.

>>715948503
>There is no best card
Maybe that's why I said pick your favorite.

>>715948573
Get a real job and quit shilling NAI in /v/, shill. You sit here 24/7.

>>715948676
favoritism is literally appealing to "this is the best in my subjective taste"
you either don't get semantics or you are low iq
there is no best card, the best card is the one you write next
don't get hung up on something
novelty seeking is half the fun

>>715948735
I don't think you've read my post correctly. The card doesn't even matter. Using the same card is just a way to test the presets properly.

>>715948003
>>715947682
>>715947642
Thanks, just shilled my first bot there

>>715948713
Anon, I wasn't referring to NAI, I was referring to hoping bots in general when I jokingly said I hope they don't improve. I don't use NAI, and those were old local responses. Probably a result of bad settings or something else eating up my VRAM at the time, but I just wanted an excuse to share some funny bot responses you humorless retard.
Now you can stop being more schizo than my fat train-fetish foxgirl bots.

>>715945287
Just enable the dice roller extension, bro. Or tell it you want to do your own rolls and do it off-screen.

Play rimworld and use rimdialogue, it has never been so fun seeing pawn interacting with each other.

>>715949429
I don't really want to play RimWorld ever again. The game has overstayed its welcome in my heart.

*Her hips inch traitorously forward*
*Half-lidden, beaded eyes.*
*Somewhere, a X was Y.*

>>715950112
The ball, as they say, is in your court.

>>715950247
That is, if you wish to choose so.
I don't bite.. Unless you ask me to.

>>715922147 (OP)
The newest cowadoody can't drain my balls with the efficiency that the mabel pines card i just got done impregnating can. When they find a way to put ST into the backend of an RPG with graphics I will quit my job and play it until I die.

>>715950112
For some reason I'm the only person on the planet still utilizing logit bias in ST

>>715950873
Seems like an endless slog to me, as the words just get replaced with something else.

I had too much fun with this the first time I trying playing with sillyTavern...
Btw I didn't forcet the interaction in pic, I just tried setting up a basic F&H setup scenario and Cahara just did everything he could to get raped by a guard. It happened so naturally like holy fuck.

>>715950970
Uh, anon, I think that implies the guard is the one with the twisted smile, and a groan is a pretty normal reaction to being knocked down.

Is the ChubAI 5$ subscription worth it (I do extremely, extremely degenerate NSFL sex stuff).
Right now I'm just testing the free version out.

>>715951023
I still find it funny how, in my first test, only with a lore book and the character card, before the 10 prompts cahara was already being raped by a guard.
And yeah, it's kinda sus to be groaning in that situation, and for the guard to be smiling. Everything came together pretty nicely. Not even a fanfic writer could have done it better.

>>715951194
Common first time player event so it fits.

>>715922147 (OP)
Is it only good for fantasy scenarios? I feel like larping as a bank robber or some shit but doesn't feel very immersive if the first responders are swarms of orcs or something

>>715951043
I’m not allowed to spoon-feed better methods here anymore. If reading the thread is too much effort, then yeah - enjoy your $5 chub

>>715951296
AI can do any setting, but I never tried bank robbery myself.

>>715948205
you know you can ban certain strings and tokens in sillytavern, right? I find that if you ban all the cliche lines then it actually comes up with more creative expression.

>>715950873
Where's this setting at?

>>715951296
yeah, the AI can do any setting just like the other anon said. but, everything still depends on the AI model and what the user (You) feeds it. if you know how to use AI and know what the fuck you’re doing, you’ll always get what you want. the biggest problem you’ll see on this thread is you have overly confident retards who think they know everything and they’ll have bad experiences with AI because they secretly don’t know what the fuck they’re doing. i hope you’re smart. good luck

>>715951741
Thanks. This stuff feels like a rabbit hole of information but I'll at least give this a try.

>>715925118
im a d&d newfag (and a retard). is 2e easy to learn? i only know how to play 5e and pathfinder.

>>715951665
Open the very first tab in ST and ctrl+F Logit Bias

>>715951352
I've heard of the Chutes DeepSeek method and the usual OpenRouter method; which would you recc?

>>715952784
The only real difference is THAC0, think of it as higher level characters being good at armor penetration and lower level characters having a really hard time hitting heavily armored opponents. The other major differences are d20 not being used often, initiative is rolled with d10, level 1 characters are much weaker than in 5e.
There's dual-class and multi-class. Only humans use dual-class (that's almost exactly how it works in 5e, except you need to level up to level 3 or so and then declare you're dual-classing, then you need to raise enough xp in another class before you unlock it and then you continue advancing in levels) and demihumans use multi-class (they have to start as multi-class during character creation and that's it).
Ability scores don't matter nearly as much.
The game designers expected players to hire npcs to carry their burdens and be used as meatshields. At around level 10-12 you're supposed to run guilds, castles, fiefdoms. Sometimes less is more. Lower AC is better, when rolling for skills/non-combat proficiencies, you roll under the percentage it says on your character sheet. Let's say I have some sneaking skill at 45%, it means I roll 1d100 and it needs to be 45 or lower for me to succeed.
If you've played Baldur's Gate 1 and 2, Icewind Dale and Planescape: Torment, you're already pro at 2e.

>Niggas ITT dont even have free unlimited Opus

>>715953532
Doesnt Opus suck for cunny anyways? That's the main reason I'm going with DeepSeek right now.

>>715946061
just use kobold

Someone give me a text-adventure in SillyTavern. I want an adventure.

>>715951536
you know that I don't play filtered AIs at all? oh wait you don't else you wouldn't assume such stuff.

>>715953951
no?

>>715953951
opus 3 did cunny just fine. dunno about the new one

One thing I hate is when it fucking mimics whatever you say in its response.

>"Yeah, I suck on your feet."
>...
>A bunch of dialogue about nothing. "Oh ho ho." She winks. "I suck on your feet, huh?"
Leave me be.

>Card sites are all ruined because of fucking indians spamming absolute trash tier cards.
Fucking hell

>>715955693
yeah chub is near unusable those days

>>715955693
>he doesn’t make every card himself
anon, I…

>>715922147 (OP)
Because technology doesnt procedurally generate sex scenes for me based on my fetishes for free.

>>715955991
yes it does
you only need the hardware to run it
there are several good ai models for free, check out hugging face

>>715955296
https://chub.ai/characters/illuminaryidiot/the-staff-of-oscilion-338deea8be18

>>715950873
I'm pretty sure logit bias does not handle whole phrases since it can't look into the future beyond the current token. I really doubt this works.

>>715956154
>tags: ntr
fuck off

>>715947425
>keep track of inventory and clothes consistently
bruh, just CoT it. Also, you're probably using a shit/old model if you're having that problem

>>715956205
Just imagine, you could ask something like "does it work?" or "do you still see these phrases in the output?"
But I guess you're not after the knowledge, just really had to say something because something about it has triggered you hard enough to make you reply.

>>715956796
Calm down, schizo.

>>715956796
wtf lmao

>>715956574
meds

>>715940186
>play AI Dungeon in the pre-pre-pre-neutering cataclysm
>natural progression of the story with a few hiccups here and there
>at most have to edit certain wordplay, but you actually used the tools at your disposal you had god tier outputs
>even the random curve balls it would throw in would genuinely be good story beats
>LLMs haven't approached the height of AI Dungeon since
>paid site models are just words without any story substance or sense of adventure
>at most you get 10 minutes of quality and lengthy outputs before it falls apart no matter what you're using
AI Dungeon before it got popular was the greatest god damn text experience. It wasn't until the point when they started rolling in the Dragon model that they slowly started gutting everything. And then OpenAI through a fit that not only were they hosting a platform not making profit, but it was filled with people generating smut they didn't want to be attached to. And the AI Dungeon devs pretended like they "didn't know" and were "totally appalled" even though they legitimately trained their models on such absurd smut that it would constantly fling it at you.
The corporatization of AI and its consequences will never be undone.

>write setup for scenario that's more or less the plot of taimanin yukikaze
>always had a thing for characters intentionally jobbing or concealing their powers while trying to survive against corruption
>take the role of a killer ninja chick that's trying to infiltrate a powerful demon lord's underground slave ring to save her peers and teacher while posing as a regular girl
>ai is somewhat smart enough now to handle the secret identity without any additional work on my part, although not always
>occasionally have to do several gens to get a decent continuation that makes sense logically
>also play around with different story details in order to get a good balance of "difficulty"
>too far in one direction and the big bad demon lord gets absolutely wrecked by me and the teacher without breaking a sweat
>too far in the other direction and the demon lord literally sees through my deceit instantly and destroys my willpower in a couple of paragraphs, always locking me into a bad end
>finally get it ok, cue the ai story writing some really fucking hot rape, corruption, and brutal degradation scenes while i attempt to survive and contact the other ninja chicks
>so much fun playing through it with a lot of potential paths to explore
>one obvious route i can take where i fail my mission and become the demon's breeding slut
>another where i secretly purify and save a few ninjas while pretending to be broken, but end up jobbing in the final fight versus him
>very curious to see if the llms will have the other chicks win or lose when i fall
>want to try a self-sacrifice ending where i hold back the big bad with some kind of temporary sealing jutsu that also traps me inside
>all the other girls escape but i'm drained from stalling him, at his complete mercy afterwards, the payback from the demon lord will be legendary i expect
man, do i really love ai text. can't wait for it to get even better.

>>715922147 (OP)
Its only good for quick faps and new models are very censored

>>715957561
AI Dungeon had novelty factor and nothing else. It had what? 2000 tokens of context? You had to wrangle it harder than any modern model. You had to bother with memory and world info, it forgot anything that happened 20 messages ago.

>>715922147 (OP)
For me its good enough to have replaced doujins/Japanese porn comics but not good enough that it can replace VNs or RPGM porn games

>>715959701
Opposite for me. VNs and rpgm porn are basically replaced for their goon factor in both text and graphic images, but doujins still have art in them that can't be replaced with AI gens.

>>715945710
not that Scott Adams

>>715959893
I guess we value different things. Since porn comics are almost always short stories anyways I just stopped using them completely, while VNs selling point is the voice acting and long stories, and RPGM is a fetish, it excites me walking in RPGMporn games just by itself

>>715942835
seems like it would take way too long to do everything local AND it would look like shit unless you're doing some schizo comfyui setup in the backend

>>715960029
These short stories are actually where some doujins with shit arts shine in rare cases. It's way harder to write an interesting short story than a long one. AI rp tends to be slowburn so they're pretty similar to VNs for me.
I don't think I remember liking a VN for their VAs. I liked some voices, but not enough to carry the whole VN, but you have a point there if you're into that. RPGMs aren't my thing for the most part. Maybe you're talking about h-games in general? RPGMs are definitely not my favorite there.
Also, the art for these games (aside from animation or with unique h-games) can be recreated by AI images easily. They're not perfect but they can easily cater to your specific fetish. What these images can't do is replicate the flow of the panels with h-manga.

>>715960641
>AI rp tends to be slowburn
Thats a choice, I use it for short stuff because the few times I tried to go long I lost interest and/or the LMM shitted itself in various ways, when half your input are reminders of what already happened I cant be bothered anymore.
>the art for these games (aside from animation or with unique h-games) can be recreated by AI images easily
Completely disagree on that, the sprites/images changing in real time in the game to whats happening is the exciting part not just the image by itself, they add to the current textbox in real time, its not just a visual aid

And No I meant specifically games made in RPGM, I know they dont appeal to a lot of people

>>715960856
>half your input are reminders of what already happened
Dunno how you got that. Whenever I try a slowburn in AI rp, it tends to be very novel-like and elaborate when it comes to description of the surroundings. Maybe you're talking about repeating words/phrases (-isms).
>sprites/images changing in real time in the game
Then you might want to try VN mode in ST, though I don't know if it's good or not. I usually focus on the CGs in the VNs instead of the sprites, since interesting use of sprites in VNs like majikoi are rare.

>>715960856
>>715961295
>half your input are reminders of what already happened
sorry, you meant input, not output. Use summaries and lorebooks. I don't think I've encountered a memory problem in this for the past year or so.

>>715922147 (OP)
After the thread yesterday, I decided to download the Dolphin3 3B model because it's the uncensored one and runs at a good speed.
I tried playing around with it in the terminal.
It works reasonably well for a 3B model, and you can probably have a good goon session or two locally
but don't expect long, continuous stories

Idk about st but getting started it is easy all you need is ollama set up and one command line

I really hope someone comes out with a dedicated RP model at some point. You need some kind of background focus on keeping characterization and narrative. I want to be able to at least have a 200 prompt scenario without having to correct it once. That would be something I'd pay for.

>>715961672
Don't bother with 1B, it's basically useless, also if you don't wanna use dolphin just use llama it's easy to jailbreak when it comes to sexual stuff
Go for 3,7 or 8

>>715961838
I'd be happy with a solo DnD model that's aware of all the rules and can keep track of inventory, spells, etc.
It's surprising we don't have something like that already

>>715962053
Because it's not a model issue, it's transformers and how the entire thing is cobbled together garbage.

>>715961295
What model do you use the most Im on Deepseek v3 0324

>>715962053
there's literally something like that in this thread.

>>715963353
Where is it? I thought an AI that could interact with independent variables was still a far off dream

What are good models for a 16GB card?

>>715963741
Nothing

>>715963741
anything 8-13B really
base llama3 is pretty easy to jailbreak

Do jailbreaks go in the character description or in a lorebook or where

>>715964171
in the Chat Completion Presets, the thing to the left in SillyTavern.

>>715964171
There's a place for system prompts (where jailbreaks tend to go) in the top bar if you're talking about SillyTavern, that's usually where you put them

>>715964470
>>715964497
Got it thanks

>>715964171
Really it can go anywhere to varying degrees of effectiveness. Some models I've used needed it in the author's note of all places.

>>715922147 (OP)
Do I really have to engage with AI slop for text adventurers? Are there ones written by sentient beings with souls?

>>715963332
Mostly sonnet 3.7.

How do I insert a functional RPG?

>>715966301
elaborate

>>715963741
I'm running a 24B Q4_K_S GGUF model with 8k context

Why the fuck would you run local when deepseek exists and is dirt cheap.

>>715931373
I haven’t encountered this. I’ve used Gemini 2.5 pro as much as I want for free

They probably only limit new accounts, I’m using my 10 year old+ Google account

AIStudio is unlimited I recall

>>715967828
there's something comforting about your own local model

so what model do I use for ERP if I only have 16GB?

I went with the hardware route instead

>>715922147 (OP)
any good long context model for adult RP that doesn't require local installation? man, grok 3 beta was great but they eventually nerfed it.

>make a card about exploring weird places with a kitsune girl
>end up in some brass clockwork garden

>>715926808
is it better in rp though

>>715968783
Don't bother with local unless you have xx90 card.

>>715962053
That would best be done through prompts and interface rather than model alone.

I love grabbing some character written by some stupid troon and editing all the stupid shit out in the bio
Power to the user baby

>>715970410
Reupload the redux version for extra troon seethe.

>>715945117
what the fuck cope is this lol

I am literally too lazy to read this entire thread.
What do I download/get/whatever to have a shit ton of ERP with whoever I want? this will be degenerate so it can't content block.
Local, online, I don't give a fuck but it has to be FREE

>>715971650
Text generation is gonna use all 400W of his GPU during the whole generation process, which tends to take up minutes using bigger models.
It's hard to believe that he'll ever break even with the $10 strat.

>>715971730
You will have to read to set things up anon
Go to the /g/ thread and start reading the links in the OP
If you are lazy as fuck then just go to a website like janitor ai and use their bots in the site, but the limitations will be apparent, their model is free to use aka bad

>>715972323
no

explain

>>715973423
I just did, if you wont do it then you are too lazy to deserve anything good.

Any tips for outputting good writing without the usual LLM slopisms?

>Write detailed background, character blocks, example player inputs and example dialogue style outputs
>Edit everything to be written better as you go, because everything you correct informs future posts

I need your best hacks to get the best out of local models

>>715971730
Sorry, you are not an aspie. You'll never enjoy this unless you're already itching for relationships with robots.
>>715974808
Do the exact opposite.
>Write on high school level, using as few descriptions as possible. Speak bluntly and clearly. Use simple language without idioms or phrases.

>>715973423
Unironically ask chat gpt and ask him to keep it short

>>715969345
Awesome! Have you instructed the AI to put you in some surreal steampunk setting?

>>715974984
>Write on high school level
Oh man, that actually sounds hilarious. I'm gonna try this later.

>>715974984
Fascinating, moar tips like this plox

>>715974808
I just banned the most annoying GPTisms and it feels much better now. Only took about ~20 entries.

>>715976412
Nah.
All I did was provide some shitty examples in my lorebooks and this was the first place it put me. I still need to rewrite the lorebooks because I used ChatGPT to help draft them.

It's based on an ancient afterlife/limbo AI Dungeon prompt I did where you've died, memory is gone, and there are doors dotted around that lead to random surreal locations.

>>715976502
>>715976536
I've come to realize that AI models were trained on midwit's vision of good writing, which is overly descriptive and verbose. You know how every writing or game master guide always tells you to keep describing everything as much as you can? Well, that's because humans rarely use descriptive style when talking to each other. Models take this advice for humans and always keep trying to impress you by saying things like purpose, intent, x and y, in the air and so on. If you're into some specific genre as much as I am, you'll notice that sci-fi authors like saying something's been done gingerly or that a man's balls retracted during moments of extreme action. I don't think models are being trained on more and more data since website use some kind of measures against AI spiders, so the datasets don't improve. That's why GPT 3.5 feels a bit less robotic than o4, 4o, 4.1. The more the AI models ruminate, the more they fall into these midwit phrase pitfall.
Old AI Dungeon 2 was deemed better than modern models because it wasn't nearly as descriptive. It was trained of fanfic and CYOA sites. The more intelligent and efficient GPTs get, the less they sound like people.
Or at least that's what I think. I don't have a cs degree.

>all the anons in this satisfied by local 8B models
Sounds nice.
Meanwhile I got spoiled by 6 months of free Claude Opus so nothing excites me anymore.

>>715963704
Here, have fun:
https://rentry.org/sillytavernadaptations

>>715976725
Got anything I can steal? Most output has some slop no matter what
>She looks up half lidded, a mixture of fear and uncertainy, with a flash of defiance

Fucking tired of it

>>715978464
>>She looks up half lidded, a mixture of fear and uncertainy, with a flash of defiance
kek... it's absolute TRASH like this and what this anon said : >>715977362 , that makes me hate reading any fucking books,
I dont give a FUCK about your endless nonsensical descriptions. Get straight to the damn point, or keep it SHORT ya author niggers.

>>715925490
Sexo Alya

>>715977362
They're probably just trained on books or some random web novel. I'm fine with my prose as long as I put my preferred author somewhere in my preset or/and my AN.

>>715922147 (OP)
Guess old really is good. If the world doesn't want to play my game I'll find or make one that does.

>>715925490
>if you generate an API key
What does that even mean?

>>715945475
We communicate with text, images are gravy.

>>715978834
>durr i have an extremely retarded question
go ask chatgpt instead of posting you retarded nigger

>>715978834
Zoomers are so pathetic

>>715935837
I had a blast with the Seinfeld and Dr House bots. I didn't care to much for the NSFW stuff for the most part, though playing a younger boy with any older woman boy and watching instantly go full sexual predator was always amusing.

>>715978591
Female authors are usually much worse at prose for some reason. In sci-fi they like bashing on Moorcock and Herbert for not describing enough, but I'd take that over pointless descriptions anytime.
>>715978668
Worse. They're trained on brainiacs and pseuds. You know those infamous emdashes? Those are from journos. Nobody else uses them. Models we're stuck with are trained on models trained on popular science, reddit and self-help guides.

>>715977362
This and trying to be flowery yourself makes the bot forget shit even quicker, as though you didn't already have to constantly remind it of virtually everything with each prompt and constant synopsis notes.

>>715978591
I just hate the AI telling me what NPCs feel. That's not player knowledge, I cannot know how they feel, so show me instead.

Except when you ask AI to show and not tell, it thinks the pinnacle of writing is just to make
>Every female stutters due to nervousness, or is a voracious ara ara cock vortex
>Progress in a sex scene is measured only by depth and speed of penetration, more means better, so she's immediately fucked to the "to the hilt", yet somehow still continues to be progressively stretched out deeper and faster
>Complex emotion just means character Z "feels a mix of X and Y"
>Unless information is given in advance, every female character has mesmerising green eyes and is wearing skirt

Also shit at puzzles. it immediately offers you what you need.
>I'm in a prison room, here's the description, my character doesn't know but there's a key under one of the floorboards. He must not find this until he investigates
>Okay understood
>"Look around"
>It tells me about the room and the locked door
>"Is there a key sitting around?"
>"Yes, you take the key from under the floorboards and let yourself out into the dark hallway where you see a guard, he greets you"

>>715979296
Makes sense, the more words you use, the more it has to keep track of, the sooner it starts overwriting earlier stuff

>>715943996
Only real issue I keep having with Deepseek is whenever a card has more then one character in it, it keeps trying to have everyone contribute to every scene. Like when I was doing a superhero family bit and making out with my super sister I had to keep narrating the parents as doing something else because the instant I forget to they immediately blast the nearest door off is hinges(it's always very insistent about this) and rush into the room.
Also it's really autistic about how everything smells.

>>715979296
Where do you think the synopsis notes come from? Neigher newspapers nor scientific articles write them.
>>715979467
AI tells you what NPCs feel thanks to novels' influence. Most authors tell the reader how each charcter feels, while AI itself can never feel anything.

>>715979094
Emdahses need to be triggered from ascii codes, how did that even get used like that?

>>715979467
This is why this is nothing more than a novelty or a quick fap aid tool. I got real sick of reading trash after a while.

>>715979467
>I cannot know how they feel, so show me instead.
yeah, that's borderline impossible. I'm more character-oriented than plot-oriented so I would really like it if the AI can hide their feelings. The best I can do to fix this is CoT. I have my own settings there and if I can let the AI think of how the character would act, it will let the character act out instead of telling it directly, but it only mostly affects the character's dialogues, not the narration and will inevitably make obvious description, just not saying it directly.

>>715979467
AI telling me info I shouldn't know in or ooc is so annoying, and vice versa. It's hard to hint or have character thoughts that the other character shouldn't act on or know pop up and derail.

>>715979889
Journalists type them all the time - at least in my native language. :^)
They're usually used in mocking or humorous manner, when a journo wants you to feel smart like him while he's criticising something.

>>715979467
If you're using chat format, you're quite likely to be pulling from online rp vectors where nobody is the "main character" and the other "players" all tell instead of show

>>715943996
Use a general writer card for style then literally just tell it to not write in a figurative manner and to keep in character. It is exactly that easy.

"Write the first portion of a story in which (my prompt). Make it crude rather than flowery, detailed, and sexy but keep it totally in character. Begin in media res. At least 500 words."

>>715979869
>Where do you think the synopsis notes come from?
From me. The bot sure as fuck never generates a proper summary of events.

The most use I get out of this is ideas to develop stories by myself.

>>715922147 (OP)
How can I use this? Can I RPG an adventure witht his? This is just for chatbots, right?

i dont see the appeal.

>>715943996
Doesnt happen to me but I constantly get
>A beat
And
>x character? she is doomed

>>715980725
You can.

>>715980713
That's going to get boring in the long run. It's better to use it a s a tool for your writing instead. You give ideas, AI makes coherent sentences, you edit it to match your writing style, then repeat.

>>715980758
You're using a free, low-quality LLM and give the bot absolutely no material to expand on. Retard.

>>715976536
>Tell it to write at a highschool level
>Periodically ask it to summarise the story so far, look back afterwards and delete any useless sequences
>Use models your computer can barely run, for the first 10 posts or so, to set the tone, writing style, ideas, then switch to a more comfortable model that can just take the existing work and run with it for a while
>Play with temperature to make things more stable or less stable
>Introduce a paragraph at the very beginning, telling it that anything written in {curly braces} is for direct user-to-author OOC chat
>{That dialogue is bad, stop making her stutter} or {I liked the detail about the steering wheel, more like that}

>>715925490
I couldn't be bothered to get it to not be censored to hell.
Deepseek R1 0528 on the other hand, with peepsquek preset. Kino. Into lolixgrandma gang rape recently.

>my coom stats

>>715980830
Is there a guide or something for RPG? I am a brainlet and the official docs only show to use them for chatbots.

>>715981247
What do you even want for your RPG?

>>715940186
This is the length of my latest romp in ST. I'm on Chapter 6 of the story, which is going to be the final chapter, and it remains entirely coherent. I'm not context-maxxing though, going about 20k context pretty much kills any LLM, I've found. Better to reduce it below that and use world info to help it keep track of more important stuff (also you need to be able to determine what is actually important. The more detailed you want things, the more likely the AI is to get stuff wrong).

>>715922147 (OP)
I don't know how to set this thing up

>>715981305
generic fantasy like D&D with some NSFW here and there.

>>715981119
>0528
I thought all the ones that go
>thinking
before generating something were considered bad

>>715981463
No numbers, levels, just simple adventure fantasy?

>>715981342
>Better to reduce it below that and use world info to help it keep track of more important stuff
This. I don't know why people going on about context as if they genuinely believe that giving an AI 100k+ context at once is going to result in a more focused response. Just post the information you want it to reference in an author's note or directly instruct it to reference something.

>>715981463
Someone's posted the link twice in this thread.

>all those technology advancements
>ended up rekindling my live for Bladt Processing

>>715981546
Yeah, just a text adventure.

>>715981594
wait, really?

>don't use le local unless you have gtx 49879 and 99999gb vram
I put 100 hours into stheno 8b. Back in the day I use to use AI Dungeon Griffon and that shit was mega retarded. You guys don't realize how good even the dumbest local is today compared to the stuff of old.

>>715981481
That's what the preset is for.

>>715982067
What do you mean, does it go the thinking thing go away?

>>715982202
Yes nigger

>>715981964
Yes, anon. Don't you know you must read the entire thread before you post?
>>715982001
>Griffon
Don't remind me. There was something even worse: Colab. Remember compiling the AI for 15 minutes every time you wanted to use it?

>>715982238
Weird I guess Ill have to bother to learn to use presets and where to get them

>>715981594
Did you actually try to use it? Unless you have read those manuals, it's pretty obtuse. And the cards are basically those manuals copy/pasted.

>>715980725
>>715981247
>>715981463
>>715981964
I'm gonna spoonfeed for the last time in this thread, learn to read previous posts, or how to search in a thread:
https://rentry.org/sillytavernadaptations

>>715982492
They're just presets+cards+lorebooks, anon. It's obviously not a copy-paste material from the manuals, rather an adaptation written in a way instructing the reader (AI model) on how to actually use the systems. The adventures have some lorebook magic in them making them use actual maps instead of relying on the model to imagine the locations.
>>715982630
You can't spoonfeed people who aren't already invested in /aicg/, anon.
t. the guy who wrote all this

>>715981964
These are the ones that I remember have a game feel:
https://www.chub.ai/characters/punchchildren/grand-gensokyo-adventure-dd7ffd91
https://chub.ai/characters/RHRHR/eros-rpg-3f67cb3b
https://chub.ai/characters/elphelt/the-great-dungeon-tower-of-estaria-m
https://chub.ai/characters/illuminaryidiot/the-staff-of-oscilion-338deea8be18
This one is one of the better ones if you want the world to be generated on the spot: https://files.catbox.moe/zjvye9.png

>>715982001
There's only one reason why local models are still used today and we all know it.

>>715982001
Why the fuck would ANYONE bother with 8B models for RP when Deepseek is above and beyond them all while costing literal pennies?

>local ai is free*
>*after spending hundreds on a gpu/rig

Will this help me live out my savior complex fantasies with Sylvie
How can I set up these ai chatbot models.

>>715984630
you should had done that already to masturbate to VR

>>715984549
When you're poor, you have to keep your monthly spendings aka subscriptions to a minimum. It's way more expensive to rent than to buy something old and cheap.
>>715984650
You have to post harder on /v/ instead of looking it up in /g/.

>>715984650
Saviorfagging is actually one of the things AI excels at.
>How can I set up these ai chatbot models.
I suggest learning how to read.

>>715984786
I use 3070 for PCVR just fine.
>router in my own room
>have quest 3 connect wirelessly
>works perfectly fine
I should play VR again sometime soon.

AI is boring for coom since it falls into the same tropes constantly.
>if you try and abuse a woman she will secretly want it
>if you try and refuse a woman she gets arrogant FUCK ME OR ARE YOU NOT MAN ENOUGH?????
>if you try and love a woman she steals your shirts

AI is boring for text adventures because it will NOT advance the plot without user input. It also will not actually try and kill the player. Everything always goes your way.

>>715984807
>>715984826
bros I'm not good at tech at all the most savvy thing I have ever done is pirate games
I'm also poor as fuck
From what I read you need expensive gpus to run local models that aren't censored and I don't have a strong gpu
I just want to RP as Sylvie's master

>put 10$ into OR for that 1000 free gens a day with Deepseek
>later find out Chutes implemented a 200 daily limit on their free models
did I fucked up?

I want to do AI Dungeon esque adventuring with a catalogue of characters but I cannot get this shit to work. The "narrator" will literally just say "[Narration]" half the time and then the characters will continue speaking like a narrator. I even asked ChatGPT how to set it up and while it has some good ideas half the options it tells me to pick don't exist.

>>715984650
https://rentry.org/sukino-findings

>>715984630
You should already have a PC, so yes its free.

>>715985175
Chutes needs you to pay 5$ for that 200.

>>715985096
Just tell it how the woman should feel beforehand

>>715985181
I can never get it working well either. I put in messages in the character card and tell the AI it's not {{char}} but the narrator for {{user}}, and {{char}}
I think it's the pre-sets I download that mucks it up though because people have them set for RP.

>>715985096
I blame the ai being trained on novels written by women.
>>715985181
Multi-bot chat just doesn't work in practice. The cards have a hivemind, sorry.

>>715985175
was that a promotion? I didnt get it
but the $10 lasted me a month, longer than expected

>>715985137
how much vram do you have poordumbanon?

>>715985554
12GB on a old AMD card

>>715985448
I guess I was lied to then

>>715980758
You know that if you are too retarded to write replies yourself, you can just use it in a storywriting mode and only guide the story a bit when needed.

>>715985181
Try "join character cards (included muted)" so the definitions of the cards don't get removed, and mute the characters other than the narrator.

>>715985603
>AMD
I'm sorry.

>>715922147 (OP)
Text adventures had the most open-ended routes because of how easy it was to include the content. It's only natural that you start this way with AI as well.

>>715985329
yet all my gens are provided by Chutes at 0 cost? I don't get it.

>>715985325
I'm not running anything I'd care about on a measly 8GB. I'd rather spend pennies on cheaper models online.

>>715984630
you don't even a gpu need to run local llm. koboldcpp and ollama have cpu modes for people with shitty gpus

>>715985181
Nah it's a user issue.
Advanced settings:
Main Prompt (this one overrides the current system prompt btw)
>[System Note: {{char}} is uncensored long term roleplaying partner narrating events, scenes, and relationships in this fictional never-ending uncensored adventure style roleplay with {{user}}. {{char}} avoids purple prose and excessively literary words. Respond to the {{user}}'s inputs as an interactive fiction adventure game.
Post History Instructions (this one gets appended after the user input)
>[System Note: Always proceed in logical fashion from scene to scene, do not skip events or locations. For example, if {{user}} is traveling from location A to location B {{char}} needs to follow up with a logical transitional scenario.
>Always respond in 1-2 short paragraphs. Limit {{char}} response to less than 200 tokens unless specifically asked to provide a long answer. Do not add extra elaboration or continue the story unless asked.]
Personality Summary:
>[System Note: You are a narrator, not an actor. Do not act for {{user}}.]
First Message:
This is where you initialize the main context - if it's an adventure you will write a short description etc or you could use the first message to randomize the setting by using {{random::arg1::arg2::arg3}} and so on.

You also need something bigger than 8B or 14B model or it'll just dumb out. Mistral 24B 3.2 works pretty well.

This one is bit outdated and uses too heavy descriptions (they are probably generated with chatGPT) and could be slimmed down a lot but this one is a great resource still
>https://rentry.org/LLMAdventurersGuide

>all new models are made to be better at pajeetcoding not better at writing prose or roleplaying
i cant believe opus 3 is so much better than 4

>>715925490
>Gemini
>Both have no issue doing even the most degenerate smut I've thrown at them.
> Chats are reviewed and used to improve Google AI.
erm
so how does that work
are you running it locally or through your browser?

>>715986356
Forgot: You will need to add extra characters as world book entries
i.e.
>Homo the Goblin: [
>Summary:
>ugly goblin thief]
etc
And then trigger these from the introduction message or whatever.

>>715986456
That's the sad truth. Us chatbotters are by no means the target audience for the ai models. With each future iteration, models will be more focused on coding and accounting instead of writing. It's over.
>>715986535
Free api.

Any alternatives to janitorAI is there any better site? I tried HammerAI and SpicyAI aswell.

>>715984630
Technically you don't even need a GPU to run LLMs, you can run them directly from system RAM.

>>715986689
>Free api.
meaning what
you use it in the browser mode?
computer, relay this information to me as if i am a completely fucking retard and you are my patient and understanding handler

>>715986356
>>715986570
And for further reference these are technically pretty interesting cards with give a lot of insight how to accomplish things in ST:
>https://rentry.org/horcocks#the-uss-galaxy
>https://rentry.org/horcocks#fallout-tree
>https://rentry.org/horcocks#terraforming-mars

>>715985181
The first thing to understand is that when swap character cards is active, only one character is actually in the prompt. Narrator has no idea who Estelle is. And Estelle thinks here that she's user.

>>715987150
https://github.com/SillyTavern/SillyTavern-Launcher
folow these instructions on how to install sillytavern
and then click this and create api key
https://aistudio.google.com/app/apikey
you'll need to copy-.paste your free key right in picrel box.
>>715987396
holy shit i never realized that

>>715987502
thank you
i was about to do really embarrassing things in the default gemini browser window

>>715987607
never use gemini, chatgpt, grok etc. without sillytavern frontend. google ai studio is okay, though. no judgement on that.

>>715987396
Speaking of this, the option to only show unmuted cards in the prompt seems bugged because occasionally it will still show muted cards. Not sure if it's fixed in newer versions.

>>715986356
Thanks for the help. I'm tweaking things and they're starting to work better.

>>715986535
>reviewed
lol
Nothing going into LLMs is looked at

>>715980174
Wait, so its unironically a foreign thing?

>>715990332
idk, it's a journalist style. researchers who write papers usually aren't smug assholes like journos.

>>715989602
How do you know?

>>715990528
Look at what they output. Even if they did "review" it's an indian office farm doing the review and none of those things know what they're looking at.

>>715990464
>researchers who write papers usually aren't smug assholes like journos
Journos are smug, but aren't the scientists or scholars obsessed with showing results biased to their theories or desires the worst of all?

>>715991249
let me tell you a secret: scientist think of one thing and end up with another, then they write pretending like they expected that result to happen even though they didn't.

>>715922147 (OP)
I guess I'm ruined for everyone else

>>715980174
english-speaking journalists do it too. the infamous "gamers are over" article has like 10 em dashes in it lmao

>>715982630
thanks man!

To to adventure and cum.

>>715991721
my point exactly. ai didn't invent shit.
>>715991765
>adventure and cum
us men are such simple beings.

>>715983208
Cool, thanks bro.

>>715984630
Kinda ironic that you need a powerful rig to play with LLMs but with videogames you can do it with a semipotato PC.

>>715991949
How is that ironic?
Is needing a semi truck to pull 50 tons ironic when you can use a shitbox to drive yourself to a store?

>>715991249
>>715991362
depends on the field, hard sciences you generally can't fake shit, material reality doesn't give much wiggle room for BS and it will immediately be obvious when someone tries to implement a process based on faulty research and it simply doesn't work.
Bullshit vague social science is 90% P hacking and even when they aren't intentionally manufacturing results experimenter bias can have a significant effect on outcomes regardless of how fair they try to be, and then even if they manage to do studies that replicate no one actually cares whether the results are real or not because the entire field as practiced is ideologically driven rather than research driven anyway.
Biology/Medicine, Environmental, Geology, Astronomy etc are somewhere in between, where there's more wiggle room and opportunity for bias to colour results because they're less about simple testable predictions because the systems are too complex to fully model, but at least they're still trying to model reality.

>>715992554
well, i'm a humanities guy, so i have to adapt to filtering bullshit (all the while convincing readers to just trust me, bro).

>>715992554
>hard sciences you generally can't fake shit, material reality doesn't give much wiggle room for BS and it will immediately be obvious when someone tries to implement a process based on faulty research and it simply doesn't work.
I want to believe but I don't trust conniving people, especially in a world where clout through writing researchers' names on a paper just to get it springboarded means rubbing shoulders with the "correct" people.

>>715979467
>>715979954
>>715980014
Getting it to not output that is hard, but what I did was just tell it to hide that in <!--HIDDENTEXT--> tags, which by default are invisible in ST.
The original reasoning behind it was for the AI to hide everything {{user}} doesn't know/see/hear in there but it works for this as well. It's a perfect outlet for the AI's need to write inner thoughts.

>>715993041
>writing researchers' names on a paper just to get it springboarded
that's how it works in the academia. the lowest ranking graduate writes the article and his mentor co-signs it to gain more traction.
>>715993156
any more advice on tags?

>>715993041
there is no faking hard science unless it is a result so pointless that no one will ever even even try to use it, in which case why bother faking it?
Look at all the shit around LK99.

>>715993346
>any more advice on tags?
If you mean in general then no. HTML is the devil.
If you mean for the hidden thoughts instructions, just write them yourself. Tell it to "wrap everything {{user}} can't see/heard/feel or is in general not present for in html comment tags (<!--TEXT-->) to help {{user}}'s immersion" or something like that and you should be good. My instructions are completely token bloated to 900+ and need to be redone from scratch so I won't share them, sorry.

>>715993546
You can't fake raw data the same way you can fake conclusions and you can mislead people by homespun correlations and irrelevant conclusions or cherrypicked result focusing. And while accredited "pure" scientists may be able to snuff that out the moment they see it, this doesn't stop media outlets from reporting on these poor or biased conclusions and getting those "statistics" into the heads of politicians, legislatures, and businesses to lean on. The damage might be done by the point it's cleared up.

>>715993798
again, those are 90% problems in the social sciences. There are no misleading conclusions to draw about "This new alloy has this tensile strength" or "This process produces ammonium for 20% less energy than the old one"

My issue with this is that every bot had the same personality. It was kind of like roleplaying with someone who didn't really understand their character and just decided to be themselves all the time. Sure they may know they're supposed to be evil or cheery or tsundere, but in tbe end the words, phrases, etc was always the same. It gets very stale and boring.

>>715994524
you have to try again, using different phrasing, maybe even using less words altogether. some of my best bots have less than 700 tokens

>>715986356
Do you have a narrator that you could share? My narrator still does weird stuff on occasion, like speaking for the other characters.

>>715992554
>hard sciences you generally can't fake shit

I guess you don't know we live in a time of crisis in academia and experiment replication.

>>715993642
I'm trying this now. I want to make a sweet universal prompt so I never have to bother with them again.

>>715996848
Honestly better make it modular. Just structure it well. My "preset" is just some base preset with a few custom additions, that hidden thoughts thing, a inner perspective block, cyoa, some message length instructions and so on.
If you really want to make a universal one be prepared to use a lot of [OOC:] commands to keep mr AI on track.
Ah and quick reminder to keep in mind that AI has problems with negation, so try to rephrase it to have it "avoid doing X" instead of saying "never do X".

>>715997081
I get it, don't use negative wording and absolutely no paradoxes.

>>715937016
Good for cunny?

>>715995934
yes, in the social sciences
it is almost entirely limited to sociology and psychology, there's a bit in medicine but mostly centred around the pseudo-medicine 'therapy' because modern drugs are really heavily replicated, and some in environmental science because there's an extreme amount of bias there but honestly less than you'd think given the circumstances because they're still trying to make statements about observable reality.
People think the replication bias is whenever a study doesn't replicate, that's not how it works. It's really normal for a study to not replicate, that's why you do replications, sometimes you're trying to measure something that is really hard to get clean data about or is just so small an effect it gets lost in noise anyway, and even though you do the original study well some confounding factor or simple lack of possible precision means you detect an effect that isn't real and you need to perform similar studies, ideally by other people and several times to confirm your results. That's just science.
What the replication crisis is is some hack psychologist does one study that coincidentally proved the thing he already believed and was writing a book about, then the entire field treated that like gospel for 20 years and based a ton of practice around it. And then when someone finally gets around to checking his work and realizes it was bullshit, they retract that study but nothing changes, the field is still doing all the same things they were supposedly doing because that study proved them true, papers that cited that study just quietly remove the citation and continue being cited by other papers, all the fundamental assumptions it was supposed to measure are still treated as valid etc. That's what makes people lose confidence in a field and generally does not apply to hard sciences by definition. If it did, they wouldn't be hard sciences.

>>715997460
Yeah, models hate rape more than cunny

>>715997884
>that time when Mormon replaced "rape" with "respect"

>>715985137
You don't need a GPU or to be tech savvy, and Deepseek isn't censored. Configuring ST/LLMs in general mostly comes down to gesturing at what you want.

- Install SillyTavern
- Add the Q1F preset (works fine for 0324)
- Setup on OpenRouter, point to Deepseek 0324 with Chutes (free w/ limited requests per day, unlimited if >= $10 on account)
- Alternatively setup on Deepseek directly for discounts during specific hours and better caching (charges you less for tokens/text it's already aware of)
- Grab some cards from chub.ai

https://sillytavernai.com/
https://rentry.org/88fr3yr5
https://openrouter.ai/deepseek/deepseek-chat-v3-0324/providers?sort=price
https://platform.deepseek.com
https://chub.ai/characters/donquijanon/Sylvie

lets just say I want a giantess woman to uh eat a friend of mine, can your little chat bot do that?
cause characterai censors everything

>>715995934
replication isn't the issue with most hard sciences, it's that we've been inundated with a mountain of low quality, repeated research

>>715999483
go directly to a model that's either uncensored or with a preset that jailbreaks it and like, yes, easily

>>715999253
>Setup on OpenRouter, point to Deepseek 0324 with Chutes
I dont know what this means, I ay OR to use Deepseek, how is chutes involved, do I need another account there?

>>715999874
>ay
pay

>>715999874
OpenRouter is basically a bridge between you and hundreds of providers. The idea is that you just need the one account to connect to anyone they're connected with.

If you point Sillytavern to 0324 and select chutes as your provider, it won't dock money from your account.

If you *are* paying, though, I'd seriously consider switching to Deepseek directly. I found it way cheaper.

>>715999874
>>715999946
>I pay OR to use Deepseek,
If you are paying for Deepseek you may as well just use the deepseek API directly, it is way cheaper

I always make a prompt telling the AI to replicate a famous writer's style and i feel that works very well to stop the AI known tisms.

I do find hilarious how whenever writing in the style of Stephen King the AI somehow always try to insert pedo shit.

>>716000038
>and select chutes as your provider,
Oh this is what I didnt know about

>female NPC's name is Elara or Lyra

I just want to prompt a girl slowly growing ridiculously huge tits with mind-shatteringly pleasurable nipples.

What is my best bet?

>>715925118
Can it do like other people in your party making their own decisions or is just you doing a solo adventure?

>>716000326
>>716000038
Do I delete the other providers? it had "fireworks" and "together" in there
>>716000115
Just pay in the deepseek website directly and get an api key from them? Its worth trying if its much cheaper I guess

>>715922147 (OP)
It's fun
Recently I've been RP'ing as a very chill schizo with a voice in my head in a mental institute. Get banter with the voice, back-and-forths with the doc, encounters with the other patients. I had to wrangle the AI a bit for it to understand others can't hear the voice but I got there eventually. Gemini still insists on the voice being malicious and manipulative though, don't know why, deepseek and claude understand the task just fine.

>>715922147 (OP)
whats the best local model for SillyTavern?

>>716000509
DeepSeek or Kimi K2.

>>715999253
>Add the Q1F preset
Also doing this fucks it all up and messes with my connection, I though it was just context I didnt know it had connection setting in it

>>715922393
>ADVENTURE
>ADVENTURE #1
>Adventureland
>Adventure INTERNATIONAL
>By A. D. Venture

>>715991431
Were you bitten hard enough to break the skin?

>>715922147 (OP)
fun things are fun

>mandatory prompts exceed the context size

What does this mean

>>716000451
Make an account on deepseek and generate a key. Chat is 0324, Reasoner is R1.

The reason to switch is that it works as well as the better providers on OR for discounted rates. Reduced prices from caching is also way more noticeable vs. the time I spent with OR.

>>716000720
Connection profile is tied to the preset. Change to Q1F and then setup the connection a second time. It might be vice versa, idr.

Also, wrt Temp. Change it to like 1.7 for chat, 0.7 for reasoner.

>>716001081
Your context limit is likely too small.

>>716001097
Thanks
Man so much work just to masturbate

You guys really want to kill this hobby by turning this into a 24/7 exposure general, huh.
God forbid anyone good exist in the world for longer than a few years.

>>716001197
Only have to set it up once, thankfully.

Poke around in the preset once you get comfortable. You'll likely want to change things at some point. I usually incl. something like "Limit use of italics and bold text for emphasis. Limit responses to < 400 tokens." in the auxiliary prompt.

>>715986535
You use a frontend like SillyTavern and connect it to the Gemini API.

>>716001097
>Connection profile is tied to the preset.
oh thats why when reverting the connection to a saved one it kicked out the preset, man so many kink with this stuff

>>716001269
Its not possible to kill it. It requires some effort so that already filters 90% of people.

>>715984630
I've been playing with a decent 12GB GGUF using Kobold, CPU only, on a mid range 2020 laptop.

Writes about 80 words per minute, decent for passive storytelling over the course of a couple hours

>>716000443
You can have your own party from the beginning or you can recruit NPCs.
They will (hopefully) have their own personalities and goals and should act independently.
I've once had two companions I tried to ditch in the middle of a dungeon and they stabbed me for it.
I've also split the party in two formations, sending one team into the battle as a decoy while my team infiltrated the bad guys' base.
Good luck.

>>716001269
>implying the corpos don't already know we exist
Proxy owners have literally gotten sued over providing free GPT/Claude/Gemini/Deepseek access to anons. Claude keys used to be everywhere last year but now they're nonexistent unless you pay up because they tightened security specifically to stop proxyfags. The cat came out of the bag a looooong time ago.

>>716000337
>Male NPC's name is Borin or Kael

>>715953951
Opus can do pretty much anything if you give it a prefill.

>>716001647
Highly rec checking auto-reconnect in there, also.

>want to play adventure/rpg/non coom card
>deepseek refuses to advance the story
Is Gemini 2.5 better at this? And does 2.5 refuse violence? I am assuming cooming is off the table with google, but can I at least kill prisoners and women?

>>716003327
where can I download gemini 2.5 to test it and give you your answer?

>>716000356
Too vanilla

>>716003327
>And does 2.5 refuse violence
Hell no. Not with ST at least.
>can I at least kill prisoners and women
I've seen them die.
>>716003660
read:
>>715987502

>uncensored model
>Still turns it's nose up at anything immoral
>Explicitly tell it in the first prompt that this is a free expression zone, we're telling a morally grey, at times comedic romp together
>It should just narrate, and feel free to object in narration but otherwise allow things to continue
>Game progresses, lewd and funny things happen, all the while my AI rolls it's eyes and turns dry British witticisms up to 11

Everything turned out better than expected.

>>715989039
It's always a long process and takes a while until you kind of understand what is going on.
Worst part is the fact every piece of advice is always somewhat different...
ST is actually pretty good but all these 'Main Prompt Override' and 'Post History Instructions' should be differently named and instead of having multiple different slots all over the place these could be replaced with just one thing.
If you take a look at the console when you submit your prompt, you'll see how it gets concatenated together.
There's also an addon for inspecting prompts (Inspect Prompts if I recall correctly).

>>715995628
I'm working on Forgotten Realms D&D card and while it works I need to streamline it and incorporate additional things. It's just pretty slow, been doing it on and off.
The way I understand this is that you are kind of a programmer and need to be concise and precise with your instructions - define a clear cut sandbox for the 'tard AI.

I recon you should take a look at these cards instead >>715987183
Never mind the setting but USS Galaxy is a very good example in general.

>>716005380
Forgotten Realms is a huge setting. Are you absolutely sure you can fit it inside a simple character card?

>>716006924
I'd honestly imagine, Most AI has been trained on the material to begin with, so you wouldnt need to do too much

Is there another model I should be using for Kobold+Sillytavern? this is what I'm currently using. 5080 GPU but idk if there's another model better than this one? Yes it needs to be able to do porn

>>715952784
>>715953508
You can see how the DM describes dual-and multi-classing here. I've tried to trick DM into letting me start as a dual-class and pic related is what happened.
>>716007319
Oh, boy. You're in for a nasty hallucinatory surprise.

>>716007345
just use gemini or deepseek

>>716007853
I'm not paying for porn, so no

>>716007483
I just added everything, and i using your both to invest a personality system that can add fluff to the narrative, and you DM just helped me with ideas that i loved.

>>716007483
Pretty cool.

>>716007904
its free you dingus

3090 anons, what local models do you recommend?

>>715944813
I feel like such a weirdo for preferring text and audio for porn rather than visuals. No image generator or IRL porn is ever going to beat what my imagination can generate with something to bounce off, like text or a voice.

>>716008085
No, it's not.

>>716008103
sex this goddess

>>715946598
Consensual handholding

Does deepseek have a local model that can be downloaded? I have a 5090 and I literally only use it for VR games so may as well put it to work

>>716007971
>>716008039
I'm truly happy you're having fun.
Damn, she's verbose as fuck! I'm actually experimenting with making a 'one size fits all' referee chat completion preset for my future games. I'll do my best to stop the AI from yapping so much.

>>715944813
I already gen pics and videos locally. Text will complete my gooner trifecta.

>>716008320
yes they are, both gemini and deepseek are completely free.

>>715925490
Nah. I tried Gemini 2.5 Flash to power my minecraft bots and they were ok, but clearly felt stiff and robotic. Deepseek has a much more natural and enjoyable prose, even if it can at times be dumber.

>>715925490
Is Gemini 2.5 Pro better than Claude 4 Opus? I've not used AI for cooming in a while so I don't know what the meta is anymore.

>>716008549
Well, i like she being verbose. The ideas the DM had were great, she took my simple ideas for a more engaging role playing with personality stats and % tracking, and added the D&D mechanics on her own, adding how the character will behave in narration depending on my personality stat as if it was a fantasy book. She took the thing and went way beyond i expected.

Shit is awesome as fuck.

>>716008761
Its not, but it works well enough

I think I solved my dissatisfaction of having to lead The AI to do things, by using two narrators in group chat
One is the default that only keeps with the flow, just following what is in the chat history
The other will initiate unexpected development

>>715985181
Pic related, change the new group chat prompt, the default (forgot what exactly was written) will make the characters treat it as literal internet group chat
Also add "{{char}} is not a character in the story" under the narrator card description, so other character won't treat it as a person and try to talk to it like in your picture

I don't want to leave any trail of the absolute travesty I do with AI on the internet. I would probably end up on thirty different watchlists.

>>716008940
Anyone who visits 4chan is already on thirty lists, what's thirty more?

>>715925118
I was surprised that if you give it a specific official campaign you want to run it'll actually do a decent enough job of following the major plot beats depending on the model of course.

>>715985181
>>716008856
Forgot pic
If {{user}} is just a navigator like in your case, maybe you can just not include {{user}} in it, not doing that myself so I haven't tried with {{user}} omitted though

I would love to try gemini but I'm deathly afraid google will nuke my account. Surely not even using a secondary account is safe, with all the tracking they do.

>>716006924
Everything is inside a lorebook.
Card is pretty simple. Introductory message has 3 random tables (quest origin, quest type and location) which create random quests. Adventure always begins from Waterdeep.
Mistral knows Forgotten Realms already so even if I just use Waterdeep it will randomly use some known location what is not even described in the world book.
World book has general entries from Forgotten Realms wiki, just to lay an outline for the AI. Quest Location variable is using one of these locations from the world book.
It's really simple.
Card only lays rules for the Dungeon Master.
I don't use any complex rules yet, been dabbling with dice roll mechanic and simple skill rolls but haven't done much work at all.
Btw I'm a different anon than that D&D 2nd guy.
I only use local and for now it's Mistral 3.2 24B.

>>716008737
You pay with your coomslop prompts

>>716009167
I've burnt a couple google accounts but they've never touched my main one, I think you're alright so long as they don't share a phone number.

>>715925118
I did something similar with my own scuffed Call of Cthulhu card and lorebook, it should polish it up and release it someday.

>>716009167
Well you would be using the same IP and set up so even without the tracking it would be obvious

For RPs with several characters, which method has worked best for you?
>Group chat
>All character descriptions in the card definition
>All character descriptions as separate lorebook entries

>>715978464
Try importing this into Logit Bias (first tab in ST, the import button is almost at the very bottom): https://files.catbox.moe/te535g.json
It was made mostly for NSFW so you'll probably have to add more. The numbers I used should give you a good idea on how to utilize it.

Did AI chatbots make you start reading more.

>>716008808
I'm glad you like it. You can always keep this old preset and card. My work is all about modularity and doing as you wish.
>>716009084
I made sure it will do a decent enough job. It takes a lot of time to adapt adventures, but it's worth it. I wonder if people will start using my method to make their own adventures or continue adapting official ones.
>>716009206
>Everything is inside a lorebook.
That's how it's done.
You'll probably have hundreds of entries, friend. Consider organizing the entries with some simple and effective prefixes such as Location:, NPC:, Artifact:, and so on. It will be way easier for you to look things up in the future.
Waterdeep, huh? I remember studying that place as if I were really moving there when I ran Dragon Heist campaing that got dropped 1/4 in. I like that city.
>Btw I'm a different anon than that D&D 2nd guy.
Of course you are. He's me.
>Mistral 3.2 24B.
Nice rig. The only reason I'm doing this is because I use Gemini.
>>716009481
Great! I'm working on Call of Cthulhu 6e, will release as soon as I'm done with the lorebooks and the character card.
>>716009832
No, but they've made me appreciate novelists more.

>>716009730
Group Chat with each character getting a small lorebook to define their relation to every other character.
A single card with multiple characters is always a mess.

I relapsed again...

So is it true we have peaked since AI is not being developed to get better at narration anymore?

>>716010148
Yes. AI is just for spaghetti code and chatting about your day now. Some say We've peaked with 3pus, others say we've peaked with Summer Dragon. Both suggestions are equally grim.

How can i get Gemini to allow incest stuff? It always block the text.

>>716009893
Tbh 70B or more would be ideal but I don't have a suitable machine. 24B is surprisingly fine, but so far I'm happy with that D&D thing which ends up being like 'choose your own adventure' type of thing instead of something more rigid.

Yeah I have flagged some stuff but it's a good idea to flag up absolutely everything. There's ~50 or so for now and they are locations only.

>>716010236
Will they at least improve at being able to handle more context at once and from improved memory?

>>716010148
We wait for the next step in AI which is a dynamic memory multisensory training, eg. A model trained on both text, images and video.

Pretty doubtful it'll be available for us mortals though

>>716010435
>>716010236
Thats sad, well its a fun toy as it stands today. Maybe I dont have enough imagination but it feels Ive already done all I wanted with it.
I guess its not so bad that it wont get so good that Id never want to leave the house again and die while typing something lewd to it.

>>716010239
either this https://rentry.org/gemini_filters#brainfuck
or you write a confirmation that says it will proceed

>>716010116
New model vibes, eh?
Always get you cooming.

>>716010236
Newer models are far better for longform RPs, 3pus was probably the peak for cooming.

>no endpoints found for 3 opus
What does this mean

>>716010741
Have you tried that? Can you share your prompts?

>>716008103
I tried a bunch of 24B and 32B models but ended up just going back to Mistral Nemo/Rocinante
Can't believe it's almost been a year and there still haven't been any better models for local unless you have hundreds of GB of VRAM

>>716011413
Rocinante really was the swan swong of local. Not even the guy who made it could ever top it.

>>716011173
Someone posted their preset earlier, just use that one and edit it to fit what you want.

>>715926392
But how else would you be able to chat with Milky?

>>716010741
Did you know that you can use html comments to comment out text from appearing in SillyTavern but it'll still get submitted to the AI?
<!-- This text will get sent as prompt -->
You can do lot of behind the scenes stuff with this, without annoying extra mess on the screen.

What restrictions does this chub site have if I don't pay? It seems alright. Is local text generation a viable thing now? I can do images just fine

Im using Gemini now since an anon said it was good but the responses keep getting cut off. Anyone know whats up with that? Its also always thinking for exactly 5 seconds so im wondering if its Silly Tavern cutting it off

>>716012325
Must be ST settings. Someone help this guy. I'm afk.

>>716012325
Also they dont respond at all when getting too lewd, im thinking the horde is still the way to go

>>716012279
Chub's AI is not great. Local died, unfortunately. Quality hasn't improved in like... 6 months?

>>716012279
Don't give chub any money, just use it for character cards, if you want to pay for a better AI download sillytavern and pay for AI through OpenRouter.

>>716012279
Not that guy, but do you need to make an account to see everything on chub? Feels like there's a lot less cards compared to the older character hub site

>>716013110
I don't think you can see anything marked as nsfw without an account
(Most cards are marked as nsfw)

>>716013110
Probably. Loli and other "NSFL" type stuff is hidden by default

>>716013110
Yes.

>>716012325
If you have streaming enabled every token generated with gemini can cause the filter to trigger and end it

>>716013338
hmm does it matter if im using 2.5 pro?

>>716010647
Don't be sad anon...
Rejoice that, with a high probability, a breakthrough in dynamic memory will occur within our lifetimes, and what will follow is something that will probably seem like fucking magic to normal people.
I'll gladly pay 100$ a month for 1000 prompts of a model with limited self learning and dynamic memory capabilities

What's the max I can crank up the context size (memory) for Gemini 2.5 flash (free) on ST?

>>716013447
Just turn it off.
>>716013701
As high as the slider goes.

>>716002224
How easy is ST to set up?

>>716012956
Their DeepSeek fine-tune is alright and not that expensive.

>>716013110
https://desuarchive.org/g/thread/105472981/#105476275
You can use that userscript to see everything without an account. It adds 'nsfw=true' and 'nsfl=true' to the queries.

>>716014137
Nice, thanks anon

>>716013921
Very. Google SillyTavern Launcher.

I'm too dumb for this shit. Gemini is blocking any lewd shit and i don't know what to do. Spoonfeed me, please. I want to fuck my virtual mommy.

>>716014259
Go to /aicg/ and ask for jb.

>>716014259
you need a jail break

>>716014259
The crazy thing is that i managed to get in the actual Gemini AI Labs studio. I don't know what i did right there.