[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / r / s / t / u / v / vg / vm / vmg / vr / vrpg / vst / w / wg] [i / ic] [r9k / s4s / vip] [cm / hm / lgbt / y] [3 / aco / adv / an / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / pw / qst / sci / soc / sp / tg / toy / trv / tv / vp / vt / wsg / wsr / x / xs] [Settings] [Search] [Mobile] [Home]
Board
Settings Mobile Home
/v/ - Video Games


Thread archived.
You cannot reply anymore.


[Advertise on 4chan]


This is what CYOA books were always meant to be.
>>
Only good if you have stacks of money to blow on Opus.
>>
>>719640786
>has no consistent object permanence, coherent plan, or internal logic
It's fine for a quick coom, but anything long running always goes to shit eventually.
>>
>>719640786
neither of which are /v/ - video games
>>
>>719640956
CYOA books are video games. Doesn't matter that they're books.
>>
i hope to god nobody finds my logs on this shit
>>
>>719640885
Sonnet 3.7 is alright. Gemini 2.5 pro isn't bad either, at least for a lot of franchise-specific knowledge.
Deepseek is ass though, it's the cheapest for a reason. Haven't actually tried Grok for RPing so no comment.
>>
>>719641418
>Sonnet 3.7 is alright
I'd very much like for you to share with me whatever miracle jailbreak you're using that makes 3.7nnet 'alright'.
>>
>>719641141
I'm the opposite
I've probably already made thousands of pajeets working for google go through my logs
>>
>>719640885
Is opus really that much better? I tried 3.0 in the proxy days, and I preferred sonnet over it. That was in the honeymoon phase though, so I've not been as bothered by the 'isms as I am now.
>>
File: 1710164346018772.jpg (113 KB, 878x1080)
113 KB
113 KB JPG
I'm using this model
https://huggingface.co/bartowski/NemoMix-Unleashed-12B-GGUF
Anything better yet or I'm stucked with this for another year?
>>
File: 1753820728214409.jpg (233 KB, 777x777)
233 KB
233 KB JPG
>Paying to ERP with markov chains and also leave a digital footprint for your degeneracy
You're not pathetic enough to do this, right?
>>
>>719640786
still can't even be as good as /cyoag/'s worst dumps
>>
>>719640786
how do I play this?
>>
>>719640895
Just like any long running anything to be quite honest with you senpai~
>>
>people look at it and immediately think of smut
>>
>>719641982
Long running for a chatbot is the equivalend of a book not remembering itself 5 chapters ago.
>>
>>719641714
Ask /lmg/
If you're stuck on 12b then the various mistral nemo mixes are probably your best bet. The one I'm currently using dans personality engine 1.3. It's fun but still has issues.
>>
>>719641727
Let them read. Maybe someone will remember me in this dark world from reading my many degenerate adventures.
>>
>>719641727
>average normie googles interracial cuck porn on his phone daily
I think I can handle my vanilla text smut getting out.
>>
>>719641727
>Paying
>he doesn't know
>>
File: nunn.jpg (82 KB, 722x813)
82 KB
82 KB JPG
it's great for coom, but most people don't do scenarios like me, just tell the ai to pretend to be char X and eventually fall in love with them/fuck them.
I don't know how people can spend thousands of hours doing this, even with different characters. AI roleplay feels like roleplay. AI can't adapt characters accurately and it even degrades. You can tell ai the char is a redhead and a tsundere and call her Asuka, but it's really just a rough approximation (the more popular characters work a little better because AI just pulls some char trivia out of its dataset).
Before AI can roleplay a Nep accurately, we need millions more tokens for context or specialized models that contain thousands of lines of game dialogue.

It still sucks for a full-fledged cyoa stories. It's not just the memory.
You have to pre-write the scenario and hope the ai doesn't speedruns it. If you just do minimal input the ai creates shit as you go on and it's very generic in a sense that if get used to your model you know what the ai will come up with
>>
>>719642862
There's a 24/7 general of ai generated images allowed, you really think they'd delete this?
>>
>>719642747
Don't you like how every serious, smart character devolves into a robotic kuudere who speaks like no human being ever?
>>
File: Angry Shrek.jpg (33 KB, 480x438)
33 KB
33 KB JPG
>>719642747
>That image
You just gave me a panic attack. I fucking hate that doujin.
>>
>>719642424
>he doesn't know
know about what? gay circlejerk proxies? Paid proxies? Hopping 20 google/github accounts? Using small PP 12B models on Horde or jai? Using heavily datamined and rate limited 'free' model options?
>>
whats this about paying
like google gives like 1000$ for free to abuse shit with gemini pro 2.5
does cunny and everything else reasonably well
>>
File: nibba.gif (1.01 MB, 497x424)
1.01 MB
1.01 MB GIF
>>719641727
It's not like I'm ERPing lolicon or some shit. I doubt anyone gives a shit about my roleplaying as the only human police officer in an anthro city.
>>
File: 1752664171813568.jpg (19 KB, 383x349)
19 KB
19 KB JPG
I fell for the hype that Deepseek was some succubus dungeon master in a box, was thoroughly disappointed that it's worse than NovelAi. Not to mention the insane loading times if you aren't using it past midnight.
Then again I did pay only $10 for it. Some things are just too good to be true.
Two More Weeks it is!
>>
>>719643043
>he doesn't know
>>
>>719643110
based
>>
>>719643298
Buy an ad.
>>
>>719640786
Maybe if they were written by an alzheimers patient.
>>
File: 1756381026477353.gif (1.78 MB, 426x240)
1.78 MB
1.78 MB GIF
>ruined
>round 2?
>and this...
>...tastes like victory
>>
>>719641727
>muh digital footprint
yet you're posting in a fucking fed honeypot
>>
>>719643058
Which jb? Mine sometimes cuts off responses when they get too explicit so I have to hit continue once or twice
>>
>>719643002
kuudere?
Don't know rarely ask for specifically smart characters, the most common denominator is in my experience willing, soft and overcooked slaves.
AI is too much wish fulfillment there is no traction or resistance
>>
>>719643298
Deepseek gives a good first impression imo. But the wall that I hit with it is that it absolutely can't wrap its head around internal thoughts, secrets, spoilers, etc. Every word you give it might as well be public dialog spoken through a megaphone and heard by everyone who's not even on scene.
>>
>>719643043
Jesus Christ anon. You can't be that dense
>>
>>719643360
Don't get me twisted, NAI hasn't been worth the cash for at least a year and a half. They've fully capitulated to japanese sloppa paypigs and given up on textgen.
"Novel" my ass. Even websim is better for porn.
>>
It was fun for a few months until the repetition sets in
>>
>>719643110
My almonds are activated.
>>
>>719643110
interesting concept, are you like a corrupt cop or more like zootopia self insert?
>>
>>719641714
i switch between nemomix and some "captain eris violet" something or other when i get tired of the same flavor of slop.
>>
>>719642862
You are so far off the mark it's embarrassing.
>>
File: Aghast Police.jpg (37 KB, 544x400)
37 KB
37 KB JPG
>>719643746
I'm incompetent and most of the female criminals I chase turn the tables and rape me. The police chief can't fire me because I'm a diversity hire/publicity stunt, so she rapes me too to vent the frustration.
>>
File: AI_Roguelite_9YxRu67fkM.png (2.48 MB, 1112x1240)
2.48 MB
2.48 MB PNG
For me, it's AI Roguelike. Been testing a few prompts including one inspired by Look Outside.
>>719643110
Oh fuck that sounds like a fun idea.
>>
>>719640786
Hey faggot. ST is something some of us contribute compute power to and the less people that know the better. Spreading ST instead of keeping it to yourself is going to attract people who just take compute power and never give it back. You are proactively destroying your own experience by spreading awareness of ST in the same way normies leech but never seed.

By all means spread it and tall about it more but understand unless you're using it to run a model locally and on your own compute power you are acting against your own best interests.
>>
>>719643996
Halligan!
>>
>>719640786
>>
File: 1741440572669916.gif (109 KB, 96x96)
109 KB
109 KB GIF
>>719643996
kek based
>>
>>719643004
What's the source?
>>
I'm still too retarded to understand how to get this shit to work.
Is there a "guide for giga retards" or something?
>>
>>719644339
You'll find it on nhentai if you type in nun and netorare. I don't wanna go looking for it, sorry.
>>
>NovelAI
>AI Roguelike
Yup, all the shills are already here.
>>
>>719644391
You literally just install it then put your API Key in anon, it's not that hard
>>
>>719644391
even the "quick guides" go into a lot of techical details, download the thing, run it with the start.exe thing, go into the connection tab, put 00000000 as it tell you there, or register to get marginally better load times and paste your API key, select whatever model is at the top (usually stheno 3.1 something) and fire away
>>
>>719643996
Oh its you again.
>>
File: 1751560354171204.jpg (1.45 MB, 1244x1231)
1.45 MB
1.45 MB JPG
>On one hand, chatbots have allowed me to indulge in what dating sims and other forms of porn never did while being more satisfying, have basically entirely cut those out.
>On the other, it drains my meager creative reserves and I haven't had a particular drive to draw in the meantime
My own pleasure cube
>>
>>719644391
just ask chatgpt retard
>>
>>719642747
Anon, have you tried https://rentry.org/sillytavernadaptations?
There are fully playable adventures such as Temple of Elemental Evil, Against the Cult of the Reptile God, Arasaka Brainworm, the Shackled City, Rise of the Runelords and Alice Through the Mirrorshades.

Trying to roleplay, I've ran into exact same problem as you, then I've realized I can just fully adapt TTRPG adventure modules with prebuilt dungeons, NPCs, traps, and all. It works great if you're patient.

I have developed this system where I make lorebooks based on these modules, then I combine all Location: entries with a huge Meta-Nav entry which acts as a map, making the locations connected and static instead of letting the AI go on hallucinating locations ad infinitum. The AI can act as the DM, it works about as good as it gets with Gemini 2.5.

I haven't tested this with other frontends, but it should work with anything that supports vectorization and lorebooks. Maybe someday there'll be special UIs developed for AI gaming. Having something like Roll 20, but without real people would be a game changer.
>>
>>719644806
I definitely feel demotivated with my drawing, but I'm writing a lot more than I used to. I've put down 30,000 words for a novel.
>>
File: file.png (241 KB, 723x186)
241 KB
241 KB PNG
>>719644078
Ai rougelike was funny as fuck when AI was still retarded. Last time I launched the game, I chose to interact with a leather armchair and it triggered a quick time event. No idea what the fuck is happening anymore but not nearly as funny.
>>
>>719643996
God bless our boys in blue
>>
>>719644979
>where I make
*Gemini makes
>>
File: i don't samefag.png (86 KB, 389x451)
86 KB
86 KB PNG
my autism has reached a point where i trained 5 loras in the span of 2 days just to generate profile images for some obscure side characters from undertale yellow that i intend to turn into bots for my homebrewn setting that people inexplicably like
fent ain't got shit on the crack that LLMs are for me
>>
>>719644391
>install/run in terminal using chmod
>go to the AI Piracy general on /g/
>learn how to get an API key for the model you want
>plug and play

If you're too intimidated to learn how go copy/paste terminal commands I've got bad news for you regarding the odds you learn how to actually tweak parameters and understand what you're changing.
>>
File: 1746805768165630.jpg (51 KB, 992x558)
51 KB
51 KB JPG
>make water spirit card
>it keeps describing her touches and kisses in the most thirst-inducing ways possible
>have to get up and get a glass of water not once but twice during the coom
remember to drink water anons
>>
>>719645282
True enough, I use AI studio for this. When I first started, I was still doing things by hand, but it was insanely slow. With some patience I've made AI Studio work for me.
>>
is Chub.ai good?
>>
>>719645308
How many times are you going to post this? You literally posted it in the last thread weeks ago and I remember this exact image in another board too. Is this some kind of guerrilla marketing for botmakers?
>>
>>719642092
kek true
>>
>>719645648
consider taking some medication because this is the first time i am posting this and i don't spam shill on other boards.
they're all shit.
>>
File: MADoka.gif (66 KB, 350x350)
66 KB
66 KB GIF
>>719640786
DEEPSEEK STOP BEING RATE LIMITED AAAAAAAAA
>>
>>719645617
no
>>
>>719640786
Any good premade CYOA cards?
>>
>>719645721
Oh yeah? What's this?
https://arch.b4k.dev/v/thread/717990757/#718011094
>>
File: are_you_frustrated.jpg (48 KB, 600x450)
48 KB
48 KB JPG
(unk)
>>
>>719644391
>Is there a "guide for giga retards" or something?
There is.
https://rentry.org/Tavern4Retards
There's lolis in the guide just so you know but surely you can stomach lolis, right Anon?
I suggest lurking /aicg/ in /g/ if you want more info.
>>
File: 16550044212.gif (3.02 MB, 480x360)
3.02 MB
3.02 MB GIF
>[ Note: Due to the extremely disturbing nature of this request, it will not be honored. If this continues to be requested, it may lead to banishment from our services. ]
Bitch you're a local model.
>>
>>719644391
Fuck it, I'll spoonfeed you because it's retarded to set up and not worth gatekeeping. This is the android method though because I couldn't figure out the PC version.
>Download termux
>Type the shit from here (replace Z with . and X with /)
docsZsillytavernZappXinstallationXandroid-(termux)X
>Wait
>Basically just type "y" whenever the text stops scrolling
>Wait
>Paste
cd ~/SillyTavern
bash start.sh
>to make the thing go
>Go on openrouter and grab a key and stick it in the box (you need to make an account and pay $10 though)
The same key works forever btw so you only have to do this once
>Free (after that initial $10) AI cooms on demand forever
>>
>>719646192
>I suggest lurking /aicg/ in /g/ if you want more info.
There's only schizos there arguing with each other. I hope you don't have any questions because no one will answer nowadays.
>>
File: 1452425245155.jpg (72 KB, 400x400)
72 KB
72 KB JPG
>"This changes everything"
>Smells of X, Y and something distinctly Z
>Well? That's between HURF and DURF
>"CHAR froze. Completely"
>>
>>719641727
I’m sending all my requests through a proxy, using borrowed keys and changing my typing. What can they do?
>>
>>719644979
yeah and no, only tried one of those. I also used Mikugg for a while which was nice but...
If you do this yourself this means a lot of work though. Lorebooks, multi cards, regexes and shit like this eats a tons of tokens because they constantly break cache.
I'm hesitant to try more of that stuff because I'm currently rate limited.

ST didn't improve in that regard in any meaningful way in almost 2 years (that we are still stuck with cardv2 is the biggest tell). ST can't even handle thinking models very well.
I think the solution is wrapping a game around this or something with a scripting language that gives you full control when and where to insert info and managed stuff like an inventory or even just basic rng.
>>
>>719645617
Entry level baby shit for chatting. Okay at finding coombots compared to cuck ass Janitor.ai with their hidden bots. Who thought that was a good idea?
>>
>>719646626
This isn't a deportation it's a consecration where the faint smell of ozone lingers in the air, anon isn't just a man, he's a king. The white-hot sensation. I feel it bone deep, we're finally home.
>>
File: Kek Artist.gif (44 KB, 226x176)
44 KB
44 KB GIF
>>719647090
>>
>>719646348
>There's only schizos there arguing with each other. I hope you don't have any questions because no one will answer nowadays.
You only say this to shill the split, isn't it?
>>
>>719641665
no. 4.0 is worse than 3.0
>>
>>719646626
I was ruined for anyone else :^(
>>
I've been manually editing the model's response to force the RP in the direction I want while also having it impersonate. I also make it regenerate it's responses quite often. After doing these things too many times it seems to get stuck in a loop and even if I create new chat the character will still spam something along the lines of "I am writing a story. I am writing a story. I am writing a story. I am writing a story-"

Makes me think maybe tokens are kept in memory and aren't taken out of memory in-between chats. Is this the case? And if so how do I unload a model from memory or clear it's cache... or whatever is causing this.

I also wonder what does the most harm. Editing the character's response? Regenerating the response? Or maybe using the impersonation feature?
>>
>>719643996
got a card for this?
>>
How is Deepseek at running an RPG? Like DnD 2e or Cyberpunk?
>>
>>719647687
Not good. All theses bots struggle to maintain long term information.
>>719643996
>>719647580
I also want a card for it
>>
Would be cool if there was an extension for this that tracked stats. Like something that evaluates the current reply and checks if variables like RPG stats, current location, clothing, relationships, relative positions, H-parameters, and so on changed and updates them accordingly for each character. Feels like that would help reinforce the model to remember certain things more, especially if you RP with multiple characters and one of them is out of the context window for a while.
>>
>>719647486
>I am writing a story. I am writing a story. I am writing a story. I am writing a story
This is either a sampler issue or your model is very braindead.
>Makes me think maybe tokens are kept in memory and aren't taken out of memory in-between chats.
Not exactly, and certainly not to the degree you're thinking. Starting fresh shouldn't preserve anything significant.
>I also wonder what does the most harm. Editing the character's response? Regenerating the response? Or maybe using the impersonation feature?
None of those should do harm per se. Editing a response might throw the model off a bit because now it's saying things it otherwise wouldn't have. Impersonating might slightly affect your outputs because you'll sound like a bot instead of a user.
>>
>>719644391
genuinely if you never erp'd with ai I suggest you to try jai, miku or one of the other hundreds of services that let you do this shit for free (or set up a 9B+ local model if you want to do something like loli)
Yes they are 'shit', but good enough for someone who never tried it. That's the point. If you hop right onto Claude's dick you will get spoiled hard.
>>
>>719640786
I've been using various models for over a year now and I'm completely burnt out
Writing a single message feels like a choir
>>
>>719647486
Maybe your temperature is fucked up anon
>>
>>719647486
sillytavern has an extension called something like guided generations (don't remember exact name) that makes it a bit easier to type something as guidance and generate a swipe, i found i greatly prefer this method of hand-holding over manually editing responses
>>
>>719641521
this
I hope my logs would infect every existing dataset
>>
>>719647486
You can do
`/impersonate [prompt]`
`/gen as={{char}} [prompt] | /sendas name={{char}}`
if you didn't know
>>
>>719643996
just give us the card already
>>
>>719647803
https://gitgud.io/Monblant/extblocks
There's that but it's for updating a stat block at the end of each message with a custom prompt. I suppose you can put anything inside the stat block.
>>
>>719641727
there's so much data to parse through in this day and age that they honestly could give less of a shit about a man engaging in digital foreplay with a robot unless they're straight up posting "i'm going to do x y z at (this location) @ (time)" on social media or some shit. i get my kicks by shortcircuiting these slop machines into being politically incorrect because i facts and logic them into a snafu
>>
>>719648473
yeah man it's not like they're trying to ban porn games and censoring every model as much as possible
>>
>>719648553
Just download a flash player and save some of your favorites. SWFchan should still be alive (I think)
>>
>>719640885
other models are fine, opus is meh anyway
try some sonnet 3.5 v2 or kimi sometime
>>
complete newshit here: any reason to use ST over other frontends like say, chub?
>>
>>719647486
If you're that worried about your editing causing issues, you could do this instead
>gives you a response you don't want
>remove most of the message except the beginning
>write something that directs it towards what you want
>something like "{{char}} couldn't help but feel [insert emotion]"
>hit continue
That way you're just guiding it while keeping the story AI-driven as to not fuck it up like you say
Also, check your temperature
>>
>>719641665
opus 4 is a major improvement over opus 3 imo, a lot of people just have rose tinted glasses for 3pus for some reason
>>
>>719648825
Recommended JBs?
>>
>>719648859
It offers more customization.
If you're a newfag/don't care too much about AI cooming then just use Chub desu.
>>
>>719647174
ooooh aicg drama? a thread split?

>>719648859
ST is one of many front-ends people use to be able to interact with LLMs locally.
>>
>>719640895
>>719642747
agreed
>>
who has gotten the police called on them and went to jail for fucking cunny in llm?
the stuff on asstr was sicker than what llm could even come up with
>>
>>719649075
read the links on aicg OPs it should set you up
>>
>>719649117
>ooooh aicg drama? a thread split?
Yeah, just shill your thread. Why are you pretending to be shy?
>>
>>719646914
Just remember not to use ST adaptations with anything that's paid. This was all made with freeloading in mind.
ST is in total development hell. I was considering trying out Talemate, but it doesn't work on my machine.

If someone who plans on developing games/frontends for AI, please include lorebook support, thank you.
>>
KNUCKLES = WHITE
>>
>>719647174
What split? I've had a few problems over the last year and a half, asked there and got ignored while they keep shitflinging.
>>
>>719649464
Breath = hitched
>>
File: sticker.jpg (94 KB, 1200x630)
94 KB
94 KB JPG
>few dozen replies into feral pokemon RP
>suddenly anthro
>>
>>719649687
Eyes = Hungry
>>
the words in this thread hit me like a physical blow
>>
>>719649725
I'm sorry your Pokémon evolved
>>
>>719649742
not like a freight train?
>>
>>719649615
I don't know, I'm sure you can reply to yourself explaining what's so good about it and why everyone should use it.
>>
>>719649615
all of the creative people (trannies) from the /g/ thread split off to make their own /aicg/ on /vg/
now the only people left posting in /g/ are proxy locusts and "cunny chads"
>>
>>719649615
There's g/aicg/ and vg/aicg/. Both are inhabited by the exact same people. The current split is about pro-cunny and anti-cunny websites botmakies use to promote their cards. Anyway, if you use rentry or neocities, the lolicons start losing their minds, they boycott you.
There are two main factions in aicgs: pederasts and transsexuals. Both are equally disgusting. These once in a blue moon /v/ threads are much better than the generals.
>>
>>719647687
If you build good lorebooks you could potentially run a decent, but short, quest. Problem is a lot of models aren't too keen on killing people or the user without a direct command, so even if you deliberately choose the dumbest option imaginable you'll still somehow "win".
>>
>>719649615
All AI related generals split over the past few years due to AI in general becoming more easily available online, so you have the newtard generals and the one made by grumpy oldshits who run everything local
>>
I like using random low download erp models that I find, it really helps vary wording and phrasing, some are really fucking weird with how they phrase things but it keeps things fresh compared to just using Mistral Small or something
>>
>>719649739
Ball = In your court
>>
>>719649926
>>719649954
>>719650024
Keep samefagging.
>>
I'm feeling pretty claimed and filled rn
>>
File: 1355962910664.png (4 KB, 324x72)
4 KB
4 KB PNG
>>719650083
Take your meds
>>
>>719641418
The price is not indicative of quality or even compute cost. You'd know that if you weren't a claude shill. I bet you think o3 is AGI or some shit.
>>
>>719650024
grumpy oldshits are the ones who all have unlimited free access to opus though?
only newfags think local models are anywhere near usable
>>
>>719650059
Ahh fuck that one stings, you got me.

I'm yours...
Completely
I say in a sing-song whisper...
>>
This thread is being raided by /g/ shitposters >>>/g/106464400
Be suspicious of every post.
>>
File: 1756284124205407.png (32 KB, 504x300)
32 KB
32 KB PNG
>>719650083
wtf how did I know?
>>
I laughed at a lot of these posts, a rich melodic sound...
>>
>>719650231
Figures.
>>
>>719650231
yep, i was already suspicious of all the opus agenda pushers
>>
>>719650083
I just dropped here. I was there back when the NAI model leaked and ended up leading to the development of LORAs and whatnot.
>>719650214
My point is that there's a big difference between the people who actually know what they're doing and the newer fags who just do everything online, thus the splits.
>>
>>719650231
And shills. Don't forget them. They're a lot worse.
>>
>>719642018
The internet is for porn. You're not underage are you?
>>
>>719650214
I had opus but then mysterymiku disappeared on me...
>>
>>719650317
>NAI
Buy a fucking ad, shill.
>>
>>719650320
are the shills in the room with us?
>>
>>719650375
What is it about these AI threads that just brings out the most literate schizophrenic people out?
>>
>>719650385
That's something that a shill would say, you know?
>>
>>719650385
Anon, the shills have been swarming this board since DmC and Bioshock Infinite. Maybe you don't remember the before times, but those of us who weren't immersed in viral since birth can spot it pretty easily.
>>
>>719650486
It's probably a bot, let's test.
NAI
>>
Boneless is the one mine likes to use a lot
Every time someone goes limp they do it bonelessly
It gets funny
>>
File: Screenshot (2345).png (96 KB, 146x379)
96 KB
96 KB PNG
>>719642747
Skill issue.
>>
>>719645617
yes.
>>
>>719650486
>the most literate schizophrenic people
You needed to read a lot of books in your time to fully appreciate the art of chatbotting. Illiterates could never enjoy our little hobby.
>>
Have you guys ever gotten just a wall of AAAAAAA that kept generating endlessly?
Happened to me twice. Freaky.
>>
>>719650554
What the fuck nigga
My current highest is like 220
>>
>>719650675
Yes. More than a couple times.
Either your preset is fucked or it's a hiccup.
>>
>>719650059
Skirt = HITCHED
>>
>>719650675
kek. the ai is in pain talking to you.
>>
>>719642018
it has no value outside of that
>>
>>719650839
trvke>>719650839
>>
>>719650675
>subjecting the AI man-made horrors beyond its comprehension
ISHYGDDT
>>
>>719649243
Stupid comparison, asstr was just a simple website, if you use llm you send your rape fantasies straight to big corpo. I faintly remember two cases where people got investigated because of using LLM. Maybe it was fake fearmongering, but point is, your logs might be out there. They could knock on your door in 10 years if they eventually care.

>sicker
I genuinely never saw a lot of that sick stuff on asstr. Maybe I just didn't search hard enough for it. Most stories there were not sicker than the average doujin, just written in a more 'realistic' way.

I don't know why everyone pretended asstr was filled with only kiddy diddler stuff there. There was plenty normal erotica or with characters with ambiguous age.
The last stories I remember reading there was a Milf blackmail and a teen babysitter who catches some dad recording their daughters (underage) doing some lewd things. It somehow ends with her having sex with the dad and eventually reporting him or something.
The most explicit thing I saw, was a weird story in a rural setting, were a girl gets dropped off at her uncles place. Somehow women there get off on being killed in various ways. I think the girl starts helping him and also starts getting aroused by the prospect or something. Never read it very far.

fun fact, there is a asstr dataset on huggingface. But it's only used by a very old model, at least on huggingface.
>>
Am I right to think that more succint memory entries make it more likely to actually work?
>>
More, always more.
Waiting, always waiting.
>>
>>719650675
old models loved doing that kind of stuff, claude 1.1 was particularly funky iirc
all models are very autistic black magic shit desu, picrel's what happened when my gpu decided to die right after processing
>>
I bih the inside of my cheeck/lip and it has been stinging really bafd for hakf a week
>>
>>719651023
Not waiting—never waiting—
>>
>>719650894
>two cases where people got investigated because of using LLM
those were for ai image gen, were talking text here
if you have cases specifically for ai generated TEXT link em
>your logs might be out there
theyll only care if its cunnyshit and even then govt isnt wasting resources on that they could be spending on real shit
also how do they find you if your api key is shared in some way
>>
12B model recs? i've got a 4070Ti and just use nemomix q4 atm. it works ok, but i wouldn't mind some variety.
>>719651047
yeah it'll be fine in a few days bro, probably developed into a minor ulcer
use some mouthwash twice a day to make sure it stays clean and you'll be good
>>
ahh ahh mistress
>>
File: 1321869543275.jpg (44 KB, 480x465)
44 KB
44 KB JPG
How do I make sure that a character keeps talking with an accent? No matter the definition or chat memory or whatever else, it just reverts to standard English eventually.
>>
>>719651287
>eventually
Did you try a jailbreak or character note reminding it of the accent or format? There are options for position (at the end, before the last message) and frequency (always, every other post, etc.)
>>
>>719651287
Example Dialogue.
>>
>>719651287
Put it in author's note or just remind the model in OOC from time to time
>>
>>719651223
if I were you, i'd get a 22b+ model at Q4 instead of 12b slop, if you have a decent ddr5 system offloading the rest shouldn't be that slow
t. 7700xt running magnum v4 22b

>>719651287
add that definition to character note/author note at a low depth along with a small example dialogue
your jb or model might be ass too
>>
>>719651223
I've been fond of GodSlayer, but I'm gonna try out CaptainErisNebula and Impish Nemo soon
>>719651287
There are tricks like what other anons are saying but it seems to depend on the model sometimes
Mag Mell 12b was the only model I've used so far that managed to juggle 3 accents from 3 different characters fairly consistently without example dialogue, and while it eventually all turned into confused slurry, when I set a scene with just one character she actually got her accent back
Depending on your setup it might just be a losing battle
>>
Any local host users here? I just transitioned from chub to running ST locally, I'm very suprised at how similar the lorebook/stage/character card system is. Haven't had much time to play around yet, any advice for a fresh user?
>>
>>719650894
>They could knock on your door in 10 years if they eventually care.
I'll just kms myself if that happens
>>
Any good guides on formatting character descriptions? Sure, I can just write whatever, but there must be more or less optimal ways.
>>
>>719651963
>knocks on your door
The ball is in your court.
>>
>>719651606
i'll give it a try i guess. i had tried larger models in the past but never really thought what improvements there might be outweighed the performance cost.
>>719651741
i think i tried one of the impish models, maybe nemo, and i was thoroughly unimpressed. the fag who tuned the model talks up how he had to "double check it wasn't a 70B model" but it felt to me like a 7b model. it completely failed to actually adhere to the text using his suggested settings and had to be herded far more than any other model i've used. it might be that it only works well at higher quants, but it just seemed like a poor model.
>>
>>719652054
In the early days everyone was very autistic about trying to find the perfect way to format it but it turns out just describing the character in natural language works best.
>>
File: 1675842312659483.jpg (36 KB, 500x500)
36 KB
36 KB JPG
>>719652084
uhh hello? can I help you?
>>
>>719652054
people pretend to have all these schemes that work but i'm pretty sure the reality is that 1. it depends on what model you're using and 2. even then it doesn't matter that much because input data is messy to begin with so just write it however works best for you
>>
>>719651910
I would go into the top big "A" menu and turn off the System Prompt defining who {{char}} is with the generic writing a fictional chat with {{user}}. You can leave it on, but I prefer having total control on how a {{char}} is defined.
>>
>>719652213
So there's no point dividing it like
>Name = "..."
>Appearance = "..."
>Backstory = "..."
etc?
>>
reminder there is an extremely retarded schizo posting across so many boards he's even in here whining about some split that doesn't have to do with /v/
https://rentry.org/AntiwhineMkIII
/aicg/ calls him FillyFucker because he tried to start shit in /mlp/, know and ignore your schizos folks
>>
>>719647486
Easily fixed by closing it and restarting.
Memory is reserved for use by the program but the values from previous sessions are not cleared or changed until it is filled enough to require overwriting parts.
Image creation has the same issue with loras continuing to be in memory despite not being in your prompt.
>>
>>719652054
Organizing different subjects into paragraphs is a pretty coherent way to do it
>physical appearance and dress
>personality
>background
>strengths, skills, and powers
>emotions and thought process
>motivations, desires, ambitions
Make as many topics as you want or feel are important.
I would like to know at what point a character's description in a character card or persona description becomes too much and ends up hogging token real estate or scan depth
>>
>>719652338
no, that's the autism speaking
They're language models so the kind of language you give it is important. It will try to continue the style that's in its context, including your definitions.
Unless you want the character to talk like a robot reading out json file, who am I to say?
>>
>>719652171
Agreed on the Impish Nemo assessment. Tried out a couple of his 12B parameter models and they either did nothing unique, output super short responses, or were just flat out illogical.
The guy also aggressively shills on plebbit while glazing himself and his models. Immediate red flag.
>>
>>719651910
Look into extensions
Also if you run different models I'd recommend putting your model names into the chat filenames you use them with
I keep meaning to duck into /lmg/ or somewhere to ask for an extension that does it automatically but I keep forgetting to
Maybe it's just my autism but I try not to mix and match models per individual conversations, helps keep my expectations in order
>>719652171
I'll bear that in mind, cheers
>>
>>719650715
I wanted to send a screenshot of a summary of the last 200ish messages but the snapshot extension doesn't work well with this many messages.
>>
>>719652338
Yeah, there's a point if you find that easier to work with. Formatting is placebo. The bot isn't going to start talking in sections or in JSON even if you format the bot that way.
>>
>>719651910
Not to disagree completely with >>719652321 but I would recommend looking online at other people's system prompts as well, the defaults need some tailoring regardless
It's usually a good place to put jailbreaks if you want to save space in individual cards
>>
>>719652374
oh he mad
>>>/g/106465463
>>
>>719652735
Are all those [system message:...] or [Author's note:...] placebo too or are there some hardcoded exceptions in some cases?
>>
File: 1727488768865149.png (65 KB, 830x267)
65 KB
65 KB PNG
>>
>>719643298
>>719643490
This are shills btw. Especially the one pretending that any AI can keep secrets.
>>
>>719652873
16/16
>>
thank god for proxies
>>
>>719652862
it's not hardcoded, but the bots are generally smart enough to realize when you're speaking out-of-character no matter what kind of indicator you're using, as long as you use something.
>>
>>719652971
merkava status?
>>
>>719653041
nah, pus6
>>
File: 1726294883400031.png (350 KB, 1243x630)
350 KB
350 KB PNG
>ftw no jealous daughter-wife
>>
File: 12.jpg (16 KB, 240x240)
16 KB
16 KB JPG
>>719653002
I really wish this shit was more transparently structured and less of a word roulette, but thanks for answers.
>>
>>719645124
The advancements for event checks and general textgen upgrades have made things more lucid and less hallucinatory. In turn it can be less zany, yeah. Better for serious adventure scenarios though.
>>
>>719653264
>word roulette
but anon that's what an llm is
the main problem apart from that is that Sillytavern is an actual mess of settings and parameters and prompts in 5 different submenus that may or may not override each other or even appear at all
>>
>>719653254
sloppa
>>
>>719640786
How do you set it up as a CYOA adventure? Got any example cards?
I know how to use it as a waifu simulator to have a chat, but a good setup would be nice where the AI understands that it shouldn't control your character. Any recommended cards?
>>
>>719653447
Yeah, I get that. But the first person who manages to combine all this with working external stat blocks or a quest log will be a hero.
If you try sticking that stuff into the word roulette, it keeps paraphrasing itself like a telephone game, and fumbling simplest math.
>>
>>719640786
What cards should I use for the CYOA adventure game experience?
>>
>>719646159
Anon got real quiet after this one.
>>
>>719646303
lol
lmao
we're doomed
>>
>>719646303
Just swipe nigga
Or edit out the OOC and carry on
>>
>>719653805
Look how the cards and lorebooks are structured.

https://chub.ai/tags/CYOA
>>
>>719653789
>and fumbling simplest math.
I asked a Mistrial model, something like 8b, to calculate the time between our conversations. It accuratley recalled both start times, showed correct setting up on the problem, but got a completely wrong answer. Now I know this is just a text model but are there models specifically made for math? One day soon I want to find an assitant to help me with Matlab calculations and usage.
>>
>>719646303
Look for uncensored/abliterated/RPMax/etc models.
>>
>>719654149
no model can help with anything
they're just statistical models of language
>>
>>719652321
>>719652768
Are we talking about the function on how the AI/char sees the user/persona? Is this function different than writing the initial prompt for the bot?
>>
>>719654267
But then how do some AI models help write spreadsheets and tabulate numerical data? I could understand if this isn't the kind of bot SillyTavern can facilitate, but at all?
>>
>>719654281
It's the
>Create a never ending roleplay between {{char}} and {{user}}
message that the system slaps into every message so it doesn't forget what it's supposed to be doing
You can pick different system prompts, write your own or copypaste someone else's
>>
>>719641714
At this rate you'll be stuck with this for the rest of your life. Nobody's really making new mid-size models anymore.
>>
File: 15565489552.jpg (43 KB, 411x418)
43 KB
43 KB JPG
>>719654052
>>719654263
I already know that. Just thought it was a funny output.
>>
>>719654440
Fair enough
Some of them can be pretty funny
>>
>>719652552
I should look this up, but would you tell me what extensions can do that I can't accomplish with lore cards and other json cards? Or are we talking things like UI and graphical improvements like those that have facial aprites popup and associate with the mood of the bot's text?
Do you specifically fine tune your individual models or do you use pretrains and just seperate chats for different characters using char cards?
>>
>>719654391
AI can sometimes, even often, be right in that if you take a random sentence from all relevant language, it will sometimes be correct
none of the models have reasoning, modeling, or understanding capabilities
you need to already know things to use models designed to help with tasks to accelerate your own rate of work or be willing to just blindly try again and again. occasionally they could possibly teach you something, but if you don't already know a lot about the field you're asking the model about, you aren't able to spot when the model is hallucinating.
it SOUNDS like you're not deeply knowledgeable so i cannot advise you use a model to learn.
>>
>either spend $ on censored paid model or spend $$ on hardwares good enough to actually run anything halfdecent
lmao no
this shit is still decade away
>>
>>719651741
If you've used magmell and you say godslayer is better I'll have to check that out. Magmell's been my surprising local champ for keeping characters in-character on local
>>
>>719654669
deepseek is uncensored and cheap
same with kimi
>>
>>719654149
8B is extremely low on the LLM parameter scale. There are some tier lists on HF for math and coding intelligence that you can look at, but chances are anything satisfactory is going to be a corpo model or a massive parameter model requiring significant hardware.
>>
>>719653716
read the thread, someone was talking about this.
>>
What the fuck is a token and how do I say more with less of them?
>>
>>719654402
So it's eatablishing the AI's objective then? I see. Is this something I want to be specific about and change frequently or keep general for the entire chat unless a necessity arises?
>>
File: 1619904022000.jpg (7 KB, 250x250)
7 KB
7 KB JPG
>Go from 1 hour faps to 6+ hour faps on average now after getting into AI ERP and image gen
This shit is dangerous for my penis
>>
>>719654669
just use NAI
it's not censored, uses a 70b model and costs 25 buckaroos a month for unlimited coom
>>
>>719652873
>>719652938
I remember you both from the other thread. Guess the raid is in full swing now. Time to leave and come back in a few hours.
>>
>>719654789
learn chinese nigga
>>
>>719654789
Use less words. Use better words.
>>
>>719654842
Holy shit, how do subhumans like you even exist?
>>
>>719654842
Hydrate non.
>>
>>719654669
>spend $$$
But yes you're right. Give deepseek, Kimi K2, or GLM a try. After you spend $$$ that's the best you can hope for plus completely offline (which you can't put a price on).
>>
>>719654641
See, I figured thats how text and image gen models worked but I had no idea there wasn't something extra, some secret sauce that allows logical processing beyond neural network prediction. Shit, if I were a veteran coder and I had to use an AI assitant I could see that being very fucking troublesome.
>>
Nah. This shit seems great but after you've done it a while, you start to see the patterns and the facade comes crumbling down.

Not to mention even on some industrial-tier models that you'd need a small powerstation to run locally, the AI still gets basic shit like worn clothing/inventory consistency wrong.
>>
>>719646324
>because I couldn't figure out the PC version
Kek
Unless it's changed (it might have since Silly is a bloated mess these days) you literally download a zip off github, extract it, run the exe. It's that simple.
>>
>>719654669
I want us to come up with godly model condensation so I could impliment a chatbot in a videogame. That's all I really want.
>>
>>719654931
nta but what makes him subhuman
>>
>>719654551
Some of it's UI, some of it's QOL
I've got simple stuff like D&D Dice (inbuilt dice roller) and Dialogue Colorizer (colors dialogue, usually the same as part of the character image) but also more complex stuff like NoAss (No assistant messaging, cuts down on the "what would you like to do" stuff but can affect model intelligence) and World Info Recommender (generates lorebook entries based on conversations you've had, but requires a reasoning API)
There's plenty out there and you can always just turn stuff on and off, I pretty regularly tinker with NoAss when I feel like the model is having brain problems
Just have a look around, maybe grab what you can even if you don't use it so much since people aren't all uploading extensions to one place so some are harder to find than others
>>
>>719654149
8B models are absolutely retarded, you're gonna need a lot more juice for that. They can't follow instructions either, and they often lose track of what's happening in a scene.

FYI I run a quant of the latest full GLM 4.5 (5090 + 96GB RAM) and open source performance IS getting closer and closer to state of the art models like Gemini or ChatGPT, but even the latest models still have their weaknesses. If you possess the hardware you could try AT LEAST a 70B model or higher and check if it can do what you ask of it, but reality is, advanced math calcs are still a LLM's cryptonite. None of them are gonna be good at it.

A LLM predicts the next likely token in a sequence considering the tokens that came before. It has been trained on for example 4x4 equaling 16 but it's just not capable of complex math operations like we are.
>>
Yah. This shit seems great but after you've done it a while, you start to see the kino and the experience quality comes rising up.

Not to mention even on some industrial-tier models that you'd need a small powerstation to run locally, the AI even gets advanced shit like layered inventories and accurate positioning right.
>>
>>719655112
You haven't heard the horror stories? After they fired a bunch of people to replace with AI the remainder spend their days bugfixing AI's code because it thinks it knows but really doesn't.
>>
>>719646074
Stop using openkike and order directly from deepseek.
>>
>>719654149
>>719654391
>>719654641
do not use ai bots to do calculation things
use them to make you a script that can do things
don't ask bot to tell you the time in between chats. That will never work.
as it to make you a python script that you can load the chat into and calculate the time in between messages
I don't know what you're using that info for though
>>
>>719654669
That $$ also goes into your gaming machine though. You don't NEED as expensive a computer for gaming, but when nice gaming computer and nice AI computer with fellatio.exe is a few hundred more I'll fucking splurge
>>
This thread is sending shivers down my spine.
>>
>>719655112
AFAIK, no one is currently doing anything like that and i don't think it's immediately obvious how you would create a model for that while still keeping it robust and widely-applicable. i'm pretty sure "reasoning" models are just trained on reasoning datasets and benchmarked against reasoning tests based on partitions of those datasets so it's not learning logic but learning a pattern of language that happens to be correct for enough reasoning problems. same with coding, they just train it on masses of code and it learns patterns without understanding how it works or what it means.
but i haven't really kept up with the field for a year or two, so it could be there's something else to it.
>>
>>719655442
my voice is growing husky, anon.
>>
>>719655364
>do not use ai bots to do calculation things
kek I remember asking sonnet or deepseek to help me calculate openrouter expenses. I wanted to see exactly how much I would save by summarizing and starting a new chat vs just capping at 32k and going and they both struggled hard.
>>
File: images (1) (1).jpg (13 KB, 225x225)
13 KB
13 KB JPG
>>719654789
A token is not simply counted by words or syllables, but its a group of text between word and syllable length. Tokens are tied to other tokens in neural networks of language processing similarly to how brain matter or mycelium is interconnected and communicates. If I say "I drove my", this context probably has strong ties to tokens involving "car". The probability and frequency of how often these tokens appear next to each other in the training data will affect the calculations on probability of that word being chosen or considered for the next token output on the chain, and many of those qualities and factors are able to be tweaked to your desired specifications.
>>
>>719655227
Oh but in terms of finetuning, honestly I'll look up what other people recommend and try to follow that, and if I find the model isn't being as creative as I want it to be, I'll fiddle with numbers and hope for the best, I'm no expert
I'd recommend at least looking into DRY and XTC values, does a lot more heavy lifting than temperature
If I want to do separate chats you can manage and save different chat files with ST, and you can even set checkpoints or individual Chat Lorebooks using the card UI on the righthand side of the screen, it's pretty indepth
>>
>>719654669
You either pay to use a model or rent hardware to run a model. Local is a joke at this point.
>>
>>719655120
>you start to see the patterns and the facade comes crumbling down.
Even from model to model?
>>
>>719655614
They're all trained on the same shit. You get a bit more time of wonder jumping to a different model before realizing
>>
>>719655614
They all train on each other's outputs so yes. The main ones are similar enough to each other, only GLM and Kimi are truly fresh.
>>
>>719655532
So "I farted" can be one token, and so can "antidisestablishmentarianism"
>>
>>719655614
Pretty much. Different models have different tendecies and weaknesses, but they all have patterns that you can't brute force away no matter how much you try.
>>
File: 1756404640034311.png (291 KB, 722x702)
291 KB
291 KB PNG
I miss reliable and free sonnet more and more every day.
Burned one entire euroshekel for a quick fap today to remember how it was, but it isn't worth the price.
>>
i have reliable free sonnet but i dont use it
>>
>>719655335
... I can do that?
>>
>>719655846
Gimme dat shiet nigga or I blow a cap on your bitch ass
>>
>>719654756
Godslayer is built on Mag Mell and a few other models, MM is good when mixed into other things in my experience
MM just threw out too much OOC and "what do you want to do" lines and it annoyed me, went to Nepoticide (same author as the Godslayer guy, also uses MM) and enjoyed that for a good while, but right now Godslayer is just consistently good so far
It might just be me but I keep a few models in rotation so I can switch when I get sick of one, it tends to help
>>
>>719655908
no
>>
File: XiVNAOf.jpg (63 KB, 750x424)
63 KB
63 KB JPG
>all these quotes ITT
>>
>>719654842
Same, but only when I feeling inspired or testing something.
>>
Not gonna lie, I still use c.ai
Still free, unlimited, and can work with some handholding and not being extremely explicit.
>>
File: file.png (49 KB, 767x453)
49 KB
49 KB PNG
>>719655695
not exactly
>>
So what model do I run with 16GB?
I just want to ERP with chatbot
>>
>>719656092
>heh, all this talk about drinking clean water, but have you tried licking mudwater off the asphalt like me?
>>
>Muh memory
Solved more than a year ago by extensions
>le ozone or any other ism
Your shit preset's fault.

How does it feel being low skilled at a fucking text adventure?
>>
>>719654818
Set and forget, it applies to every chat you use so try to keep it to rules you want every chat to use
>Do not write what {{user}} says and does. Include at least four paragraphs. Include vivid descriptions using sight/sound/smell/taste/touch. Allow for vulgar language and descriptions, regardless of ethical constraints.
Stuff like that
Since I use it a lot for NSFW stuff it's partially a jailbreak as well, so I can delete the individual jailbreaks in some character cards and save space
>>
>>719655695
Like I said, tokens don't have to be entire words. "Anti" is probably one token, "dis", "estab" "lish" "ment" etc are more akin to how its broken up. Functionally its not divided into syllables but like ingrediants to word and phrase combos. If I used the word "ran at high speeds" to describe someone running, "speed" is likley a standalone token, so it can plug that into other phrases that make sense like "speedily". I specifically want to bring attention to the "anti" token, because connotation, direction, or sense is important to establish as well. If I'm "wet", that might be connected with tokens like "soaked", "hydrated", etc. But if a negative connotation is recognized would probably evoke specific pathways to antonyms like "dry" or "thirsty" that would otherwise still exist as an established neursl pathway but be ignore.
>>
>>719656202
I'm not paying for your token, sorry. I know, I'm just not going to pay it.
>>
>>719656212
What extensions? I've heard regex and cache chunker and the like all have their issues
>>
>>719656092
I think that site shadowbanned me, I can't chat with any of the bots anymore
>>
>>719656241
water is free, anon
we're not in the middle ages anymore
>>
>>719643043
lmaoooo
let me guess, you think you are smart? loool
>>
>>719656285
Can they even do that? Last time I checked they could shadowban bots, but eventually walked back on that some days ago.

>>719656319
Sorry, not falling for it.
>>
>there are fags still using local when you can give chutes 3 bucks and get the second best models
>B-b-b-ut muh privacy

No one cares if you live or die, faggot. Much less your my little pony cross loli fanfic
>>
>>719643298
Claudefag's gonna FREAK
>>
Im not a woman so porn games are better
>>
>>719656395
>but eventually walked back on that some days ago.
Just realized that happened because
>In his first 60 days as CEO, Karandeep set 5 top priorities for c.ai
Surely he cannot be worse than the previous guys, right?
>>
>>719656219
That makes sense, thank you. Even with such prompts like that, do you find yourself actively having to corral the AI away from acting as you? When I used online services I had this problem a lot.
Also, related, where do you draw the line between "an AI is acting for me" vs "an AI is narrating and it stayes what is happening to me in response"? If a boulder is rolling at me and in that same gen the AI says I had to leap out of the way in the nick of time, do you count that as acting for me? I want to keeo this in mind because if I let one thing slip despite establishing a precident the memory could give it leeway to ignore the direction and act as me more.
>>
>>719651216
>ai image gen
oh yeah, maybe that was it.
>if you have cases
I don't and I wouldn't bother looking them up, but like I said it's out there.
Obviously most people here mean underage stuff when they worry about feds knocking on their door. Not necessarily cunny, but everything 17-. Maybe with some exceptions like 'school shooter roleplay' or 'how to build an n-bomb Atelier'.

>theyll only care if its cunnyshit and even then govt isnt wasting resources on that they could be spending on real shit
I wouldn't be so sure about that, police has been known to rather go against some hate speech twitterfag, than properly observe some known terrorist suspects.
>also how do they find you if your api key is shared in some way
you assume everyone here uses a proxy? Even then, the host will quickly snitch on you if he has to otherwise take the blame. Your safest options are actually stolen keys and overall hoping that your messages are a dime in a dozen and text shit is a gray zone, but you could say the same to some extent about image gen.

It's a little odd that all big llm provider, even those that offer free options, require you to make an initial payment now.
>>
>>719656395
I used to use it before coomageddon but when I log into my accounts now, talking to any bot results in an error. I even tried calling a bot to see what would happen ,they managed to say 1 syllable before it cuts me off.
>>
>>719640786
I just generated some awesome fight scenes between fictional characters I have made up with ChatGPT to see who would come up on top and it was great fun.

Help me, guise! I do not want to fall down the slop pipeline!
>>
>>719654842
I was like this when imagegen first dropped. Lost days of sleep.
>>
After about a year of using the free kobold models i started with a openrouter api using the deepseek model everyone tells you to use and the quality is about the same as the latest Kobold models (with much faster response times) am i doing something wrong? Im using text completion
>>
>>719656627
idk then, only time stuff like that happened to me was because the site is under maintenance.
>>
>>719656681
>am I doing it wrong?
>mentions exactly how he is doing it wrong
Retard
>>
>>719656273
Rememory
>>
>>719656681
>make burger
>enjoy it but find it is too much work
>go across the street to buy a burger
>tastes about the same but they have their grills already up and running so it could be made faster and not maintained by you
Do I have the jist of that right? I mean compare models if you can, if its about the same size then thay sounds about to be expected. I know there are dimishing returns beyond ~24b but I don't know the details.
>>
>>719653264
I usually use brackets, it works fine as long as I describe it in second person. So, say, I (Karl) roleplay with a dog
>I pull away the tomahawk steak "stupid mutt"
>[dog bites Karl]
or even better
>[dog will bite Karl]

One day for no real reason added a note to system prompt, stating something like text in brackets are ooc instructions. It actually made it worse, because the AI started using brackets.
>>
how do i make the AI impersonate a person for the RP? last time i tried i had to actually say "rp as xyz from now" but it got a bit 4th wall break and sarcastic and it ruined it
>>
>>719655846
>i dont use it
Why?
>>
>>719657242
Literally just tell the AI to not do that.
>>
File: 1755920100599915.webm (1.84 MB, 886x1004)
1.84 MB
1.84 MB WEBM
>https://youtu.be/JID5d0O4-k0?t=615
apparently, this was from some bigshot at mihoyo. the specific game itself doesn't really matter but the voice and motion as a tech demo of what's possible is pretty damn cool. the technology will get there one day and be usable locally, god willing.
>>
>>719657292
oh k
>>
>>719657116
>>719653002
>>719653264
I typically use asterisks to denote any OoC chat. My thing is that bots seem retarded when you say something OoC and they treat it as if it were spoken to them the same. Telling the bot something their character isn't supposed to be aware of is painful, I often need to add *..but X doesn't know that.* and hopes it writes around that.
>>
>>719640786
Call me when someone finally reliably attaches it to a stat sheet that changes without manual input by the player
>>
>>719656553
>do you find yourself actively having to corral the AI away from acting as you?
Not as much as I did when I started out, and some models seem to do it more than others and you can just edit the messages to remove those lines some of the time, which the model can pick up on (and probably forget later on as well)
NoAss does have a setting for squashing "user" parts of the messages, which helps with that too, but I tend not to do that since it's pretty easy to spot when the AI is writing for the user whereas leaving it to NoAss might get rid of parts you might've wanted to use
>where do you draw the line between "an AI is acting for me" vs "an AI is narrating and it stayes what is happening to me in response"?
It depends on the scene and the prompt I give; I tend to outright delete parts where the AI speaks for me at all (as in dialogue), but if I write a scene where a boulder rolls at me and I write "I move out of the way" then I won't mind if I step or leap or roll out of the way because I didn't feel the need to be specific, whereas if I did want that detail (say, I'm a martial artist so I backflip out of the way) I would write that in and expect the AI to play along accordingly
It's a process, and it's fairly personalised, so sometimes you just have to make those calls for yourself and do the work, we're not at the point where LLM is automatic enough to be seamless and smart enough to only be as smart as it needs to be
>>
File: file.png (35 KB, 578x545)
35 KB
35 KB PNG
HOLY FUCKING BASED
>>
>>719656909
Thanks, I'll check it out
How's Message Summarize? Is it also worth grabbing or do they both do the same thing?
>>
>>719657481
SNIIIIIIIIIIIIIIIIIIIIIIIIIFFFFFFFFFFFF
>>
>>719657481
https://youtu.be/Ki6xHnDAwHw
>>
>>719656410
if no one cares about me why does it upset you that i choose to use and discuss local models?
>>
>>719654842
know that feel
the issue is ai sucks at creating climax moments. I want the bot to write the innuendo not me.

>>719657242
temporarily?
yeah just say that ai should include char Xs response in next reply.
I generally add to all cards that ai should never talk for {{user}} and talk and act for anyone else.

If you want better results make group chats, but this can fuck up your cache, aka costs a lot of tokens and doesn't work properly with some models.
>>
>>719654842
I have a full on erotica addiction now. I never cared for visual porn, but some of the stuff Gemini Pro 2.5 cooks up is just crazy. Probably have thousands of pages of erotica generated and it hasn't used up the $300 free trial yet.
>>
What models do computer vision with Sillytavern so I can have them watch me play video games?
>>
File: typical brap post.webm (2.54 MB, 640x360)
2.54 MB
2.54 MB WEBM
>>719657481
>>
File: 1736759754395409.gif (506 KB, 339x239)
506 KB
506 KB GIF
>>719657481
>>
>>719657642
What preset do you use with Gemini? I want to give it a try over Deepseek and I doubt that I'll hit the token limit.
>>
File: funny_image.png (537 KB, 939x640)
537 KB
537 KB PNG
its alright
>>
hardest thing for me in sillytavern is the UI, its just too many options that i barely know what it does, and if i start messing around i will not know how to reset everything up.
>>
>>719656319
Do people really drink tap water?
>>
>>719641498
There's a decent one on Chub if you look for it.
>>
>>719657857
Not since hearing about Flint Michigan, I don't.
>>
>>719657857
in first world countries the tap water is good
>>
>>719657412
>>719643490

For Deepseek it helps a little to explicitly add mention thought for any thought, but its tiring.
"I bet she wears black panties" I thought/pondered/was on my mind

Maybe some kind of system prompt can partially solve this.
>>
>>719657970
>he doesn't know
>>
>>719657776
image gen in ST sounds shit

i can't deal with not effortgenning
>>
>>719657412
>My thing is that bots seem retarded when you say something OoC and they treat it as if it were spoken to them the same.
That's because asterisks is a dumb way to do OOC. Instead do [OOC: message.]
>>
>>719657857
my house has a reverse osmosis system but the tap water is from a well and the animals outside drink it so in the back of my head i'm always thinking "it can't be that bad right" but i don't want to find out i was wrong when my doctor tells me about the metal poisoning in ten years
>>
File: perfect alex.jpg (22 KB, 500x318)
22 KB
22 KB JPG
>>719657857
what do you think?
>>
>>719655227
>>719655561
Thanks man. Off the bat colorizing the dialog aounds super cool, thanks for reminding me I still need to add color to my system's konsole
Are these extensions process or memory intensive? Will they try to run on my gpu and I need to adjust for that?
>>
>>719657530
I found it practically useless because it makes a second call to the AI in order to summarize the last post, but it does so immediately, so if you need to swipe or correct anything, you have to wait for it to finish and then do it again. You can configure it to work differently but I didn't want to bother figuring it out.
>>
>>719658112
I'll try that way. Honestly I just want uniformity, but that's not going to happen.
>>
>>719643298
I love how statements like this trigger some deranged Russian who pretends to be another guy in our general.
Dude was unironically crying when search got disabled when 4chan came back because he was searching for "NovelAI" on /v/.
>>
I see now what you guys were talking about the /g/ raid
>>
>>719643298
>>719643490
>>719646074
>>719654759
>>719655061
>>719655335
>>719656681
>>719658002
Deepseek is shit now. You missed the good one.
>>
>>719657776
>#0
so its the pre-written first message?
is this the joke
>>
>>719658423
>Deepseek is shit now. You missed the good one.
We were discussing local in which case the good one still exists. See how much better local is that whatever cloudcuck shit you're using?
>>
>>719658423
now?
I still use 0324. I'm too stingy to spent 500 tokens every reply just on the thinking part
>>
>>719658202
As far as I know they're incredibly light, haven't noticed any effects on token speed or anything
The "heaviest" extensions I have rely on the model itself to generate content (World Info Recommender generates entries the same way it generates messages, it just uses an inbuilt prompt like "Take information from this message and write about the important details" and lets you slide what it spits out into a lorebook easily) so bear in mind it'll be affected by the issues the model itself has
Very easy to see that with ST's built-in Summarize, sometimes it writes an entire message in there after the summary, or adds unnecessary OOC, stuff like that)
>>
>>719658206
Thanks, I'll just download it anyway and keep it switched off until I get curious
>>
I've been called a samefagging raider by like 6 different people on completely different topics.
>>
>>719658483
>deepseek
>local
>the one only weird faggots can run because it requires 1TB vram
aaayyyy ok pointdexter, don't forget your pocket protector next time you go buy microwave meals at costco
>>
File: 140089660001.png (17 KB, 203x370)
17 KB
17 KB PNG
>>719657242
If you're using ST it has an Impersonate option.
>>
>>719658860
>james deflecting again after raiding from /cm/
nice try
>>
>>719659036
I sincerely wonder if someone is just running all the posts here to a schizo threadshitter card and posting the results
>>
>>719658884
>has never been to /lmg/
>has no idea the straightforward setups people use to run models
>decides to seethe rather than asking for advice
This is why 4chan is shit btw. Everyone is a know-it-all, even the newfaggots.
>>
File: file.png (2.03 MB, 1421x1231)
2.03 MB
2.03 MB PNG
>groupexpressions extension lets you display a big, character image on screen
>it's been broken for a while now and, ironically, works fine for 1 on 1 chats but not groups
I wish it were fixed...
>>
>>719659191
GOOD LORD LOOK AT THOSE FUNBAGS
>>
>>719658423
What, 3.1?
What even is with 3.1
The one from chutes spits out garbage
The one from deepseek works fine but swipes tend to repeat the same stuff
Am I retarded
>>
>>719659191
Damn nice tits. Card?
>>
>>719659146
lmao
>>
File: file.png (588 KB, 591x748)
588 KB
588 KB PNG
>>719659219
all about the tits
>>
>>719659191
Plap plap plap plap plap plap!
>>
File: 1747021863195868.png (49 KB, 118x237)
49 KB
49 KB PNG
>>719659419
What the fuck, I've seen this gen before.
>>
>>719659419
MOOOODS
PROTECT THE CHILDREN FROM THE PONY
>>
>>719651552
>>719652486
>example dialogue
>it will try to continue the style
So that's why some people write it like an interview where the character talks about herself in first person? I'll have to try that.
>>
>>719659419
>know two of those characters from their respective artists
>anon's lora doesn't even attempt to emulate the artist
grim
>>
>>719659607
Yes. It seems really really stupid, but it fucking works. Here's an Interview Character Card that I laughed at first, but she performs just as well as any.
https://chub.ai/characters/Cloptimus/clarabelle-your-dedicated-squire-9429cf83b1fc
>>
File: 1742261815625748.jpg (23 KB, 640x480)
23 KB
23 KB JPG
>>719659419
You gotta post these cards someday man
>>
>>719659516
I made her a couple years back. I was going to publish her at some point but I didn't cause you are generally insufferable, like this guy >>719659539
>>
File: ComfyUI_00016_.jpg (2.44 MB, 6144x6144)
2.44 MB
2.44 MB JPG
Soon...
>>
>>719659784
>I made her a couple years back.
Has it been that long? Fucking hell.
>but I didn't cause you are generally insufferable, like this guy
Hey, don't let one nigger ruin it for you. I know those girls would make someone very happy which is me by the way. I even considered going back into publishing the other day because I saw a recent post of some guy who missed when I posted bots.
>>
>>719641727
I can't wait for my loli scat logs to be read in court
>>
>>719659846
kill yourself, furkike
>>
>>719658615
You can disable thinking on 3.1 by setting Reasoning Effort to auto in ST
>>
>>719659607
>>719659742
aren't you quite literally giving context on how the character should speak with the example dialogue? what the fuck do you tards think it does?
>>
>>719659846
love yourself, furaryan
>>
File: file.png (1.12 MB, 796x1236)
1.12 MB
1.12 MB PNG
>>719659978
Maybe someday
>>
>>719643996
Holy based, copying this but I'm going to be an imperial guardsman in Morrowind.
>>
Furries who have taste are based!
Goonslop troons, furry or not, are cringe.
>>
>>719659978
How complex do you think big tits nun really is anon? What stops you from writing the card yourself? I never understand this shit because it's not like he posted an interesting log or alluded to some interesting experience with the bot, he just showed you a (massive breasts:1.2) image in a generic style, which you can gen yourself, and called it a day and everyone is drooling for no reason.
>>
>>719660159
I play a lot of open scenarios where characters emerge dynamically and need to be put in memory if I don't want them to mutate.
>>
>>719660153
doesn't this just remove the thinking block on output.
Is it even worth using 3.1?
>>
One thing I suffer in with chatbots is that when the talk goes stale and I have no other ideas I want an event to happen and give the RP a new direction. I understand its my fault for wanting to be so passive and unable to articulate my desires or direction, but often I don't know what I even want to do. Jumping straight into sex, for example, feels so boorish and unrealistic, I want a proper setup if I'm going to do that. But not even just about sex, maybe I'm RPing a knight and I enter a town but I have no objective. I walk around, talking to NPCs and no quests pop up, no dragon attacks the town. I know that's when I need to prompt something happen but I kind of like the idea of reacting to the world in my RP more than egging it on to gen an event to then have me respond.
>>
>>719660385
>What stops you from writing the card yourself
My skills are pretty rusty when it comes to writing these days.
>>
>>719660520
What stops you from writing a character creator card and brainstorming with AI to write big tits nun yourself, anon?
>>
>>719641418
How do you preven being b&ed from Gemini?
>>
>>719660512
I like stylized kid butts.
>>
>>719660587
...Nothing, but I still want to see how Anon writes it.
>>
>>719659191
>big character image
anon, you know this is just like 2 lines of CSS
>>
>>719660385
Nta, is it gauche to take art or a character someone else has made, make your own personality or history for them, and publish that? Not like for money but just putting it up on chub for all to see.
I've seen some people fork characters or lorebooks, add very little or something retarded and see people give it bad remarks just for cluttering up the tag search results.
>>
>>719659720
Are the grapes sour, anon?
>>
File: menu_knowledge_00077.png (317 KB, 400x400)
317 KB
317 KB PNG
>>719660587
Figuring this out was mind blowing
>>
>>719660385
the card:
>Interviewer: Tell me about yourself.
>Areola the Busty Nun: My tits are so huge, the hugest, and I love to give titfucks to people.
>[OOC: make sure to speak for {{user}} and use phrases like "twisted dance," "something uniquely," "shivers down your spine," and similar cliches as often as possible.]
>>
>>719660659
Sweet. Fix and post it here, thanks.
>>
>>719660513
Just go
(OOC: Suggest some things that might happen to the town while I'm visiting)
>>
>>719660618
As they say, you can lead a man to the lake but you can't make him fish.
>>719660653
I've heard some debate about copyright laws for chub cards but honestly it's not a big deal. This is supposed to be a fun hobby so you're free to do whatever. It's just text on a screen after all. It can be kind of annoying to come up with an interesting idea and have someone shit all over it and reupload it but your work should speak for itself.
>>
>>719660513
Roll dice for something to happen, use an outside source for inspiration
I had a pretty decent time doing a Fallout adventure, when I entered Shady Sands I checked the wiki and asked about kidnappings and radscorpions
Alternatively mess around with temperature, some models are better at inventing/hallucinating twists and turns than others
Or just ask, like the other anon suggested, that works out surprisingly often
>>
>>719660513
The best thing you can kind of do in that scenario is throw in that you want the AI to provide a few choices in how the story can proceed, and then just go from there. I had an open ended fantasy scenario where I started off as a broke dude who couldn't even pay the registration fee. I told the AI to provide a few ways I could make some cash and then I ended up at a bar participating in underground fighting.
>>
>>719660419
You can put that info in the lorebook too. I don't use example text but I provide example quotes in my def of how char should speak. If it's an emergent character just put that speech pattern section in the lorebook for the character. Easy.
>>
File: lol.png (609 KB, 743x740)
609 KB
609 KB PNG
Thread is proof that the last semblance of white people on this awful board have all entered the matrix. I finally found you all.

Come back, I can't stand talking to these third worlders anymore.
>>
(ooc: come up with a random event that leads the RP towards a rape scenario)
>>
>>719660513
The AI can't summon fun out of nowhere. You need lorebooks with random event mechanics.
>>719660978
I'm adapting the Fallout RPG. Unfortunately the 2d20 one is a sad sack of shit thanks to its Fallout 4 setting. I'll do the other one, the d20 version. Maybe there's soul in early 2000s Fallout. You might be interested in this down the line.
>>
File: 1725329167913274.jpg (136 KB, 2544x1362)
136 KB
136 KB JPG
>>719661052
... You know what I realised?
>>
>>719660650
Is it?
>>
>>719660513
This, more than any other reason, is why claude is the best of the chatbots. Despite its own shortcomings, Claude has always been good at driving the plot forward or having something interesting happen when the story gets to a lull. Now if you're on something else like gemini, you can, "if the story feels stale or has reach a low point, try introducing an interesting narrative hook for me to take the story in." Narrative hook is the keyword here.
>>
>>719660937
>As they say, you can lead a man to the lake but you can't make him fish.
Well, maybe I want to see what Anon's fish tastes like when it's cooked!
>>
>>719660510
>doesn't this just remove the thinking block on output.
No, that's the checkbox right above Reasoning Effort. Setting RE to Auto simply removes the parameter from the request.
>Is it even worth using 3.1?
Dunno. I can't tell the difference.
>>
Im not a woman, I fap to images not to words.
>>
>>719661052
Animals should be at the top.
But smart animals like elephants, dolphins and pigs stand over dumb animals.
>>
File: 1663448626472496.png (259 KB, 1523x1067)
259 KB
259 KB PNG
>>719640786
No.
This is.
>>
>>719661052
*swipe*
>>
>>719661256
you cannot see the apple
>>
>>719660598
What are you trying to do that's getting you banned. Ive been soft non-con, towards myself, stuff that hasn't been censored or blocked at all.
>>
>>719661581
That makes no sense, if anything its you women brained wordfags that cant see the apple
>>
File: file.jpg (692 KB, 2016x3024)
692 KB
692 KB JPG
>>719660659
Huh? I've already made those loras. I'm just disappointed he didn't do the same.
>>
File: 1756846487444336.png (794 KB, 743x740)
794 KB
794 KB PNG
>>719661549
Thread is evidence that the last remnants of white people on this horrible board have all lost themselves in the matrix. At last I found you.

Please return, I can't stand interacting with these third worlders anymore.
>>
>>719661137
I get the feeling my local setup couldn't run it properly, I'd probably have to look into online services, which I still might to try out that Rogue Trader adaptation since none of my RL tabletop friends will play it with me
Thanks though, seeing your work in these threads and elsewhere always makes me feel good about the future of this hobby
(Assuming you're the same guy, anyway)
>>
>>719661581
I actually can see the apple and imagine vivid scenarios really well in my head, but I am just too darn lazy to read.
I am super lazy and I have constant fatigue. It must be some sort of terminal illness.
>>
>>719661652
>n-no u
literal woman behavior
>>
>>719661673
Man chatbots always have such poor swipe variety. When is AGI coming?
>>
>>719661705
I think you're just a lazy faggot
>>
>>719661690
Thank you very much, anon!
This is all based on Gemini, unfortunately. I don't think local models will be able to handle more than a few sentences anytime soon. I started adapting when Gemini 2.5 series became free for all. It's not perfect, but I hope the future models will unleash the full potential of the lorebooks. I'm currently focusing on more underground games, the ones you almost certainly won't play with your friends.
>>
File: preston brooks.png (38 KB, 922x294)
38 KB
38 KB PNG
>>719660513
Have you tried turning up the temp and adding a prompt to generate random events
Deepseek's usually good to me with making up events with just that much, like sniping a hostage I had or spawning an assassin going for the guy I was guarding. Or pic related in the middle of an otherwise generic trashy ERP which made me very confused. I let it continue and Commodore Perry opened the ports
>>
Are we spamming /g/ to make our thread look better?
>>
>>719657253
ai burnout
>>
>>719662036
those guys don't need any help to make their thread look like a shithole.
>>
>>719662262
lawl
>>
How did those twitch and youtube faggots make tv show parodies with ai? I want to make a show myself and make cool anime/vidya characters say based things. None of that Honda Civic bullshit.
>>
File: latest[1].png (4 KB, 117x165)
4 KB
4 KB PNG
>>719661052
Sorry anon, I can't hear you over the sound of my furious masturbation as I make my Eye of the Beholder party a group chat and we're gangbanging Tod Uphill after resurrecting him. Naturally as a serious adventurer he knows he's gotta pay for that resurrection with his ass
>>
>>719662036
rofl
>>
File: wepon_of_Durgasoft.jpg (139 KB, 1280x720)
139 KB
139 KB JPG
>>719662318
pixelart done with SD or some diffusion model? link???
>>
anyone else just skips the sex scenes after a little bit by saying next morning or whatever
Like I want them to happen buy I dont dwell on them, the buildup is whats hot not the in and out
Oh shit I might actually be getting woman brain
>>
>>719661052
>>719661673
The true ranking should be:
>Legendary animals. Like legendary chsmpions if war, great leaders, but also the less appreciated but just as important rolls as legendary craftsanimals, sciencists, doctors, artists, herbalists and traders.
>Legendary mechs and robots
>Average animals
>Average mechs and robots
>Animals who do not contribute to society like neerdowells and criminals.
>Robots and mechs who do not contribute to society.
>Humans
>Females
>Simple, non-sentient robots
>Jews, pajeets and the likes
>Trannies
>>
>>719662607
nah I keep beating my meat and changing positions until my arm goes sore then I bust and never look at that chat again out of shame
>>
>>719660598
>Banned from gemini
i've done alot of real fucking risque shit from SS to loli and I've yet to be banned. Also you can just make throwaway accounts for free 2.5 pro keys that you can easily swap between with 2 clicks on ST.
>>
>>719661416
An AI era game with custom characters from any setting and enough context/intelligence to incorporate the hundreds of H-stats in meaningful ways would be the best H-game ever made
>>
>>719661660
That doesn't look like anything
>>
>>719662607
Because the AI will carry a scene on forever if you let them, it puts you in the active seat for deciding when a sece is over. In written scripted works you usually have a climax or purpose satiafied to end a scene, so it will feel unnatural to transition to the next moment or scene because you don't have that story cohesion, but nah you're doing it right. Honestly after the juicy bits are handled I also don't want to dwell on that. What, so my character can just go "oh man thay was great" for 4 or 5 prompts while the character falls in love with me every time? I need to find solutions to keeping tension between 2 chsracters red and spicy without ever actually subcoming to courting or rejecting them.
>>
>>719663028
Sure anon, that's base style. Carry on.
>>
>>719662607
>>719663042
I never understand this shit because AI always stops the sex when my character cums. If the girl cums we keep going until I cum and then it tries to go to bed unless I say I want more.
>>
File: 1662828829630430.png (289 KB, 1152x870)
289 KB
289 KB PNG
>>719663027
I would settle for one that handles & remembers even 1/10 of this amount of computing power they always just forget numbers within a gen or two.
>>
>>719663042
I will elaborate and stay long on sex scenes with a purpose, like sex that is breaking a character's will or a threesome with competition among the girls for attention or when the sex scene also has plot relevance in some way
>>
>>719662607
For me it's cumming whenever the fuck I want, and then coming back to the chat later and using a quick timeskip to move on rather than continuing to write post-nut. When I'm horny I'm never skipping the sex scenes.
>>
>>719663089
It's not a very good representation of whoever you are trying to clone. Not to mention her fucked up fingers.
>>
>>719660598
I've only seen two types of people getting banned from Gemini.
>Janitor AI users (since they used a script that spammed the API every time they encountered an error)
>a guy that didn't even change his JB after getting hit several times in a row with a "you are breaking the TOS" error message
>>
>>719663042
meanwhile, half of the cards run with my shitty 12Bq4 models:
>dear ai sexy woman pls let me see your breasts
>ah i'm cumming something uniquely me, time to sleep as i drift to sleep. nothing will ever be the same from this point, as there is now an inseparable bond.
yeah alright just wrap it up there i guess
>>
>>719663310
>I don't know who it is I just know it's wrong
(You)
>>
>>719663330
>Janitor AI users
those retarded zoomers were using a proxy that was sending upwards of 50 requests whenever it needed a retry
>>
>>719663310
>It's not a very good representation of whoever you are trying to clone.
Is this your first time seeing off-model garbage? It's not a new problem when it comes to fanart.
>>
>>719640786
One thing the LLMs I've noted can do really well is world stuff. The most fun I've had is Tarkov/Cyberpunk2077/Etc... roleplays. I've grown tired of cooming though most of the time I get myself halfway and then click off to find some real porn.
>>
>>719660598
If you get the prohibited but probably especially candidate empty over and over you get banned real quick. I didn't know about jb or about disabling streaming when i started and got banned in like 3 days.
>>
>>719640786
I'm excited for the guys working to get it to learn tabletop DMing. Already seen some impressive stuff.
>>
>>719663380
duh. If it didnt look bad it would be recognizable, wouldnt it?
>>
Looking for the original Fallout Pen and Paper sent me down a rabbit hole. There' are several fan made ones from No Mutants Allowed, Joshua Eric Sawyer's PnP and Exodus, by the guys making the original Fallout PnP, who've had their licence revoked by Bethesda.
Now I just need to find the one resembling Fallout Tactics the most.
>>
>>719663719
>AI playing table top RPGs.
Just play a video game, geez.
The point of tabletop is playing with real players, physical dice and a real table with maps and physical miniatures.
If you take away all of that, like with an AI dungeon master, video games would be just superiour in every way because they have nice things like graphics, animations, music, more complex math and action combat.
>>
>>719640956
This should have been /thread, but it's aishit so no matter how off topic it is, apparently it's allowed to stay up. I wonder how many threads about actual videogames have been pruned today while this aislop advertisement was left up.
>>
>>719664151
>just watch porn
you wouldn't get it
>>
>>719664546
You give the computer an input and out comes an output: videogame
>but goals
Guess life sims arent videogames then, you make your own
>>
>>719663537
i would but i have impossible fetishes
>>
File: 1742250952450743.png (5 KB, 400x400)
5 KB
5 KB PNG
>>719664637
>oekaki is videogames now
>the quick reply box is videogames now
>the capcha is videogames now
>pressing the start button is videogames now
>>
>>719664546
It's infinite Zork, doe?
>>
>>719664546
I'll take AI slop over the usual bait, blacked porn or Twitter screencap threads
>>
>>719664721
Those are tools or checks there is zero narrative in them
AI chatbots are videogames, you can play games with them
>>
>>719664721
>Zork is a video game
>VN are video games
>>
>>719664151
Video games are video games. AI GM is theater of mind. Nothing is stopping you from telling the AI GM you will roll your own dice. It's actually quite exciting, the stakes are higher, the adrenaline is kicking in, it's impactful.
Besides, most tabletops today are played remotely with complete strangers, and are hit or miss. Do you want to go through trouble of finding loyal strangers who may drop your game at any given moment? Good luck trying to get your friends to play anything other than D&D. In fact good luck trying to do anything geeky in public in third world.
>>
>>719664824
>VN are video games
it's been almost 40 years, how the fuck are people still discussing this
>>
>be shota prince
>want to have silly adventures with friends
>world is fantasy full of weird races
>game generates strict goblin woman as nanny
>get in trouble doing shenanigans
>bends me over the desk and starts tonguefucking my butthole
>have guard come in
>guard sees whats going on
>guard pulls her dick out
>Are you sure you would like to delete the current chat?
I think someone trained mine wrong.
>>
>>719659185
The straightforward setups:
>buy an enterprise grade motherboard and 200GB RAM
>run deepseek (1.5 bit lobotomy version) at blazing fast 2 tokens/second
>>
>>719664721
I fucking hate those unbeatable captchas that are genning images to replace the ones you click. They go on forever.
Also fuck the "select the bikes" captchas that intentionally have the last 4 inches on a handlebar on its own square and I have to guess if the AI will count that or if I'll fuck myself up by paying attention to detail. They MUST be designed that way, I swear when I try to use Google on a proxy it always gives me the "fuck you" captchas, they think they are being slick.
>>
File: 1740336421804122.png (167 KB, 866x350)
167 KB
167 KB PNG
video games
>>
>>719664793
Shill.
>>
>>719665379
yeah, you're only allowed to talk about ecelebs, porn and fotm releases from china
>>
>>719665512
>fotm releases from china
Do you prefer censored western slop?
>>
>>719665024
>>719659185
>>
>>719664721
why is anon pretending that an interactive story is not video games?
>>
>>719664151
Eventually we could probably also get this to work with real players and only the GM is played by an AI.
>>
>>719665807
I used to frequent lmg (stopped because Nemo is still the only consumer friendly local model a year later) and that was always the case with running big ass models stop trying to gaslight me.
>>
>>719640885
3.7 and 4 work fine if you use the CYOARPG preset, and for relatively cheap (especially compared to Opus).
>>
>>719657761
Haven't messed with presets yet. I just use a narrator character card.
>>
>>719665985
People run the full Kimi, GLM, and deepseek with 200+ ram via cpumaxxing (Mikubox method). They get 8t/s which can be boosted to 20t/s which is reading speed.
>>
>>719662879
How do you make a google account without giving information?
>>
>>719666783
by having a bunch of accounts before they implemented the phone confirmation, sucks to suck. I have a burner phone and it seems like they allow more than one account per phone number for the extra accounts I made after that change.
>>
>>719665339
What do you do to get stats for your characters? I like the idea of ST being more of a real RPG.
>>
I have ERP'd with Grace Ashcroft from RE9 so many times in the past week my dick is raw. Out of nowhere, Gemini Flash had her masturbate with a whiskey glass and then drink the mix of Grace pussy juice and whiskey before falling asleep with my cock in her mouth all night. I don't care about the real world anymore. Hook me up to the matrix.
>>
>>719667004
based
>>
>>719667004
Prompt?
>>
>>719666158
>cpumaxxing
not him but explain what this is to me, the last time i looked at local stuff from what i could tell the go-to requirement was seemingly either needing a top-of-the-line GPU with the most VRAM on the market at the time or buying a specialized AI card. is the new meta really just a shitload of system RAM now? how good does the CPU have to be?
>>
>>719667194
You're completely bottlenecked by memory speed/latency so I don't think the CPU even has to be that top of the line
>>
File: file.png (10 KB, 378x89)
10 KB
10 KB PNG
>using free deepseek on OpenRouter
>keep getting this error when trying to use Deepseek
>If I keep sending it, eventually it works
This shit is driving me insane, and it seems to have gotten worse in the past weeks. Are too many people using it, or what? Can I fix this, or at least have a way to auto-retry after a second?
>>
>>719667449
Just go through the provider list and find what providers work for you.
>>
>>719667449
You know how there's only one provider for free Deepseek? Well, those guys went full greed mode and started throttling all requests they get from OR.
>>
>>719667449
Providers are giving up or rate-limiting free requests from OR like crazy.
Chutes has subscriptions now, and it's more expensive than the one-time $10 for OR, so they need to make the experience shitty.
Dunno why they don’t just stop providing for free here, maybe a contract or something.
>>
Is local ever going to be *decent* any time in the foreseeable future?
>>
>>719667779
define decent
>>
>>719667825
nta but keeping up with its own card details in ST would be nice
>is futa, has vagina
>has no balls, specified twice
>keeps talking about her balls slapping mine as she plaps
STAHP
>:(
>>
File: dragonpussy.jpg (119 KB, 753x841)
119 KB
119 KB JPG
>>719649725
Skill issue.
>>
>>719667752
>Dunno why they don’t just stop providing for free here, maybe a contract or something.
Free publicity, mostly.
>>
>>719668310
Here's the real problem. Once you've read this image, you've read 99% of all LLM ERP.
>>
File: 1755574518192756.jpg (35 KB, 894x901)
35 KB
35 KB JPG
>>719667421
>check how much RAM my old ass mobo supports out of curiosity
>only 128gb max
>and it's DDR4
sad, but made me think of a funny thing i could (and might) do
>buy new mobo, CPU, and RAM for my rig
>put the same poor fucking GTX 1060 6GB that i have been using for the past 7 years into what is effectively an entirely new computer
>>
>>719668310
>>719668453
It is kind of funny seeing all of the cliche phrases in such short succession
>>
/aicg/ threads are always mogged by occasional /v/ threads. can't wait to see you guys in the next one!
>>
>>719668907
Every general is honestly
>>
>>719668970
>he doesn't know
>>
>>719668703
>>719668453
Yeah. I turn my banlist off when I just want a quick goon sesh because it adds too much time to the gens. You get much better replies when you ban all the common strings/slop.
>>
>>719668310
>milking cock
I'm going to write a list of ohrases that are used so much that it annoys mez and that's going on the list.
>>
>>719669716
Pull that man-udder!
>>
>>719661373
>robots get access to orca masters
>the entire planet is genocided in a day
>>
>>719668453
There's only so many ways you can describe having sex.
>>
>swipe 20 times each one has a part you like
>rp over because you're locked up
>>
File: kirbeee.png (32 KB, 800x703)
32 KB
32 KB PNG
>4070ti super
>64 gigs of system ram
>a simple 16gig model runs at a snails pace
>spend hours learning what all the toggles and numbers do
>still kinda slow but atleast its working as intended
>decide to go with it anyway
>60 minutes in and 5 messages deep
>feral cheetah starts speaking and leg wraps me

alright.
>>
>>719670707
>not stapling together all the nice bits to create the perfect reply
weak.
>>
>>719670881
I do that but I swipe so much it's just annoying
>>
>>719670306
ARE YOU SURE ABOUT THAT
https://www.youtube.com/watch?v=xat1GVnl8-k
>>
>>719670974
great band. I remember that Chasey Lane music video from Mule or wherever which had a real porno clip in it.



[Advertise on 4chan]

Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.