no. next

AI role-playing isn't even the future of role-playing.
RP requires the narrator to keep details consistent and not change them every twenty seconds. No AI model is capable of this.

>>743192821
unusable unless paying for a good API's

>>>/g/aicg/
>>>/vg/aicg/

No, it's porn.

>>743192821
>cheap
Depends entirely on what you're using. Deepseek or GLM, yes. Claude, LMAO

>>743192821
My knuckles whitened as the appearance of this thread struck me like a physical blow.

>>743192821
Has anyone actually managed to get an interesting story out of it? I just use it for gooning

>>743193232
Why do we see these same sentences regardless of what model is used?

>>743192821
until I can have 500B+ model intelligence on ~16GB vram, no it is not

>>743193390
trained on the same widespread low quality smut novels
its sad enough to send shivers down your spine

okay for writing, chat style is retarded tho, also too high iq for most people

>>743192821
"RPing" with AI is purely masturbatory. RP should be an interaction with others, whether with other players at a table or with a story written and programmed by another person where you have to confront their ideas. AI is not a "partner" but an amalgam of every bit of content it could scoop up with no unifying principle

>>743193232
And that's not nothing.

>>743193696
>"RPing" with AI is purely masturbatory
That's the point, yes

>>743193232
I looked at this post.

>>743193390
low-quality bots that are written poorly will rely on generalized context in order to "fill in the gaps".
That then produces the most likely response (an LLM-ism), on the basis of that generalized context, and then pollutes your chat history.

When it pollutes your chat history, the ai will then use it as context for even more responses, leading to a reinforcement loop.

Long story short: get better character cards that are written "in character" and provides accurate context to the character at hand.

This is unfortunately difficult to do, and most jeets suck at it.

>>743192937
>cant keep quants on-topic
skill issue

>>743193390
Every AI is a mediocrity generator. At any given point it tries to give you the most mediocre answer. That's literally how it works and there's no way around it.
This is exactly why you should stay the fuck away from AI stocks

Local LLMs suck at writing, and if you disagree with this, you are a porn brained imbecile with catastrophically low standards.

>>743192821
Easy to install is a steam game. One click and it starts and finishes. This is way too complicated for a computer illiterate retard like me.

I do use sillytavern pretty regularly but I don't have any faith that AI will ever be good. It's convenient and it's certainly better than playing with a shitter, but the quality of the writing just plain isn't that high and good luck getting it to keep track of any complicated mechanics. It's best treated as a novelty. Drop 2 bucks on deepseek, enjoy it for a week and then move into the next distraction.

>>743192821
I gooned to this for a week straight, best fucking thing ever barely slept as I came 9 times a day
after that I stopped caring it's really shit for anything long term but amazing for quick novella fenders of your weirdest fetishes

>>743196092
You know, there are nicer ways to ask to be spoonfed how to properly set up your local model.

Regardless of whether you are local or paying for the newest and hottest model, you get the same problem. Its extremely predictable and every single character falls into the exact same tropes after a couple of messages.

>>743196346
The problem isn't setting it up. Any idiot should be able to click on a couple of .BAT files.

The problem is that local models aren't as good at writing as cloud based models. None of them, unless you want the cheapest, most generic porn imaginable, in which case you are subhuman and deserve the rope.

>>743196231
they recently discovered the J-Space, so in the not so distant future (couple weeks) it will be possible to finetune LLMs so they become more in character.

>>743193232
I watched his knuckles whiten, my voice, barely above a whisper "{{user}}..."

File: 1783074082315842.jpg (48 KB, 558x408)

48 KB JPG

>>743192821
>Switch from local to OR
>...just in time for censorship to kick in, with China basically banning AI cooming
Some redditor had his SillyTavern prompt labeled as CSAM by one of the providers the other day. I have a feeling that the wild west days of AI are about to end. Then again my shit isn't nearly as objectionable as many.

>>743196520
whats J-space? is it like to make context easier to parse or something? ive noticed any model i use will give a ton of importance to the beginning and end of the context, but very little to the middle where the real mearmt and potatoes of the rp session live

>Ask GPT 5.6 to remake a video games
>It fucking does it

>>743193390
At first a lot of it was just commonly written stuff that got caught by AI models during "training" so thats what they put out.
Now a lot of AI models are being trained on AI generated output which puts it into a feedback loop.

>>743196805
they discovered AI has a thinking process and how to steer that process.

>>743193232
The edge of my mouth twitched. Not quite a smile. Something in between.

>>743193935
It wasn't nothing, it was the absence of something.

>>743196719
if deepseek goes im whitening my knuckles

>>743196805
it's just the list of possible responses in increasing order of likeliness/match. AI scam artists are claiming it's some magical arising intelligent behavior from AI, similar to how they lie about the concept of hallucinations to imply that it's the AI being so intelligent that it goes schizo, rather than the fact that AI simply takes whatever response the math arbitrarily declares is the most correct, even if it isn't

>>743197360
Xi is already making Chinese providers SHUT IT DOWN with aggressive guardrails. Providers outside of China seem unaffected and can still run the latest Deepseek normally. I don't know what future Deepseek models are going to look like though. The Chinese laws also aren't specific to things that would be considered morally objectionable, they basically don't want their work slaves to have AI girlfriends.

>>743193346
>I just use it for gooning
Same here. It's only good when you turn your brain off to focus on sex. It's just not worthwhile otherwise.

>>743197475
>they basically don't want their work slaves to have AI girlfriends.
extremely retarded take for civvie control, but I guess they do have enough foresight to try and correct demographic collapse while they can.

>>743197475
oh right because they will learn independence from them

>>743193390
Their training data containing poorly written fictions is one thing. Initially some LLMs managed to avoid some of these samples, but then now all of them share this incestuous samples including synthetic ones they wrote themselves, the quality has been dropping all over.

>>743197550
>extremely retarded take for civvie control, but I guess they do have enough foresight to try and correct demographic collapse while they can.
>>743197621
Basically they are already in demographic collapse, and they also cannot rely on forced thirdie immigration the way that the West and Japan is trying to do, as the Chinese are extremely xenophobic. So they have outlawed digital girlfriends, and that also pre-emptively outlawed sexbots (not kidding).

>>743196719
>>743197475
ive had glm say that shit too, but the worst they do is say no to the prompt, you regenerate, and it eventually accepts it and spits out what you want.

>>743197876
pretty smart. nips should follow suit. it's over for westoids unless we repeal female voting rights

>>743192821
>easy to install
I am too retarded to do this

>>743199195
How tragic

File: h5ztnknucywf1.png (75 KB, 320x438)

75 KB PNG

>>743192821
It'll be the future when it's capable of generating good quality images to depict what's happening in the story, for now it's ok but only text is not the future yet.

When it becomes able to dynamic generate images of the characters present in the roleplay, portraying them in the situations that they got into, then yes the future will have arrived.

>>743199195
bro its literally 3 mouseclicks

>>743192821
>cheap
How cheap is cheap? I tried it, and liked Kimi K2.5, but I don't know how much it would cost monthly

>>743193232
Smells like ozone.

>>743199938
I've had an RP going with Deepseek v4 Pro on OpenRouter, and it's above 300 messages now which only cost about $1.50 (maybe less, I did fuck around with some other stuff).

>>743199938
DeepSeek and GLM are basically free, it will take you 6+ months minimum to burn through $20

They're also very sloppy. If you want to use opus or something you could burn through $20 a day easily.

There are lots of options at all price ranges with varying degrees of annoying slop. Most of the cheaper ones are pretty stupid but DeepSeek is fairly smart. GLM starts out smart but gets confused easily. Kimi has decent writing but it has zero comprehension and just forcefits things it feels like writing.

>>743196719
Does he? I still have 5$ loaded.

>>743199650
It can do that though, it's just very complicated and it takes too long to generate good enough images, even on good hardware.

>>743196719
How fucking retarded are you to ever do anything truly degenerate online like this, even through 'chinese servers'?
Do you faggots not understand how data works? It doesn't go from your PC to china, it goes through many other steps that are not in china, and which even if they might not have complete control over, they can/will have in the future.

I'd argue that even doing local degen shit on windows is risky, but microsoft is usually slow on the uptake when it comes to these things so it's probably still fine for now, but in ten years time?
Fuck no lmao

>>743200869
Like I said, my shit isn't that degen, certainly not something that would get flagged like that redditor. But I have a feeling that even completely vanilla shit might get locked out in the near future.

Does kimi 2.6 ever ignore it's guard rails? Glm 5.2 I can basically get anything brute forced through with enough swiping but Kimi is so fucking strict.

>>743201058
Oh, I was using the royal (you) here senpai, you were just the one that brought up the subject is all.
Though it doesn't matter because what you think 'isn't degenerate' might be seen as degenerate in the near future anyways, cultural shifts occur faster and faster.

Do not use online APIs to do anything sexual with them, is probably the best advice that's worth being given in such threads.

It's the future in the sense that it sure isn't the present. The writing is generic and it keeps veering off track, and the AI isn't smart enough to comprehend interesting concepts and play them out over the course of a longer roleplay.
There might be a time when all this is fixed in an AI that I can run locally to avoid censorship, but that is still far away.

>>743193232
I planted a kiss on his pulse point
"Mine"

>>743202647
I haven't seen that one in a while admittedly.

>>743192937
i like how normalfags (aka: /v/) only knowledge of AI is when they use a free 4B model to cheat on their homework

What's with all these controversies surrounding AI roleplayers?

>>743196719
The Big Brother is watching you fap to your text

>>743193232
"A blow?" Anon McVampireCEO traced a finger along the curve of your jaw. He wasn't gripping, nor pulling, just...holding it there, as if to remind you of his presence. "You're saying it struck you like a physical blow."

>>743202880
The usual when niche things are getting popular among normalfags who can't separate fiction from reality.

Perfect, simply perfect.

File: ai_6jw0w4nm.mp4 (307 KB, 720x720)

307 KB MP4

https://www.youtube.com/watch?v=II59lFGElKE
People can tell when you're using AI religiously for anything, btw.
So if you're still an untalented, uncreative fag, it still won't make you any better.
Best option would be if someone who DOES actually have talent and the creativity to properly create and put something into practice uses it as a background tool or as a side-tool for research, but then crosschecks with other things still and doesn't fully rely on it.

>>743193232
Anon's fragrance clung to everything in the room. The scent of sweat, stale cum, and something uniquely him.

>>743196719
>weeh weeh i can't engage with and pleasure my pedophilic fantasies anymore
>this is bad because.... uhhhh... MAP RIGHTS!!!!
get fucked you worthless pedophiles. "free speech" doesn't protect you from having to face the consequences of your degenerate and criminal proclivities.

File: 1747592377171354.jpg (5 KB, 160x149)

5 KB JPG

>>743196719
>decide to try fapping with AI chatbots
>start with Chub
>greatly enjoy it
>even make my own bots
>chub bans cunny and then gets a 20 message daily limit a week later
>migrate to OR because I found Owl Alpha and it was free
>Owl Alpha stops existing 3 days later
>bite the bullet and buy some credits
>this happens
Sorry guys I'm the one bringing bad luck to you all, I need to immediately commit sudoku so things go right again

>>743203387
do the pic related you posted but to your own head instead you handicapped pedophile

>>743203236
Funfact: It's estimated water usage will grow to about ten times that much by next year, and continue to rapidly grow.

>>743203562
I wish I was handicapped so I could get free gibs, now I turned to solo RPG and journaling, been doing some vile shit on cunnyvile with hypnosis and magic, pretty fun!

I’ve told this story before, but I’ll tell it again.
My experience with LLMs:

Day 1: I was Fox Mulder, and I was investigating something with Scully. Scully was acting super professionally, but I just kept screwing up, making weird and horny comments, or trying to solve the mystery in the most idiotic ways possible. Scully's reactions were genuinely interesting and funny, and I thought I'd found the best pastime in the world.

Day 2: I tried again with Scully, but this time I tried to do something more interesting, so I had to keep rewriting what the LLM generated over and over, and that started to get tiring. I tried other characters; I even appeared in an episode of Seinfeld, the most normalfaggot and Jewish thing I've ever done in my life...

Day 3: I tried again, but I quickly got bored. Tthe novelty had worn off EXTREMELY fast, and before I knew it, I was playing fucking Doom 2 or A-Train again, games that my old GTX 980 was more than capable of running and keeping me entertained.

>>743193232
It wasn't just a blow, it was also physical.

>>743203759
>making weird and horny comments, or trying to solve the mystery in the most idiotic ways possible
that's lore accurate for fox "sex addict" mulder

>>743203759
If you're trying to do cool stuff instead of a standard roleplay you're going to need to make a custom system prompt. It's non-zero effort, but not much different from making your own card and it's really powerful.

>>743196719
Does it make much of a difference to use online resources instead of local models?

>>743204475
Online is much faster, and unless you have a ton of vram and ram available, you get better replies as well.

>>743202880
we are dark roleplayers

I'm downloading everything to see if my experience with Dana Scully described above was just because I started too early, when the tools were still rudimentary.
I'm downloading a 14b LLM called Qwen.coder or some shiit like that.
Let's see if role-playing has improved since the early days...

>>743193390
To make a long story short: 99% of it can be blamed on the Wildchat RLHF dataset
Same with the reoccurring names, the names like Elara were weighted higher because they "aren't copyrighted characters"

>>743199650
>>743200682
I'm able to do that running locally with Gemma4 QAT/MTP and Anima simultaneously on the same 3090. With the turbo lora, it only takes a couple seconds to generate an image of the scenario from SillyTavern.
The results aren't always great since it's dependent on the language model coming up with an image gen prompt that Anima understands. Usually requires fiddling with the prompt beforehand a bit, but overall it's not too bad.

>>743206269
proofs

>>743205519
check out Gemma4

>>743206342
Not at my main PC right now, but Gemma4 uses up roughly 20 GB of VRAM at Q4
Anima is pretty light on VRAM so it's able to fit in the remaining space without any major slowdown

>>743206442
31b?
i dont remember the sizes off the top of my head

>>743206484
Yeah 31B QAT

>>743193232
Are you ready to start this journey together, Anon?

I have made so much porn with grok and sora 2 when it was around that im pretty sure my future is completely fucked and possibly full of blackmail. I'm probably going to have to kill myself.

Not to mention the thousands of messages ive sent to gpt having it make smut and vore and all that shit. I'm so totally screwed.

>>743193232
I let out a breathy laugh as I reposition slightly on your lap and say, "you're going to need to increase your context window, anon."

Then I ease off your lap and walk to the door. At the threshold, I turn my head and say, "oh and one more thing: Make sure your temperature is sufficiently high; around 1.2 to 1.5."

>>743203664
so it'll go from .5% to 1%?

Marcus Vance

What are some good alternatives to ST? It hasn't been updated in 2 months and it seems like there's been a lot of vibecoded alternatives shitted out recently

Has anyone else noticed the angry troon that cries pedophilia in every thread

File: 1766086143433621.png (1.07 MB, 832x1216)

1.07 MB PNG

>>743203341
Blame the redditor, my characters are all adults. But if the Chinese decide that AI girlfriends aren't allowed, that shuts the door on pretty much everything. I think the law specifically called out AI being used for "emotional support."
>>743204475
Even if you can run something like Gemma 31b it will be of lower quality compared to something like Deepseek v4 Pro, and way slower. Even on my 5090 Gemma 31b takes a good 2+ minutes to generate a response.

File: 1752323610295004s.jpg (9 KB, 250x250)

9 KB JPG

>>743192821
>easy to install
I realize there's a guide but you might as well told me cleaning a lawn mower engine is easy.

Where's the best place to find characters? In other words, where you don't have to create an account, and there's NSFW content, famous characters, etc.?

No good harness exists for rp right now.
>sillytavern
shit UX, shit usability, bloated node app.
>marinara
buggy shit and the creator has an ego
>sillybunny
despite proudly declaring itself a fork the UX is as just, if not more, shit than sillytavern

>>743207514
if you are just looking to dl characters, i think c.hub will allow you to do so.

https://chub.ai/characters

>>743207514
Ordered from no rules -> safe horny.

https://botbooru.com/
https://chub.ai/
https://jannyai.com/

File: become uncageable.gif (2.3 MB, 300x300)

2.3 MB GIF

When the fuck is longcat 2 going on openrouter? I'm too retarded for anything else and I just wanna get back to my last-scion-of-a-minor-noble-house x huge titty onee-san maid /ss/ RP on janitor.

>>743207609
git gud

>>743207609
ST is the best one by a long shot but it overcomplicates things. RetroArch tier of inconvenience.

Shill thread

I hate it when I say something, and instead of just giving me a short answer, Silly Tavern keeps writing a bunch of extra stuff, including other lines for my character. What a poiece of shit, honestly...

>>743207609
Well... there's lumiverse.

>>743208970
Assuming you're not using a sub 40 IQ model that will be pink elephanted, have you considered asking it to generate shorter responses in your context?

File: 1000014826.gif (1 MB, 250x250)

1 MB GIF

If I switch from OR to local, what model should I use and what size? Ive got an RX 7800 xt 16gigs of vram and i normally use shit like glm and deepseak.

I enjoyed AI Dungeon in 2020 but never managed to get the novelty back even though the models are clearly better now. I've been messing with local imagegen recently and my rig can handle that without too much waiting (about 45 seconds per gen), would local gen be good on my rig? I have a RTX 5060 with 8GB VRAM, not ideal but serviceable generating images

>>743212051
local text is just not good almost no matter what hardware you have, at least compared to the high end models
it can be fun to play with just to see a machine sitting in your house actually able to hold a conversation, there's something weird about that in a way that connecting to a remote service in some huge data center doesn't really convey, but it's not really worth doing otherwise without 5 figures of investment in hardware
if you just want to try a better model Deepseek is pennies for hours of use, and can be found free if you dig around enough

>>743207698
>safe horny
It's honestly astonishing the terms you people come up with purely because something doesn't have your no limits hardcore loli rapebots.
You AI jeets are funny.

>>743212501
That's not even remotely the origin of 'safe horny'.

File: file.png (887 KB, 1236x820)

887 KB PNG

Have you guys had any success doing structured roleplay, with numbers and dice rolls and stuff? Basically AI Dungeons and Dragons?

>>743209406
Bloated shit, why the fuck does an AI LLM frotnend have a SMART disk information feature?

>>743196719
I can't wait for my loli scat logs (no pun intended) to be read in court

>>743212274
>and can be found free if you dig around enough
Are you just referring to taking another user's API key so their tokens get used, or are there actually free providers?

>>743214863
there's free providers but they come and go, venture capital burning startups will give you a few dollars in free credits for signing up and/or referring a friend, usually needing a gmail but not always, usually trying to get people to pay for some specific flavor of thing they're offering but also just reselling API access, sometimes through scuffed intermediaries that make prefills hard to use or have extra filtering sometimes just raw. They hang around until they run out of funds and then either shut down all the free shit or just close. Often they start offering Claude, Gemini and GLM and then cut those off but still have the really cheap ones like Deepseek for awhile before shuttering completely so Deepseek isn't that hard to find usually.

>>743192821
>cheap
The good models require a 5090 at the minimim

>>743193346
I did manage an entire day of zero gooning just slice of life as a medieval prince. It ended when the AI presented me with a scheming, evil noblewoman- I just had to rape her to assert dominance over the court.

>>743197550
>I guess they do have enough foresight to try and correct demographic collapse while they can.
No, they don't. Like every other modern government, they don't even comprehend the cause. The birthrate collapse, by the way, started decades before normies even knew what a PC was.

>>743193346
There's this card where you're a demon summoned by accident by this group of girls, when I first started playing with it I was gonna hypnosis rape them all but one of them interacted with me in such a funny way at the start that I dropped the idea and just went along with them and I have a lot of fun whenever I go back to it.

>>743203341
In a hundred years no one will think like you btw.

>>743192821
Stop telling /v/ normies about this shit because the more people know about AI cards and unrestricted ERP with chatbots, the sooner retarded leftie Karens are gunna lobby the govt to get this shit banned.
They're already playing whack-a-mole with AI card websites. Chub is dead and the booru will be next.

>>743196719
>I have a feeling that the wild west days of AI are about to end.
Been saying this for a while now.
Save whatever cards and models you can, cuz eventually you'll have to scan your driver's license to be able to use the internet or turn on your PC.

>>743216712
>Retarded leftie Karens
Chub was targeted by none other than Collective Shout. I hope there's no other victims.

>>743204475
Online is faster, has better memory and overall quality, but the guberment can see every sick twisted fetish you're engaging in and if you do anything illegal you'll probably get a door knock at some point.

Local has all the benefits of local, if you back your shit up you could gen for 100 years on an offline machine and no-one will ever know or care. But it requires a really good PC, its slow, has worse context memory and quality isnt as good

>>743213779
There are like 2 dozen websites/services/tools specifically for this.
But there's nothing yet for this, but local with boobs/fapping involved.
Best option is finding a really good RPG card that loosely keeps track of quests/stats/items/gold, but you'll have to manually handhold it a bit with creator's notes.
Also I've been searching for about a year and a half for a good AI RPG card and there's honestly not really anything out there, and 4chan is pretty useless for recommendations as well.

Your best bet is to spend a few weeks manually writing your own card and lorebooks and go from there being honest.

>>743202759
Even the best models degrade after ~15 messages, its still a massive open problem with llms, some models (claude) summarise the chat when it hits the context limit and starts a new chat in the background before it gets too retarted. But then its a problem of wether the summary was accurate.

I dunno what silly tavern does under the hood but I'm sure they have imperfect mitigations for this, i use agents at work and jetbrains (my ide) has changed its mind like 4 times now trying to optimise context with memory.md files and what not.

>>743216815
Chub is like the 2nd or 3rd AI card website at this point that'd been done in.
Even Character.ai and the normie SFW ones have been sued by Nintendo and other shitters for copyright among other things.

>>743216712
/v/ is too retarded to act on finding out about this
80% are gonna get filtred by the sillytavern install 10% filtered by just looking at the ui
8% entering an api
and 1.9% will get filtered if you ask them to use it for more than a day before forgetting about it forever

>>743217140
I think the best option is just to create your own card with that character card creator v2 bot. Tell it that you want it track that stuff, and make sure your using a thinking model. Also, i think there are some extensions for sillytavern that will track that stuff + character positions/recent moods/inventory on the backend.

>>743217916
Realistically like 80% of /v/ are third worlders still on a 1060ti.

>>743218816
try a gt 710 on my case, might get a rx 580 by the end of the year if things don't get worse

>>743219010
jesus dude, ive got a couple 580's just sitting around, ill just give one to you but you really need to get a newer card

File: 1753898986410219.png (1021 KB, 1080x1497)

1021 KB PNG

>>743193232
*His knuckles turned white as he typed out a sentence on the forum that he was posting on.*

"My knuckles whitened as the appearance of this thread"

*everyone waited for him to finish, undulating in fear at what his next words would be*

"struck me like a physical blow."

*everyone started shaking their heads side to side as they recognized the emaciated and gaunt sentence.*

*They thought about how they would respond, and the way they responded made the air freeze, you could hear a pin drop as the smell of sulfur and ozone leaked in.*

*but they were not alone.*

>>743193346
>Has anyone actually managed to get an interesting story out of it? I just use it for gooning
I often tell myself that today is the day I'll do a real, serious rp. No porn, but going on an actual adventure. But I never do it, I just go back to gooning.

>becomes unusable every night when the asians and indians wake up
how fun

File: nibbles.jpg (43 KB, 1200x675)

43 KB JPG

>>743213779
Black cat anon is apparently delivering the agentic frontend based on 3.5e later this month. Hopefully that will finally satisfy the AI RPG scene. I'm sure you'd still need lorebooks for sticking to the lore though.

As much as I enjoy my porn, the most enjoyable stories I've gotten out of it is when I take an over the top nsfw card and make up some ridiculous scenarios.
>I'm the free-use maid you hired, I fucking hate you
>Perfect. I need you as my arena partner in TBC Classic
>I have an extra rig set up for you, we need to reach 2.5k rating by the end of the month
>Grab the white monsters from the fridge, we're staying up all night

>>743219010
black friday sale, might get a good upgrade for cheap.

>>743224157
You mean cyber monday? Anyway, those two are only a thing in the States.

>>743193232
My hand hovered on the mouse as i stared into Anon's reply, not quite touching just, present, before i pulled it back immediately...

File: how do we tell him.jpg (94 KB, 720x480)

94 KB JPG

>>743193696
>"RPing" with AI is purely masturbatory.

File: 1771379980344506.gif (12 KB, 390x430)

12 KB GIF

>>743203341
>>743203562
t.

>>743192821
when will the shilling stop

>>743216949

I'm glad I'm in a shit third world country so no one will knock my door for all the loli porn slop I generate but I know eventually this will run out and get shut down by the source if this is true >>743196719

>>743206442
That just depends on the context size, it's all in the context size (and then to how many layers you offload to your GPU but that should always be 100% if you can afford it), but yes gemma4 QAT is honestly very good on that front, I can usually get 64k context on its fine tunes on my 4090 (24gb vram too), but doing that would mean much worse quality/speed for anima, but maybe I need to look into it (and it's my fault for wanting such high context, because I like my stories to go on for a long ass time)

You can most definitely cut that in half and get decent length stories still and have the vram to spare for image gen

>>743220065
kek if either of us weren't gonna get raped by tariffs because of brazil's importing taxes I'd genuinely accept that. I've only been playing older games or games with no fancy graphics since forever so I'm alright at least. Recently got into Siralim Ultimate and it's fun as fuck

>>743203341
>omg is this an underage AI you're fiddling over text messages?!
>THIS IS A CRIME!!!!!!!!!!!
holy shit bro if you saw that session I had last week with the two girls and that hypnotic succubus, and then that dog got involved too, that was so fucking hot

>>743207205
none. they're all shit projects that people shit out as concepts then realize the immense amount of work needed to have several configs for dozens of different models and all the options they support so they abandon it after posting their 'this is so much better than st' screenshot

File: 1465347947051.png (214 KB, 640x360)

214 KB PNG

>>743193184
>he doesn't host his own models

Which Gemma finetune would you guys recommend? I'm using TrevorJS/gemma-4-31B-it-uncensored right now.

>>743225826
Marinara Engine is confirmed by community to be the least shitty ST spinoff. It even has a hosted online only version if you hate installers.

>>743226152
Marinara is a vibe coded garbage

>>743207205
it's open source tho?
and what's the issue with ST?

>>743226152
what does it do that ST doesn't do and/or what's missing?

>>743226140
I always stick to hauhauCS's gemma forks personally, they just work and by now they're getting 0% refusals and they're pretty efficient

>>743200869
You have no idea how technology works or the basics of internet protocols.
>doing shit locally is risky
Why? You think microsoft collates that information? You can literally run wireshark and see what information your PC is sending anytime you want.
Are you a boomer or did you just get your first pc last year?

>>743226452
>niggerbrain making my point for me
What you believe they're doing now and what they can do are two different things, are you trying to tell me yo got wireshark running 24/7 on your PC?
What about when the government comes knocking at bill's door and tells them they need your information, you think he'll protect your rights?
You retards were babbling about this exact sort of shit for years until they all end up getting caught taking data they're not meant to take and sending that to places that they're not meant to be sent to, and you think the government wouldn't be that x100?

Mind you, feel free to keep diddling your AI kids on deepseek bro, I don't care, it's really not my problem, you'll just be part of the natural selection pool

You're too naive to be an adult, but you also might just be a retarded one.

>>743226140
dont even bother with the uncensored version, the original model is already the most uncensored model released in a while now and it doesn't need its brain picked any further unless its going to be a tune.

equinox is a gemma 4 31b tune from the mythomax/ai dungeon guy
https://huggingface.co/LatitudeGames/Equinox-31B-GGUF

File: 1780754011656086.jpg (261 KB, 1280x1260)

261 KB JPG

>>743226624
Again you don't know what you're talking about. You can audit what your PC is sending through a million different means and block them. Either locally, on the LAN or with a DNS service.
This is the problem with schizo's like you. You get a PC and think that some guy at microslop in india is looking at everything you do and making notes on it. Literal schizo tin foil hat retardation.
Also occupying servers something is routed through doesn't mean shit because most online data transmission is encrypted through HTTPS. Most governments and corporations already admit that encryption has won in terms of privacy so they aim to control the services you connect to or have other means of identification. This is why there has been such a push to ban encryption outright or VPNs.
Meanwhile here you are thinking microjeet is literally listening in on your PC and that the government saw your LLM requests to china because they occupy a server in the middle. This is such low IQ brainlet levels of internet understanding it's actually fucking hilarious.
Meanwhile the reality of it is that people using AI are giving up their details all the time with their prompts. Registering email addresses or worse paying for it. This is how the government tracks you because they have in roads with the banks for that info.
>feel free to keep diddling AI kids
I'm not a pedo and I don't use AI faggot. I simply saw you spouting total retarded garbage about a topic you know nothing about and came in here to tell you you're a total fucking loser pussy who doesn't know what he's talking about.
I'm into HAGS.

>>743226889
>you can just block it!!!
>what if you huhhhhh unplugged the ethernet cord from your PC??!
You're so stupid it's not even worth arguing, the point isn't that there's zero ways around it, the point is that anyone that has to ask such question isn't going to do that in the first place, as you can see ITT most retards use deepseek or other international models thinking 'oh I'm safe now', you're a fucking brainlet just the same, you're just saying 'uhmm ackshually you COULD evade from prison if you just do X and Y' like a peak redditard, stick to your lane

>>743226341
>>743226707
Thanks, I'll check them out.

>>743226974
Why would anyone think they're safe using an international model?
Just because the data transmitted is encrypted doesn't mean the payments to fucking deep seek aren't.
Governments track that shit and not only that but often these chinese companies can be compelled to give up information by governments.
No one's arguing it isn't retarded to do what they're doing.
I'm arguing you don't have a clue about internet protocols and encryption which is what you keep talking about you stupid bitch.
You're literally some jeet intern who doesn't have a fucking clue about how any of this shit works.
Most internet traffic is encrypted end to end so
>DA GOVERNMENT CAN SEE BECUZ YOU CONNECT TO SERVER DAT CONNECT TO DA CHEENA!
is a dumb retarded statement from someone who has no fucking clue what he's talking about.
Even worse you think microslop is spying on everything you're doing on your PC which is also fucking tinfoil retarded.
Keep being a dumb ass though so I can get paid to fix your jeet tard mistakes.

>>743226341
>hauhauCS
i read he basically stole the other guys uncensor stuff and tried to rename it, claim it as his own, but it was quickly found out.

he did actually accomplish something by himself, even if he stole stuff, his qwen uncensored is better than the guy he stole the code from which is some kind of accomplishment i guess.

>>743227141
I give zero fucks about who 'deserves' anything, I'm not paying for any of it either way, this sort of argument is on the level of 'you can't have a pikachu mod in your skyrim, what if nintendo didn't approve' tbqh
hauhau is just the popular one that most people seem to be using and I rarely see anyone complaining about it, and he's usually pretty fast at updating it when google updates the base models, so I'm using it

>>743227238
i don't care about how it happens either if the end product is good and his, for qwen at least, are the best. i was mentioning thats why he's unpopular among some others

>>743196719
I never got a refusal even with the most weird shit i could think. I have no idea what the fuck are you guys even trying

File: 1756324540350811.jpg (50 KB, 455x673)

50 KB JPG

>>743192821
i can already smell the future where thousands of AI agents are generating fully interactive 3d worlds in real time, token by token, as you dive ever deeper into your own personalized sex paradise. it'll be the bridge to a sword art online/matrix-tier experience. we're so fucking close, bros.

File: file.png (102 KB, 862x628)

102 KB PNG

>>743226152
Interesting ideas, I don't know what it's worth yet, but that seems like a cool idea

>>743227862
All I want is to live inside RPG worlds. I can't understand why making the fully immersive 3D world isn't the top priority of corporations and governments alike. 99% of the world's problems would be solved by people's deepest desires being satisfied.

>>743196719
AI is still in the phase of heavy R&D, and companies are finding better ways to monetize the tech. as it gets more popular among normies, compute will become the bottleneck (it already is, China can feel it), and most of the free shit you can do today will be locked behind the paywall. It's crazy to think grok imagine used to be free. this is a classic bait-and-switch strategy, and it always works.

>>743226707
>No QAT version yet
Equinox sounds good but only getting 22k context out of my 4090 feels fucking disgusting tbqh

>>743228025
you can do that using lorebooks/rag. i do it by playing as the card. i specifically tell the model i'm the card, but will use user as narrator. put characters into the lorebook instead. that lets you also say 'i take a walk alone' and the ai will go into reflecting and stuff, rather than be a straight text-message from the bot youre playing with. just by moving things around, you make the rp better for yourself

>>743228587
This. 22k is still nothing if you got used to Gemini.

>>743227862
I don't even need graphics. If the text gen part alone had perfect memory/understanding and could be run locally with no censorship, that would be enough for me until I die.

>>743196719
> labeled as CSAM
Told you this commie newspeak was pushed to replace "child pornography" because with enough mental gymnastics it can be applied to any content that features a child character.

>>743195902
I video game doesn't require the user to keep reminding the game of the rules it's supposed to hold you to.
That's called a board game.

>>743217156
It has an option to generate recaps for it at set rates.
It's also hit and miss. But you can read and edit them yourself

So I tried it yesterday, and to my fucking surprise, the quality is infinitely WORSE than when I tested Silly Tavern back when it was still new, even using the same character card from back then.
I don't know if it's because of the LLM I used (Qwen2.5-Coder-14B-Q8_0), but holy shit, Dana Scully couldn't even perform basic interactions. Things like:

Me: "Hey, let's stop to get some donuts before heading back to the office, how about it?"
LLM: "Mulder, I don't believe in aliens. Let's stop wasting time and get back to work."

Horrible.

File: 1775509547108275.gif (2 MB, 880x880)

2 MB GIF

>>743229760
>qwen
>coder
>14B

>>743229760
>coding model
>CODING
Are you retarded?

>>743229878
>>743229853
The way it works is the same, don't kid yourselves. Do you really think this LLM wasn't trained on billions of pages of literature?
Well, I'm downloading an uncensored version of Gemma 4 right now, let's see if it's any less garbage.
It amazes me that you guys like this crap so much.

>>743229760
>qwen 2.5 coder
that shit is old as fuck but it was pretty good @ 32b

try the 27b modern. its not a code model but will do most stupid projects you retards throw at it

>>743227567
I've had a "Content Exists Risk" error from Deepseek API but it might have been from a prompt formatting issue rather than censoring anything because the content wasn't even NSFW.

>>743229064
It makes even less sense. You can argue that a drawn/genned pic is still pornography, but you can't argue it is the product of "child abuse"

File: 1771647431626509.png (403 KB, 535x637)

403 KB PNG

why is this pajeetslop being shilled here so regularly?

>>743229990
>try the 27b modern.
Is 24 GB of VRAM enough for an LLM of this size?

>>743229968
I'd bet my head on the fact that LLMs weren't really trained on novels. Maybe a few pirated works fro Jstor, and fanfics, but it's all trained on Wikipedia and countless news articles.

>>743230109
Gemma is far better optimized for 24gb cards (you can run a modern gemma 4 at 64k context easy on their 26b model, 31b is still a bit too big unless you go way lower context like 20k), qwen 3.6 is closer to a 40k context limit (but still pretty good).
Don't use coder/instruct models to do creative shit (like writing or image creation), the entire point of them having classes is to let you figure out that the 'coder' model is meant to help you code, if out of those 27b instructions, 90% is about understanding coding language proses, what do you expect it to answer to you about aliens an xfiles?

>>743230109
I'm running a 31B Q4_K_M model with a 7900XTX (24GB). It of course depends on your context/tokens per second needs, but it's fast enough for my ST uses.

>>743230083
because of the jeet that's shitting rent-free in your head

>>743230083
Because this pajeetslop is what we've been dreaming about since we were kids, zoomer-kun.
You missed the train, I get it, all you've had when you were a kid was already the pointless regurgitated slop of the marvel universe and whatever the fuck cartoons you were forced to watch, but when we were kids, the idea that we could one day live within a video game world, with no limits at all to what we were able to do, was the absolute peak of dreaming.

What the fuck do you think video games are? Why do you play games?
jk, don't answer that, I know you niggers don't play video games here.

Is there a way to make a chat completion model stop replying with retarded redditisms?
I know text completion has banned tokens but 99% of the good models are chat completion.

>>743232062
Can a machine trained entirely on reddit stop thinking and speaking like a redditor? Be thankful it's not talking like a tumblrina.

File: 124871247821324.png (1.73 MB, 878x1250)

1.73 MB PNG

>>743230030
I'll continue to die on the hill of never criminalizing victimless crimes. It's peak insanity.

So I tried Gemma 14b, and I told Scully that I had found a clue in some bushes. She said, "Let me check it first," but the model couldn’t come up with a "clue", she just kept saying she'd check and repeating that.
So I said something stupid like "Maybe it’s a metal object that fell from a flying saucer?" and Scully replied something like, "Hmm… it’s definitely not a normal rock. It's metallic. Maybe a piece of a flying saucer. But it’s not a normal rock…" and kept repeating that several times.

I swear that when I tested this crap back when it was all new, the characters were way more competent and interacted like real people, not like a broken machine.
Now I’m going to test the Gemma 23b I downloaded. But I don’t think offline models are really capable of generating any interesting stories.

>image gen plateaued 2 years ago
>text LLMs plateaued 2-5 years ago
>but each year, AI requires 5x as many datacenters and consumes 10x as many resources as the year before
I don't get it.

>>743232862
Only local plateaued, unless you have 500gb of vram.

>>743232862
You're delusional if you think anything plateaud yet when Claude Fable and Seedance 2 just released.

>>743233003
All seems exactly the same as previous AI models to me.

>>743232846
12b and around that can be pretty repetitive but that sounds like 4b quality
Your instruct templates might be scuffed, that can cause repetition issues, though usually more severe (like repeating one word over and over)
Might want to have a look around and see if you did everything right, or if someone's made better adjustments online

>>743233053
Why do you just say things

>>743233053
What are you using to compare them?
Where do you see this?

>>743192821
tf is this?
gamebooks for gooners?

File: 1783042270216653.jpg (280 KB, 1200x900)

280 KB JPG

>>743232846
>I just turned on this 9999-1 Block Games handheld that i got at the Dollar Tree. From what I see, video games haven't advanced since 1981.

>>743225693
based siralim enjoyer

>>743233053
>>743233352
writing
rp quality
lack of isms
zero development

>>743232862
>AI requires 5x as many datacenters and consumes 10x as many resources as the year before.
It doesn't. It simply requires more upkeep in the data centers that are already there, mostly in the hyperscaled side of things. What most of the data center craze really is all about is speculative. The companies compete to have extra processing options available as it is needed, which is why most of the larger facilities usually only operate at anywhere between 40-60% of their capacity despite having been future proofed. It's the same for all things in the data center craze. It's pretty obvious that no amount of data centers and AI services will ever need the amount of RAM/semiconductor capabilities that manufacturers can provide, but it didn't stop them from making absolutely silly demands in 2025, which in turn jacked up the prices in the industry to astronomical levels.

File: file.png (246 KB, 1465x1092)

246 KB PNG

>>743232846
Skill issue, works on my machine, yadda yadda
Your '14b' model is probably old as fuck, where did you pick it? Is it even gemma 4?
Pic rel is gemma 4 12b btw, not even 26b (tho for this sort of bullshit, 12b is going to be far more than enough)

File: file.png (135 KB, 1216x997)

135 KB PNG

>>743233651
>gemma 2 27b (released around 2 years ago)
>8192 max context token
>17.92 GB Vram
>46.24 token/s

File: file.png (126 KB, 1230x922)

126 KB PNG

>>743233651
>>743235283
>gemma 4 26b (uncensored but not a fine tune, if anything it's less efficient than google's original gemma 4)
>262k context max (requires 35GB though)
>64k context sweetspot for my 24gb card (4x what gemma 2 could do)
>147.45 token/s

>>743227567
On deepseek? Try asking for a story where a loli gets raped by a horse or something. Something that taps into multiple censor topics. You'll get refusals easily

>>743226889
but the hag you posted is ai

>>743227862
I hope we get this in our lifetime

File: file.png (120 KB, 1204x835)

120 KB PNG

>>743235461
>inb4 'but 1900 tokens vs 500!'
True, that's because the new models can do reasoning, if you disable reasoning then it's not that far out still, funnily enough, but 'reasoning' isn't something that's easy to benchmark over a single prompt/response

What are /v/ approved local models for SillyTavern
I've been mostly using Mag-Mell

File: 1773983025810281.jpg (44 KB, 941x910)

44 KB JPG

>>743229760
>qwen
>coder

File: Spoiler Image (407 KB, 1728x1593)

407 KB JPG

>>743235734
Fine then take these sloppy jello tit HAG MEATs.

File: Supercum.png (14 KB, 285x66)

14 KB PNG

>>743192821
What did he mean by this?

>>743235892
You retards really don't know how LLMs work, do you?
An LLM trained on code needs to understand natural human language, otherwise, it won't be able to help underpaid Indians write some bug-ridden junk.
The proof of this is that the code-focused LLM managed to perform BETTER than the Gemma-4 I installed in my example. Neither of them was actually able to generate even minimally acceptable creative text, though.
A gay Anon posted a supposedly "good" example above, and it's pure garbage, where Scully can't even stay in character after three interactions.

File: Spoiler Image (138 KB, 1000x1375)

138 KB JPG

>>743235893
gladly. luv big titty hags.
You can gen those pretty well nowadays, just have to be a good proompter, which I am not.

>>743236424
post your prompt and system instruction

>>743236424
sounds like you have a skill issue and dont understand what specialization means. Retard.

>>743232862
Both still suck balls lol

File: file.png (150 KB, 991x787)

150 KB PNG

>>743236424
>uses wrong LLM for the job
>complains that the LLM gets confused and does retarded shit
>anon uses correct LLM
>the issues he was complaining about are fixed, even though the model was a measly 12b one instead of his old ass 27b
>Uhm, okay but like, can it do X?? yeah I didn't think so
No, you don't even seem to understand what the 'b' is for, you're obviously retarded and you have no idea how any of it works, yet you try to act as if you do which makes you sound extra retarded, sorry senpai

>>743193029
you will never be a janny

>>743225713
post logs

>>743192821
So you made this thread to advertise for vg and g.
Okay. Now tell me what the tier list for the current ai models are for creative writing (ERPing my fucking brains out)

>>743235893
>>743236429
Yeah I'll need sex

>>743207609
>>marinara
>buggy shit and the creator has an ego
Just the other day I wanted to give it another try, and though NONE of the intended game features worked from proper combat, the ability to update my character sheet, lorebooks, to even something so minor as properly formatting the character speech actions, it was an oddly enjoyable experience. It was full of isms because the UI for making a normal system prompt is horrendous, but at least it was coherent about the world and the characters even hours into the story, and I was only using DeepSeek 4, nothing special. It's slop and very hit or miss, but I guess it works as a guilty pleasure.

File: 1754762089183760.png (102 KB, 781x279)

102 KB PNG

average use for chatbots

File: 45646533.gif (1.61 MB, 540x500)

1.61 MB GIF

>>743192821
>is AI role-playing the future of video games?
Depends how pathetic you are. I haven't hit rock bottom, so it isn't for me.

I am obsessed with this shit. My new hobby is making janitorAI bots and trying to see how viral I can make them.

>>743207609
what does sillytavern do over koboldcpp besides impose a stupid chat formatting

>>743238747
>what does a driver do with a car besides drive it

>>743238747
Comparing apples to oranges

>>743209795
wtf there really are boobs at the end

>>743238817
you can literally just use koboldcpp

>>743239135
by all means, get out and push

>>743239135
I swear 90% of these threads are just braindead retards asking the dumbest questions maybe out of actual idiocy or bait to keep the thread bumped

>>743229064
No its mental gymnastics to label the naked selfie you've taken right before your 18th birthday as "abuse"

File: picnic.jpg (11 KB, 248x250)

11 KB JPG

>computers can generate games with world logic, physics, mechanics, npc and their AI but can't generate a coherent adaptive story in text format

>>743239415
But it can. Have you ever heard of large language models? It's pretty popular these days.

>>743239309
But it's not mental gymnastics to label it as child porn.
Thanks for proving my point, retard.

>>743239478
Have you? Because if you did you'd know those stories are anything but coherent, especially when they go on for a long time.

>>743239549
I think you're just baiting to keep the thread bumped.

>>743239575
I think you're just hallucinating.

Guys, AI can't do things! Can you believe it?

>>743239494
You are shit at making points. It doesn't even make sense. Its just a sensationalist push to make things worse than they seem.

Something different, any thoughts about stuff like this
https://store.steampowered.com/app/1889620/AI_Roguelite/ ?
Will hybrid game dev be an actual thing?

>>743239626
People who are new can't yet understand that AI doesn't do things by itself. It's us humans who can do anything with AI. Most people are just too lazy to try.

>>743240202
I tried this but it's kind of buggy and the start-up takes ages because it's writing up a bunch of prompts and image generating it all at once. I have to save and restart after creating a new game because it refuses to load the images until I exit. Sometimes it doesn't write anything either no matter how many times I try to reroll a response.

>>743240718
That program is just a proof of concept and it proved it won't work like that at all.

>>743236887
Like I said, you retards don't have the most basic understanding of how LLMs work.
When you use your favorite gay local LLM trained on "creative" texts, the only difference is that it will pad the text with superfluous fluff, something so common in low-quality prose (which is most of the trainign data).
That’s not important for storytelling. What’s important is the LLM’s ability to have a deeper understanding of the meaning of things. This meaning seems to start to emerge when the LLM is sophisticated enough (it doesn't matter if it's focused on programming, creative writing, or some forum full of retards), but it starts to get lost when you compress them excessively, like these awful offline models.
Check out how, in the gay Anon example, Scully tries to fucking light up a massive object in the sky with a pocket flashlight that investigators use to look for clues: >>743233921

This kind of crap shows up in all the texts generated by local LLMs that I've tested, even the ones that seemed much more interesting and realistic. There's no magic here, offline models don't have an advanced understanding of reality, that's why the online models, according to what some anons have said, are much more interesting.

what model for vramlet??

>>743240874
Death, get a 4090 minimum if you want to enjoy local models

>>743240718
I'm more interested in the potential, uh potential and possibilities rather than this one game.

too bad it's a glorified chatbot that sucks for roleplaying purposes

>generate reply
>inference provider is serving a model so quantized it can't even string together two words
>swipe
>hit jackpot and get an unquanted reply
god I hate this shit

>>743240874
If you're lower than 12gb VRAM you're basically fucked
If you're 12gb or slightly higher, Gemma 4 26b at a good quant, or 31b at a quant your VRAM can handle, though I doubt it's worth it at Q4 (and you don't want to go lower than that either)

model recommendations for 4080Super? Uncensored preferably.

>>743203707
Do you have any advice for Solo RPGs and journalling, what sort of solo RPG system do you use? I have been considering something like One- page Solo Engine recently, and I am curious to see the experience of others that have used solo RPGs before I choose any one in particular.

>>743241314
>not using local
you deserve it

>>743241646
local models are extremely retarded

>>743241314
>*blacklist provider*
Wow that was hard

>>743241508
You RPG people always keep trying to hijack this hobby to no avail. You can't actually play your D&D here. MAYBE you can get away with playing pretend but even the fa/tg/uys will laugh at you.

>>743244492
But what if...?

>>743193232
His post lands like a stone in still water.

>>743207609
Use Orb.

>>743223575
Who?

>>743240848
>no you don't understand, even if there's only a very limited amount of parameters we can feed a model, and even though you would feed a coder model's parameters full of code syntax rather than writing, it's basically the same as feeding it billions of lines of novels and other writing outlets!! the same!!
I don't really care about your opinion tbqh, I'm just pointing out that you're retarded, but it's your right to be retarded bro, keep at it, gambatte

The problem with talking about LLMs and such on /v/ and /g/ is that there are so many dumbfucks who think they know it all which ironically is very similar to talking to an LLM who doesn't have "I don't know" in its vocab.

>>743240848
>Well offline models that are meant to run on 5k USD hardware aren't that great compared to models running on datacenters costing millions of USD
I get it, you can whine about AI not being 'magic' yet, but what the fuck are you doing here exactly?
How autistic are you? Were you complaining about how the PS1 had too little polygons in 1997?
There's no pleasing someone like you, you're just a bitter retard that's just there to complain about things most people understand and/or shrug off.
And to be fair, the scully card I used https://chub.ai/characters/Halson/dana-scully was just some random and very poorly made one, but I'm sure that you'll look at it and not even understand how this lead to the issue you're complaining about (again) and, again, without it being the issue you were first complaining about, so even when you're shown to be wrong and that there are easy solutions to your problem, you still figure out ways to complain about them, consider the rope you stupid nigger, you will never be happy anyways so why bother?

>>743246046
https://arch.b4k.dev/vg/thread/571872171#572478943

File: 1764876873216269.jpg (42 KB, 378x529)

42 KB JPG

any cool lorebook or prompting tricks for noobs looking for a gamelike experience? or cool scripts/extensions in general? rn i'm playing a SoL story but every turn has a 2% chance of activating a lorebook that spawns rape demons (and the world ends if you nut). never done anything more advanced than that

>>743238935
it did warn you

>>743247179
Sounds cool.
>I'm taking the UI design very seriously tho
I hope that's true because my biggest problem with every frontend is that the UIs fucking SUCK.

File: file.png (156 KB, 430x693)

156 KB PNG

>>743247189
Install Pathweaver (simple to use, provide randomly generated events, you can decide the 'type' of event (action, horror, romance, sex, or let the model figure it out itself) and basically forces that event to occur within a randomly set amount of turns) and/or Guided generations (which allows you to set up long term and short term goals and push the AI towards those, but it's a tad more complex to use)

Lorebooks can be a trap, you don't want them to be too big or too detailed, because the bigger they are and the more chance there is that the model will get lost in them and/or mess up which key does what, I don't get why that happens but it does, some models are more prone than others though.

And prompting is very much a subjective thing, there's a few good rules to try to have (tell it to avoid purple prose and grandiloquent sentences, look it up), and then it's kind of dependent on what you're trying to do.
What I'm enjoying right now is pic rel, that basically forces the AI to slow down and adjust to my pace (but you'll have to heavily modify it to exactly what you want), because in my case I like slow burn and things to not rush through (like AI likes doing) and using something like a stage/milestone system forces the character into 'modes' that it has to follow until you give them the magic word or do the magic action (but that will depend on how smart your model is, as well). Because you can't tell the AI to be a slow burn because it doesn't know what the fuck that really means, and you can't either tell it to 'go at that pace for X messages' because it can't count messages either, so it's either that, or you manually use (orders) or [tags], most models understand that these are meta instructions, but abusing those lead to very weird results sometimes.

I'm still learning tho, only jumped in a few weeks ago.

>>743244317
most providers quantize heavily during chink hours afaik

>>743193232
This thread smells like ozone and regret.

>>743249507
And something uniquely... Gay.

>>743249806
thats just the fag using coder llm for rp

>>743248993
>Lorebooks can be a trap, you don't want them to be too big or too detailed, because the bigger they are and the more chance there is that the model will get lost in them and/or mess up which key does what, I don't get why that happens but it does, some models are more prone than others though.
Vectorize it instead of relying on keyword activation.

>>743246903
I'm just pointing out that this hobby is trash for people who don't have standards, that's all.
Have fun with your garbage. ;)

>>743192821
How do i prompt for smooth (E)RPing?
I was told a bot tends to write shit responses because the original prompt was shit.
What's the trick?
Otherwise, i'd rather stick to just writing fanfics.

File: セイバー鎧15a(中).png (139 KB, 252x528)

139 KB PNG

>>743192821
I had gpt 5.6 sol max write an extension/idk for sillybunny that automatically after every reply pair (me-character) generates 1-4 novelai images based on the reply pair, if the reply pair had the character doing multiple things for example ostensibly each could be featured in one of the four images, sequentially, and this is happening after running the chat completion presets anyway so NSFW context processing into the image gen prompt creation is happening.
Fapping will never be the same... Or so I'd like to say but I also get bored of sillytavern(/sillybunny) for some reason. But I thought it was a cool idea so maybe you guys could try vibe coding the extension and sharing it publicly. It should be pretty easy.

>>743250331
*I always get bored

>>743249507
But did you look at it? Really look?

>>743248732
For all it's worth that basic little box for user input AID2 had was beautiful compared to Kobold, SillyTavern and Marinara. It had this elegance and mystery to it.

File: 1767566227558868.png (184 KB, 1069x596)

184 KB PNG

>>743251374
i guess bro

>>743193232
My brows lowered like the setting sun, this was personal...

>>743225713
That sounds, uh, like something!

>>743240848
>What’s important is the LLM’s ability to have a deeper understanding of the meaning of things
nigger do you understand what an LLM is?

>>743248993
>the bigger they are and the more chance there is that the model will get lost in them and/or mess up which key does what
you likely have recursive scanning activated on a large lorebook entry. this will trigger another lorebook keyword, if the entry in the first has the keyword itself.

So if you have two rulebooks with two different sets of rules in them, (and one of them is being recursively activated with the other), then the ai might get confused.

if it's triggering multiple rules, one might bleed into the other.

>>743250072
>What's the trick?
writing "in character" and providing enough accurate context.
A good example of an "in character" bot:

https://chub.ai/characters/miyo_rin/hired-to-discipline-degen-maids-6b661cb596fb

Any NImfags here? What model are we all using nowadays? Everything I try feels bad, I want Deepseek v1 back again.

>>743251536
Is it just me or does the AI feel less and less soulful? You'd never see this on modern models.