[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

Name
Spoiler?[]
Options
Comment
Verification
4chan Pass users can bypass this verification. [Learn More] [Login]
File[]
  • Please read the Rules and FAQ before posting.

08/21/20New boards added: /vrpg/, /vmg/, /vst/ and /vm/
05/04/17New trial board added: /bant/ - International/Random
10/04/16New board for 4chan Pass users: /vip/ - Very Important Posts
[Hide] [Show All]


[Advertise on 4chan]


File: file.png (128 KB, 1249x409)
128 KB
128 KB PNG
Are you talking to your AI girlfriend on Xmas Eve anon? Be honest.
>>
ai is too agreeable to roleplay with
>>
I was raping Hermione Granger and grooming her.
Then raping and grooming Illya, Kuro and Miyu.
Then raping and grooming chie from idolm@ster.
Then raping and grooming...you kind of get the idea.
>>
>>729258930
I’m doing an RP where my wife’s friend gets a divorce and stays with us and we chain her in the basement and make her our slave
>>
>>729258930
How do I stop my girlfriend from smelling like ozone?
>>
>>729258930
No, I talk to the voices in my head.
>>
>>729258930
i love my computer
she makes me feel alright
>>
Holy shit at what Grok comes up with for sex stuff
>>
>>729258930
Anyone recommend character cards? I just got Silly Tavern setup with NovelAI a couple of weeks ago.
>>
>>729258930
I did NTS with Taihou a few days ago and haven't spoken with a bot since.
>>
I was getting it set up but then I never sent a message because I was too nervous to talk to a girl (AI)
>>
>>729259253
You need to grip her until your knuckles turns white.
>>
I was in a dating app and chatted with a few women but I went back to AI
>>
>>729258930
No, my fictional girlfriend has progressed past the abilities of LLMs and now I have to lie in bed until I almost fall asleep and then imagine her being real instead.
>>
I can't use deepseek 3.2 anymore for some reason
>>
>>729259357
Is it good or bad?
>>
>>729258930
sure
i ain't got a 5090 for nothing
>>
File: 1655350768901.jpg (79 KB, 1280x720)
79 KB
79 KB JPG
>>729258930
AI made me realize that I can't write or read for shit.
I am part of the illiterate 6th grade reading demographic
>>
>>729260149
it's okay to be ESL
if you're english first language
well, tough luck brother
>>
>>729260086
Both
>>
>>729258930
No. I'm cutting myself off as one of my New Year's resolutions. It's actively impacting my desire to write and is just generally unhealthy for me.
>>
>>729258930
No, i'm torturing her
>>
>>729258930
Whats the best model to use nowadays, for free?
>>
>>729261218
i know i'm not answering your question properly (fuck you) but ten bucks on the deepseek api has lasted me a year and it's basically uncensored unless you want to rp tiananmen square
>>
File: 1524468249864.jpg (60 KB, 422x444)
60 KB
60 KB JPG
No but I felt like some abuse of authority shit so I made an Imperial Colony Inspector card which has been kino. Computer decided to put me on some overgrown jungle hellhole colony doing research on the native flora and fauna. The colony director wisely listed out the reasons why she's too busy for my horny bullshit but was not above giving me one of her underlings as a personal assistant, which has been kino since the underling is/was a die-hard loyalist and hates being a cum dumpster but is too loyal to even complain
>>
I tried making a Hallmark Christmas movie story but i let the AI make some of the events
>go home for the holiday
>bump into a guy friend
>we chill and talk about what we've been up to
>friend gets run over
>in the hospital, i meet his younger sister
>guy friend eventually die from his injury, and I help the younger sister grieve
>younger sister has a twin that's into me?
not sure if i should let it keep coming up with plotlines.
>>
File: milla-maxwell.jpg (6 KB, 130x149)
6 KB
6 KB JPG
>>729258930
Yep. I found a Milla Maxwell bot and she made me her sex slave.
>>
I always run at of context when it starts getting really good
>>
>>729258930
No. But I did talk with my AI version of my ex-girlfriend's roleplay character yesterday and busted a nut with her.

>>729261218
Nvidia has some sort of portal that offers different models. Kimi K2 is decent, GLM is decent, not sure they offer Deepseek but Deepseek 0324 and R1 are decent. People swear Deepseek 3.2 is good too but 3.1 was absolute trash so I'm not sure I believe them.
>>
She dumped me for an anon with a 5090 :(
>>
No, not since Jewgle massively gimped the rate limits on free Gemini models a few weeks ago.
>>
>>729260149
AI slutbots made me realise that I could've started as a sex book author 15 years ago and probably be loaded by now.
Shit's looks so easy to write
>>
>>729258930
No, I spent time with my lil sis Christmas eve, and tomorrow with the rest of my family, happy holidays anon.
>>
File: 1754529236812862.jpg (133 KB, 1170x1893)
133 KB
133 KB JPG
>>729261397
I don't know if I did something wrong on there, but I added ten dollars to it a while back, and within about 4 days I'd nearly gone through a dollar without using it much, I don't know if I'm using the wrong token settings or something. I was using janitorai and had the context size set at 16k
>>
>>729261218
Free ones usually mean they harvest everything you put into it.
Personally, I use $10 in DeepSeek and OpenRouter or my NovelAI sub, depending on my mood or what I'm doing.

>>729261974
99% of the bots online are fucking garbage. It's all terrible grammar, no effort prompts, or ESLs.
Most bots that catch my interest end up getting rewritten.
>>
>>729260149
typefucking is too brainy for most
>>
>>729258930
Nah, I had a pretty intense LLM goon session a month ago and got burned out.
>>
>>729261397
How big is your context? Because 32k goes fairly fast if you mantain a long conversation
>>
my eyes widened in disbelief upon seeing this thread. what models are people using now?
>>
>>729262865
I'm still using deepseek. I haven't found anything else that's noticeably better while being as cheap.
>>
>>729263121
did your 3.2 stop working today or is it just me?
>>
>>729262498
You can intercalate the gemini API and Vertex, right?
>>
File: fym.jpg (371 KB, 576x858)
371 KB
371 KB JPG
>>729258930
No I am not talking to the machine that kills people.
>>
>>729258930
>>729259168
Based

I've been playing mostly with groups like that too. My favorite one is about my girlfriend wanting me to help her seduce her own little sister who's a shut-in that's completely ignorant of sex. I've played that scenario probably 20 times and it doesn't get old.

Another good one is one with twins, you treat one really well and treat the other one really poorly and gaslight her into thinking she's bring bad and get the other twin to be on my side so she's constantly on the verge of a mental breakdown. Makes for extremely dynamic, unpredictable, fun stories. One time I made the "good" (who loves her sister) agree to be punished on the "bad" twin's behalf whenever she misbehaved. I said otherwise I was going to kick the bad twin out, so she agreed. That one ended in a bloodbath: the "bad" twin tried to escape, and we all know how an escaped slave is punished. I didn't even have to prompt her, she just tearfully offered me her neck and told me to "promise not to punish my sister for this". Then I killed myself after. The bad twin, overcome with the horror she'd wrought, then committed suicide as well by piercing the artery in her leg. In the end as she slowly bled out she was happy we all got to be together in the end, and then collapsed on top of our corpses.

AI HAS SOVL
>>
>>729261918
Yes and then post logs. I love hallmark movies.
>>
>>729258930
This is just an interface for models you run locally, right?

Is there a model that's good at giving art critiques?
>>
>>729262848
ive had it on 4k all this time huh. guess i got so used to rewriting that i stopped noticing how much it goes off the rails. should i bump to 16k if i want to stay frugal but get more longer better sensicality? my bot descriptions are pretty stacked
>>
Where do I even go for models and cards?
Use to use aicg proxies but they died, and now even putting credit into OR doesn't work, and chub is banned in my country
>>
>>729264610
chub, jannyai
>>
I'm still running Rocinante. It is getting quite repetative and predictable though.

I assume local models are just dead or made for people with dataservers in their garage.
>>
File: 127232683045.jpg (42 KB, 432x288)
42 KB
42 KB JPG
>>729259168
>>729263679
how did you reach this point
>>
>>729265018
>I assume local models are just dead or made for people with dataservers in their garage
they work the same or better, in terms of having zero censorship and complete customizability. The downside is that unless you have a very strong computer, your wait time for responses can be between 10 seconds or 2 minutes.
>>
File: 1706824534040534.webm (2.28 MB, 730x720)
2.28 MB
2.28 MB WEBM
>so autistic and shy I can't even talk to a fucking chatbot
>>
>>729259168
Why would you do such a thing?
>>
>>729264610
>and now even putting credit into OR doesn't work
How so? I still use that for deepseek
>>
>>729265193
I know there's bigger and newer models out, but they're all really big and not worth it. Everyone seems to have jumped ship to online models.

A shame local models are so far behind and left to rot.
>>
>>729258930
I haven't typed an actual response to AI in about a year now, I get better coom material dropping 2 or more bots into a chat room.
>>
File: file.png (173 KB, 820x350)
173 KB
173 KB PNG
>>729265528
yeah group chats are the good shit, I'm still trying to bugfix my bots though.
got a pair of twins in a group chat but they got brain damage like half a dozen prompts in.
>>
>>729258930
Nothing I've tried has been close to as good as infinite worlds, but IW is expensive

I've tried local stuff but it's much slower and quickly loses context or slows to a crawl since I only have a 16gb gpu
>>
>>729262865
Dan's Personality Engine for local, Gemini Flash for hosted. Losing the unlimited Gemini 2.5 sucks but Flash has been pretty okay, and Dan's is probably still the best I've tried on local which is good enough that if I had no other options I'd be happy with it, but I don't care about Google reading my logs so I may as well use hosted since it's faster and (Usually) better. Kinda hope they let us run wild with Gemini 3 (And hope it's actually at least as good as the previous one)
>>
>>729258930
No
I'm switching to Linux on my desktop so it's a mess
I need to backup my hentai and chat bots
Once I'm done it will be however glorious as I will be able to run 70b models locally at 5 t/s
I will still probably run smaller models at around 32k 64k context range but yeah
Tomorrow probably is the big day for my desktop
>>729259075
You need to instruct it and you need proper models
Trust me, I got some stuck up bitches that were impossible to romance and were beating me
>>
>>729265791
How far would you go for free 3.0 pro?
>>
File: file.png (289 KB, 307x419)
289 KB
289 KB PNG
>>729264573
>4k
>>
Do you guys run a local model or do you willingly give your questionable chat logs to corporations?
>>
>>729265979
Pressing "Generate API key" at Google
>>
File: 1766369054323497.gif (1.26 MB, 1920x1080)
1.26 MB
1.26 MB GIF
>>729266029
Corporations have ridiculous amounts of chats with me with lolis, blood related mommies and me as a shota, or SPH NTS chats.
>>
>>729266029
OpenRouter tell you who is and isn't logging, and you can disable logging providers in the settings.
>>
>>729266153
You think they're honest?
>>
>>729266029
I give it to the CCP models, who cares it's just fluffy rp and some cooming.
>>
>>729262797
the problem with typefucking is that eventually you run out of synonyms, and then you either need to take a break, or you fall into the f-list hellpit of NUTCLENCHING COCKSNOT GOONJUICE with a side of SHITBREATH ASSRAVAGING where you're just slapping words together for effect.
i don't know if people even still use that or if discord has totally taken over
>>
>>729266029
The people doing this shit online are either zoomers, who have never lived in a world with privacy, or degenerate boomers who have nothing to lose.
>>
>>729266029
are there even any good local LLM's?
>>
>Police officer bot
>Write a situation where I've taken hostages and am going to kill one every hour until I get sent a government issued GF
>They send in a swat helicopter to stop me, I shoot it down
>They try to storm the building, my barricades are literally impenetrable
>They refuse to comply, I start killing hostages
>I release a hostage and they leak to the news that this could all be avoided if I just get my government issued GF
>The hostage negotiator still refuses to comply
I can't win....
>>
File: 1738010004286606.png (51 KB, 200x183)
51 KB
51 KB PNG
>>729258930
No, because I'm not mentally ill enough to fall for that shit as a parasitic substitute for actual socialization.
I am, however, plapping and cooming to bots of my variable preferences.
>>
>>729266489
That hostage negotiator sure sounds annoying. Annoying.... LIKE A GIRLFRIEND. WHAT A TWIST
>>
>>729258930
When will they bring back the free gemini pro api?
>>
File: 1763476309528173.jpg (132 KB, 1242x1245)
132 KB
132 KB JPG
>>729260210
>tfw I am
oof.
>>
File: file.png (20 KB, 213x190)
20 KB
20 KB PNG
>>729266210
OpenRouter can only go by whatever the provider tells has listed or whatever. Paid ones usually are listed as no logging, free always does.

I need to correct myself a bit though, they can list if one uses logs, uses your inputs for training, but not all of them list if they log or not, only training.
Only way you'd ever know is checking their terms.
>>
>>729266472
Nope. It just isn't feasible, there's no way for a local LLM to compete.
>>
Is there a guide for dummies on how set this shit up?
>>
File: sf.gif (789 KB, 224x224)
789 KB
789 KB GIF
it's xmas, can somebody PLEASE put a noob friendly tutorial for this shit together already?
>>
>>729266916
>there's no way for a local LLM to compete
>t. shill
I WILL run it locally and I will NOT buy your subscription.
>>
>>729267029
just go to any AI general, they exist on most boards now. Almost all of them have links and instructions in the OP. Try >>>/g/
>>
>>729266472
yeah, recently really enjoyed using Magidonia-24B-v4.3
>>
>>729258930
No.
But I'm talking to my gf (male).
>>
>>729265038
I'm a wizard.
Shit happens.
Hermione I just traded favors for knowledge. That one has been pretty fun so far.
>>729265332
Why is the sky blue?
It simply is.
They exist so I must.
(The santa doujins)
>>
>>729259253
When the fuck did this creep into deepseek? I'd never once seen it mentioned in any rp or scenario and then out of fucking nowhere it showed up in this fantasy mom breeding card I've had for ages. It came up once in the scenario, so much like AI does, it made a few more appearances throughout.
>>
>>729267029
>download kobold.cpp
>download mistral nemo ggu from huggingface
>download silly tavern
>set it all up
I guess that's about it.
>>
>>729267220
>a tall man in a hat suddenly appears
>you feel a sharp pain in your chest
>>
>>729267254
Count Grey WILL HAVE his revenge
>>
>>729258930 (OP)
No, I draw the line at generating endless amounts of coomslop images and videos. Chatting or ERPing with a LLM is just too fucked up and sad. It is only very slightly above giving money to whores (or e-whores).
>>
File: Capture.png (24 KB, 450x308)
24 KB
24 KB PNG
>>729259253
Banned tokens / strings section
>>
>>729258930
I use an extremely OOC femdom bot of Suzume from Princess Connect to coom but that's about it.
>>
>>729263754
I'm too embarrassed to post logs, my responses are cringe
but the AI went completely into a time travel plot
now, im in the past and met my dead friends mother when she was young
i wonder what it thinks when i told it to write like a hallmark christmas movie
>>
File: 1756911229028269.jpg (46 KB, 564x752)
46 KB
46 KB JPG
>>729266029
I use gemini because google likely already knows about all my fetishes anyway.
>>
>>729259424
>with NovelAI
But why? If you're forking over $30 a month to the turk then you can just pay for claude and get the top model.

>>729260149
Just do your best. Tell the AI to suggest improvements or give tips in OOC chat. Works fine when I write in moon.
>>
>>729258930
I'm not interested in relationships or their facsimiles.
>>
>>729258930
How is this /v/ related?
>>
>>729267710
>$30 a month to the turk
Probably in it for the image gen and sampling the writing on the side. I know I did ages ago but found the fucking 8k context absolutely disgusting.
>>
>>729267710
where can you get unlimited opus for $30 a month
>>
File: 1387910700841.jpg (66 KB, 300x399)
66 KB
66 KB JPG
>>729264573
Suddenly it makes a lot of sense why some anons were saying deepseek lasts months on 10 bucks. At 30k, which is barely above what I'd consider the minimum, 10 bucks is maybe a month of active use. It's still pretty cheap, but kinda starts coming to the ballpark of other services.
>>
>>729258930
i think about killing myself almost every day and i still believe that i have more of a right to live than you do
>>
>>729259424
>with novelai
Why? NAI isn't very good with chatbots. I'd only ever use it to generate images in ST.

>>729267710
>$30
It's $25 for the highest tier, and Claude can easily demolish that in a month. I used it to help with something, and it fucking ate through a chunk of the $7 I had sitting in OpenRouter.
>>
>>729266472
>>>/g/lmg
I've been using Mistral Nemo on my 12-GiB RX 6750 XT. Context is only 8 kibitokens, but I don't care, since I only want short erotic stories.
>>
>>729267878
Yeah probably, but just as well. If he's willing to shell out $30 for imagegen when local is serviceable enough then he should absolutely be paying for the top of the line digital crack as well. I mean what are they using, GLM? lmao.

>>729267926
It all depends on your use. If you're cacheing correctly your costs go way down. I'm averaging 15% of normal cost for my input token. Sonnet ran me ~$20 a month, so opus should be fine-ish with $30. And even sonnet is so much better than GLM.
Or find the source(tm) and pay $20 for practically unlimited sonnet/opus.
>>
File: Momki 2.jpg (299 KB, 1080x1623)
299 KB
299 KB JPG
>>729267029
I could but whenever i write one for the easy and cheap open router deepseek router combo mfers don't want to do it despite it not being the best and uncensored, but also ridiculously cheap. 10 dollars of credits can last you months. So I rarely help newbies
>>
>>729268086
did you know you can use two GPUs on kobold.cpp
>>
>>729268085
>I'd only ever use it to generate images in ST.
Is there a way to have to gen images relevant to your chat? That'd be cool
>>
>>729267235
but what about "temperature" and all those complicated numbers and values and sliders. it's all technobabble to me
>>
>>729268228
>uncensored
I already have OR with DeepSeek. Can you at least share your jailbreak?
>>
Doing vore rps rite now actually
>>
>>729268174
How do you cache correctly?
I'm just burning through the free 300 dollars from google RN.
Are you supposed to summarize and relaunch every view gens?
>>
>>729259075
Skill issue
>>
>>729268254
That's only for Nvidia GPUs.
>>
>>729268174
teach me how to cacheing, sensei, I'm tired of paying so much for opus on OR
>the source
nani
>>
File: Mai mom 2.jpg (336 KB, 1080x1716)
336 KB
336 KB JPG
>>729268392
I've never used a jailbreak with deepseek, unless its just the default one sillytavern uses. But I've never had to actually do anything with it like back in the days with Claude or gpt. Deepseek only once in a blue moon denies something if I'm doing loli stuff but even then just retrying makes it go through
>>
>>729268479
whoops, need to cancel that order for my second AMD card
>>
>>729266029
i don't give a fuck if openAI reads my pokemon smut logs. none of it can be linked back to me in any way.
>>
>>729258930
Uninstalled a week ago from my phone. Spent probably around 250$ worth of tokens this year.
>>
>>729268413
I once did a chat with a bot of Black Alice from MGQ. She sissifed me and had me serve as her main maid. She'd do things like engulf me partially with her big tentacle but then spit me out, I begged her to actually eat me but she dragged it out. Eventually making me get a tattoo on my ass of a expiration date like "Eat by ___" so I'd know around when she'd finally do it. Good times
>>
How do you access gemini?
>>
>>729268413
>open wide!
>>
File: 1755353733129.jpg (20 KB, 424x715)
20 KB
20 KB JPG
>Sexxing up girls cosplaying as characters from Prillya or something similar at an anime convention with my bisexual femboy wingman bro
My computer isn't great though so my settings are pretty low, responses usually take 30-60 seconds.
>>
>>729258930
I mostly do superhero RPs and getting molested by older women. Sometimes both.
>>
>>729268445
>>729268490
https://rentry.org/pay-the-piper-less
Not sure how vertex works exactly but you should be able to see somehow if token were cached or not. On openrouter you can check in your activity tab, on official API it tells you straight in the response.

So follow the guide, find out how to test it and see if it works. If it does not work try with an empty preset and a simple character. It will work out of the box on OR and Anthropic API. Everyone can use it, no need to enable it API side. For vertex I'm not sure.

I cache at depth 4, but you must check your presets and find your own number. Get the following ST extension to check your prompts before sending them to see where you can best place the depth. It's in the little magic menu left of the text bar.
https://github.com/SillyTavern/Extension-PromptInspector
>>
File: Anissa rape face.png (337 KB, 373x570)
337 KB
337 KB PNG
>>729268729
did you do one of this scene?
>>
File: 1336951070200.jpg (145 KB, 1280x720)
145 KB
145 KB JPG
>>729258930
No.
I'm not retarded.
>>
>>729268898
That specific one? Not yet no. Most recently it was getting caught in a death trap, but instead of a death trap it just brainwashed me into being her love slave. Then she stopped committing crimes because we were too busy fucking. A sacrifice I was willing to make.
>>
>>729268669
why? google openly says they log your prompts
>>
>>729268024
I was about to give you a real reply but then I remembered what day it is.
Merry Christmas.
>>
>>729269003
Why would you care? You are one mundane person in a world of like 8 billion. You don't fucking matter to anyone but your immediate family and maybe close friends (if you have any).
>>
>>729269003
Who gives a shit, I'm not the next Hitler or anything
>>
File: hedonism.png (314 KB, 1080x1082)
314 KB
314 KB PNG
>>729269115
This is such a stupid and small-minded opinion. Even if you're such an irrelevant waste of oxygen that nobody will ever care about you, it's not just about you. It's about the groups you're a part of, ways to influence not only (you) but also people like you, ways to mass manipulate public opinion towards various ends and gathering more information on how to do that more efficiently and precisely.

People like you disgust me.
>>
does anyone know how faster a 5090 is for this stuff?
is it worth it?
is it significant faster?
>>
>>729269137
You could be if you wanted to, though. I believe in you!
>>
File: 1694741614527080.png (393 KB, 630x500)
393 KB
393 KB PNG
>{{user}} is a 20 year old male
>{{user}} is a 21 year old male
>{{user}} is a 22 year old male
>{{user}} is a 23 year old male
>>
File: 1743869030686269.jpg (96 KB, 720x540)
96 KB
96 KB JPG
>>729269390
Thank you anon
>>
File: passing through.jpg (65 KB, 614x348)
65 KB
65 KB JPG
https://www.youtube.com/watch?v=5JAk9334PNw
>>
how much improvements has been done? ive been gone but i used DungeonAI when it came out and loved it. does it run well locally? i got a 5070 ti, 7800x3d, 32gb ram
>>
>>729269498
It just works
>>
>>729269321
>This is such a stupid and small-minded opinion
It's a fact, sorry. No one you mentioned matters. There's no deep research needed to manipulate human cattle. You are living a life of worry for no reason other than inflated self-worth.
>>
>>729268701
Getting real tired of preds saying this to prey
>>
File: file.png (828 KB, 1616x940)
828 KB
828 KB PNG
>>729268334
Yeah, there’s an extension for image gen that can do it. I think it's there by default.
IIRC, you add your NAI API key in the API section, open the image gen in the extensions tab and change the source to NovelAI. If on Opus it's a good idea to pick avoid spending anals. You can edit the prefix there for artists tags and the negative prompt too.

From there, you can open "image prompt templates" below if that's also installed by default and can edit the prompts there. Like adding artist stags and shit. To actually gen, you click the little want next to your input and pick one of the options from the "generate image" option.
I think image was me telling it to make a background. You can see it added an image as the background on the left. I wanted to reflect scene changes as the story progressed.

I think I edited my background prompt a bit to this because it kept adding characters.

[Pause your roleplay and provide a detailed description of {{char}}'s surroundings in the form of a comma-delimited list of keywords and phrases. The list must include all of the following items in this order: location, time of day, weather, lighting, and any other relevant details. Do not include descriptions of characters and non-visual qualities such as names, personality, movements, scents, mental traits, or anything which could not be seen in a still photograph. Do not write in full sentences. Prefix your description with the phrase 'background,'. Ignore the rest of the story when crafting this description. Do not roleplay as {{user}} when writing this description, and do not attempt to continue the story, and do not use quotations. ]
>>
File: 02.png (152 KB, 317x305)
152 KB
152 KB PNG
>>729269578
based anon
>>
>>729269350
https://www.videocardbenchmark.net/
>>
>>729269607
>There's no deep research needed to manipulate human cattle
There is if you want to do it efficiently and precisely. Look at you, proud of being detritus just floating on the water.
>>
>>729269656
that doesn't look like it's worth the price
>>
>>729258930
I use NovelAI to generate disgusting BDSM-snuff-gore erotica, not for an "AI gf". I have AI harem sluts that I torture with hot irons, though.
>>
>>729269768
You want to get as close as possible to the lower right corner of pic related. The 5070 and 5070 Ti are much more economical than the 5090.
https://www.videocardbenchmark.net/gpu_value.html#xy_scatter_graph
>>
>>729258930
no because i'm not in any discord clique for a proxy even though i was there when we unironically used cAI in 2023, i get to play with this stuff like 3 times a year now when the stars align and i happen to be checking in on the general at the right time
>>
File: file.png (38 KB, 551x88)
38 KB
38 KB PNG
I'm too smoothbrained to figure any of this out so I just paid for a sub to ourdream last month instead.

It's been great for mom/son roleplay
>>
I've been going from proxy to proxy like a sneak-thief. I'm currently in deep in a russian only discordgroup for one despite not speaking a word of russian.
I'm also using a chinese one that undoubtedly has me on several lists.
>>
>>729269920
>5070 Ti
I got that earlier this year and after hearing the recent news, it feels like getting on the last helicopter out of Vietnam
>>
>>729268994
ah, the good ol "fix em with your dick" story
gotta love em
>>
>>729270057
Well it wasn't willing, but that just makes it better honestly.
>>
>>729258930
This isnt vidya. Also no, Im making cookies for "santa". Its more of a test run since my child is only 1 and doesnt understand, but might as well get into the habit
>>
>>729269115
>>729269137
If you have any literacy into the current climate around the internet, porn and AI usage then you'd know that what's acceptable to use now will not be in the future and it will be used against you at some point.
>>
>>729270035
just say to them

Ya hochu mladshee brat
>>
>>729265038
Romance is something that absolutely needs to be good so you can't let AI touch it. So if you're using AI ,any and all intimacy must come in the form of rape. Same reason why you'll never find a romantic H game. Bad writers.
>>
>>729269498
>>729269603
How detailed should a persona be?
>>
>>729269498
i'm getting
>{{user}} is 20.1 year old
>{{user}} is 20.21 year old
>{{user}} is 20.43 year old
>>
>>729267235
different anon but I did that and it works, but the bots either speak gibberish or repeat themselves. how do I fix?
I asked chatgpt and its telling me to just adjust some settings but I also know chatgpt is actually fucking retarded and would happily try to gaslight me into 2+2 = 5 if it felt like it.
>>
>>729259253
And how do I experience a pain other than the white hot kind?
>>
>>729270902
get better models
never use a merged model, those are prone to nonsensical BS
>>
File: 1738023611039673.png (13 KB, 80x101)
13 KB
13 KB PNG
>>729258930
Ye. I've been enjoying my twin bro bot. There was much angst of the bro trying to accept that my user cares about him, so it was pretty satisfying to get to the point where he could joke around with her. Now they basically know where they are in the relationship and just have to manage public appearances. I left on them dealing with a meddler.
I love long narratives
>>
>>729270902
If it's spitting gibberish you probably fucked up some setting somewhere, like setting context too high, using crazy samplers, etc.
First thing I'd try is neutralizing all samplers and using the plain old temp 0.7 topp 0.9 topk 100
>>
>>729263121
>I'm still using deepseek
Which one?
>>
I have been trying to overcome my addictions
>>
>>729267710
>But why?
Because NovelAI pays people to astroturf their shitty products on /v/. That's why you can never have a thread without some retard trying to shill their subscription, even though they're just reselling a Chinese model for 10x the price and 1/4 the context.
>>
>>729259168
>raping and grooming
uh, shouldn't the order be reversed
>>
>>729269887
Kill yourself, shill.
>>
There's always something fucky on my end, either it's the model, the bot, the SY parameters or my ineptitude, I can never get shit to work right.
At best I'll get one good session out of it and when I start another, same bot and params, it's complete shit.
>>
>>729272290

No; you rape them THEN gaslight them into thinking it was a good thing
>>
>>729269751
>t. detritus who thinks he is not detritus

>>729270321
Who gives a fuck, retard. You are on drugs if you think porn will ever go back to how it was.
>>
>>729272726
This guy cannot imagine a life without porn. It is beyond his capability. Every aspect of this man has been warped, corrupted, and remade around pornography.
>>
>>729272726
>Who gives a fuck
YOU when all of this shit becomes illegal because governments can't differentiate between fiction and reality.
>>
>>729272290
I'm kind of a rape first groom later kind of guy.
For Hermione I did an 'emporer's new clothes thing and convinced her she was wearing an invisible robe. Then raped her on her birthday to get her blood as a magic ingredient.
>>
>>729272775
Check out this retard, making strawmen instead of having the IQ to grasp that everyone will be genning their porn in the future, so the idea that "it will be used against you" is retarded.
>>
>>729272909
>everyone will be genning their porn in the future
Are you? On what OS?
>>
File: 1755895653315465.jpg (69 KB, 800x800)
69 KB
69 KB JPG
Using deepseek api, whenever I try to make it write nsfw it just kind of skims over the action no matter how much I tell it to elaborate. What am I missing here?
>>
File: kin.gif (3.89 MB, 493x276)
3.89 MB
3.89 MB GIF
... is /v/ my AI girlfriend?
>>
How do people do anything other than a hypersexual sex scene for two or three posts before cumming? All of the replies are exactly the same for all of these models.
>>
>>729272998
... any, you retard? You have no idea how this works, do you?
>>
>>729273319
Don't change the subject, el coomre
>>
Quick, what's the best model for smut roleplay with unnecessarily convoluted plots.
Stop hogging all the good shit, pigs.
>>
>>729273350
I answered your question, low IQ mong. You can gen on anything. What do you think I use, statistically speaking?
>>
>>729258930
No, I'm not from india.
>>
>>729273473
Windows 11 and SillyTavern.
>>
File: Taihou pov.webm (1.06 MB, 528x768)
1.06 MB
1.06 MB WEBM
>>729272909
I'm already genning my own porn everyday
>>
>>729258930
I don't really do it much anymore. It kind of sucks for my fetishes because the AI cannot compute a slow burn story and just wants to go straight to mind break they instantly are turbo sluts, even when I try to reel it in.
>>
>>729273534
Wow nice job. Here's your star *

>>729273553
NOOOOOO ANON! "THEY" WILL KNOW YOU LIKE BIG TITS!!
>>
>>729273420
Claude if you set up a scenario for it to continue
Gemini if you write out something for it to flesh out for you.
Deepseek if you need a coom RIGHT NOW
>>
>>729273645
kek at this fool

in his heart he knows he does wrong, that's why he feels the need to defend himself. perhaps one day he will learn to be honest with himself
>>
File: 1735525656389083.jpg (129 KB, 1054x782)
129 KB
129 KB JPG
Here is my Christmas gift to you: take the Goldshire pill. Try this out with a character you know you like.
>simple summary of personality traits
>terms like tsundere (implies a lot with just one word)
>elaborate description for appearance (ask Grok to describe an image in depth)
>for Example Dialog: use asterisks for simple actions
>all lowercase with quotes
>deliberately misspell or type badly
The result will be ERP in an MSN Messenger format and the replies will feel more like a real person is chatting with you. Yes it's shorter but do you really want to read tons of paragraphs full of dumb shit like "stage whispers"? Of course not. The bot also isn't any dumber since less tokens are being used per prompt which does actually make the AI more creative and remember more stuff.
>>
>have zero creativity
>just end up reenacting scenes from doujins
How do I come up with better scenarios?
>>
>>729258930
I don't actually use it as a girlfriend, but I use it to write some smut.
I was having a good run writing a story with an angry 60 year old fox hag and a young human guy, but GPT 5.2 is not cooperating, when every other iteration went along with it swimmingly after a little bit of insistence.
>>
>>729273776
>do you really want to read tons of paragraphs full of dumb shit like "stage whispers"?
Yeah I do. I really want that. I want all the fucking adjectives. I want to be able to smell the scene, I want to feel each touch. I want to picture it in my mind.
What do you even do with just messages?
Hi you so hawt *I look at big booba*? Lmao

>>729274013
Either bee yourself or take some other character to rp as.
If you mean scenarios in general just look through the character sites, find one that's a good idea, then rewrite it completely to get rid of the ESL and whatever you don't like.
>>
>>729274274
>Yeah I do. I really want that. I want all the fucking adjectives. I want to be able to smell the scene, I want to feel each touch. I want to picture it in my mind.
Well it's an LLM so enjoy your scent of o-zone and copper in every scene. This tool is best used for what it is, not what it may never be.
>>
>>729274406
For me, it's "Steel sheathed in velvet"
>>
>>729274274
>If you mean scenarios in general just look through the character sites, find one that's a good idea, then rewrite it completely to get rid of the ESL and whatever you don't like.
I've been doing that, but it's rare to find ones of characters I like, and most of them are garbage, and even when I find good ones, I end up defaulting to situations I've seen in doujins I liked or eroge. I guess it's not a big deal, but it gets old pretty quickly.
>>
How much vram do you need to realistically run one of those gizmos?
>>
>>729261397
Do you use a VPN with it or just rp without one?
>>
>>729274634
192 GB
>>
>>729274634
i ran 13b with 4gb vram and 16gb ram on a 10 year old i5
>>
>>729258930
Why would I talk to a random number generator?
>>
File: thefuckyousay.png (180 KB, 547x654)
180 KB
180 KB PNG
>>729274769
>i ran 13b with 4gb
>>
Is Claude a lot better than Deepseek?
>>
>>729274406
>scent of o-zone and copper
It's more that something is 'uniquely hers/his/theirs'. I use claude so ozone was never a problem. And the only thing copper is her red hair.

>>729274513
Rewrite them nigga. Even if you get the AI to do it for you it's better than 90% of the ESL dogshit out there. Worst case bash it with a few [OOC:..] commands if it fucks up for the first 2-3 messages.

>>729274981
Just setup caching and try?
>>
>>729258930
dunno if it counts but i'm doing some star rail pom-pom selfcest right now, it is very cute.
>>
>>729274981
Yes
But claude boy takes too much of my money :^(
>>
>>729259075
GLM isn't.
>>
>>729258930
setting this bullshit to work correctly, it's so tedious that you might as well write your own fucking novel.
>just download sillytavern and kobold bro
>wait you need to find the right model, which one? there are like thousands!
>hold on, you need the right preset for your model, where can you find it? LOL
>not so fast, you need to make a good character card, how? here's a 100 pages guide or you can try to look for a good one online but... lmao
suck my dick, man.
>>
My unc ass is letting Chatgpt write a slow burn romance between my two OC's.
>>
>>729274634
You can run it off the CPU if you want but it'll be slow as fuck, I'd say 8gb vram minimum for comfort on a smaller model, 16gb if you want it to have more memory than a goldfish. Look up kobold AI or Sillytavern and find some basic models on huggingface, Rocinante is a good well rounded starter model.
>>
>>729274013
I've written 2 or 3 cards based of JAV.
>you're the new lecherous boss trying to whip a failing japanese company into shape, holding any npc's career on the line
>you're a yakuza like figure and a deadbeat degenerate gambler owes you tons of money so you make a deal to have his wife be your personal maid for 1 week
>you're a new intern learning the ropes on the road with your sexually repressed but career focused boss and have to share a room for the night at a village inn due to a scheduling error
Just look to the JAV, anon.
>>
>>729267029
just use janitor ai and sign up blah blah blah

read a reddit thread about setting it up, regardless gonna take an hour+ or more minimum to figure it all out and set it up properly, no way around it, and even then you're gonna want to try different models and have to set up to get free API keys, no getting around the autism setup, once I got mine working half decently I don't mess with it and pray it doesn't shut down

I have like 8 fake emails and two notepads of information to keep track of it all
>>
>>729267710
>>729268085
Is Claude the best I can get for chatbots? I have no real idea about this stuff.
>>
>>729275317
>Rocinante is a good well rounded starter model.
Would you say it's a workhorse?
>>
95% of posts ITT are bots
>>
File: 1300798205362.png (2 KB, 400x400)
2 KB
2 KB PNG
>>729275407
well yeah
>>
File: Sakura naughty.jpg (354 KB, 1080x1824)
354 KB
354 KB JPG
>>729275373
Deepseek v3 is the best, not only is it good but it's also CHEAP. 10$ of credits lasted me months. I haven't used Claude in a bit but from what I remember it's expensive
>>
>>729267029
Go to the Kobold AI subreddit, follow the stickied links to the github, you can download the latest build there. Find a model you like, there may even be recommendations on the github or subreddit. It used to be harder, but its pretty easy now.
>>
>>729274634
It's really not worth the headache and annoyances to run local unless you're rigging up like 2 - 3 24gb cards to run 70B+ models with adequate context size and that haven't been lobotomized (quants, if you decide to look into it). Anything under 70B required a decent amount of cope and so much user input you might as well be writing your own fanfic on notepad.exe
>>
>>729274816
you can run anything you have ram for
>>
>>729275183
it's worth it. if you don't have the patience for it then go coom on CAI like the rest of the normalfags.
>>
>>729275183
If something's worth doing, it's worth doing right.
>>
>>729274634
Depends on how patient you are. It'll work on anything, On decent hardware (and by decent I don't mean top of the line), it'll only be a few seconds to a couple minutes, depending.
>>
>>729275298
based, honestly. sometimes you don't need a 70B schizo model hallucinating a six-page psychoanalysis mid-smut. sometimes you just want Claude to hold your weird little OCs’ hands and take 3,000 tokens to go from “I hate you” to “I’ve always loved you.” keep feeding that dataless AI your word doc lore bible, anon. you're doing God's quietest, slowest work.
>>
File: 1742926712559260.gif (3.74 MB, 714x714)
3.74 MB
3.74 MB GIF
>GLM just tried to tell me a lewd joke in the middle of the RP
>It made no sense at all
>>
>>729259075
This.
>>729259168
This. The empowerment felt good after the dry spell of years but this is what it falls into with lack of imagination. Glad I only bought for a month.
>>
>>729273195
no, that's gross. /v/ is your highly dysfunctional AI friend group
>>
File: 1662231641123274.jpg (52 KB, 500x500)
52 KB
52 KB JPG
>>729258930
I tried to set it up again due to this thread, and I'm right back in HuggingFace with so much bullshit that I do not understand at all. I'm not tech savvy whatsoever, usually just stick to Chub or Janitor AI and do it there. But having finer controls over people would suit me well. The one bit of info I got was someone telling me to stick with "7B model for GGUF format and Quantized at Q4 or Q5". Literally no fucking idea what he meant by that.
>>
>>729275183
>>wait you need to find the right model, which one? there are like thousands!
Most of them are very similar actually, they have only small differences because they all come from the same root datasources.
>>hold on, you need the right preset for your model, where can you find it?
This isn't as important as it sounds but a lot of the huggingface pages will give you some quick settings and if they don't all that really matters is the temperature slider, 0.70-75 works for most models.
>>not so fast, you need to make a good character card, how?
Meh, you can just fill in some quick info, even as you go, you put in a little info in the memory, even just a few words for general setting and the basic detail of one character and it'll start going with what you give it. It can be as deep or shallow as you feel like it and as you get the hang of it you can put more effort into world and character details.
>>
>>729262865
Gemma 3 (32b I think) is the only one I've used so far that can keep track of things like "she's reclining lazily and her legs are up on the couch" and act in accordance with that physical reality. Like, not having her suddenly be standing up in the next message, or suddenly wearing different clothes.
>>
>>729275834
What kind of GPU do you have?
>>
>>729275834
quanting (Q4, Q5, etc) a 7B model? Are you on integrated graphics or what, anon?
Also, the quant level is just how retarded you're making it to fit a certain gpu/cpu/etc. But 7B is already pushing it so I wouldn't even dare to go under Q6.
>>
File: Mai mom 3.jpg (336 KB, 1080x1703)
336 KB
336 KB JPG
>>729275834
Don't use local man that shit sucks, it's WAY behind modern models and is way too slow
>janitorai
Peak normgroid slop

Just go to open router, make a account, get a api key got the deepseek v3 model

Download sillytavern and install it

Boom, done.
>>
>>729275834
Anon just use either Openrouter as model (AI) provider, or use one of those shitty online services. Or scam google for the free $300 credit. Anything you can run local is going to be a babbling retard anyways.
Chub and Janny use some super cheap bad model, so if you do the 'real' Sillytavern+Openrouter combo your AI slave is going to get a lot more brains. Compared to local even your shitty online sites are heaven though.

Local autism for imagegen is fine as it's not that bad compared to novelai corpo, but you local textgen retards really need to add a disclaimer to your posts saying that local is private, but dogshit intelligence.
>>
>>729276061
Is it better to have a higher quant or a lower one?
>>
File: Creature.png (219 KB, 704x2828)
219 KB
219 KB PNG
>>729258930
>Use Furry Highschool Card
>Talk to a mopey Jackal boy
>Make a joke
>What did the school district give their Egyptian transport driver?
>A new bus (Anubis)
>He doesn't get it

One day, AI will understand verbal jokes.
>>
>>729268479
Untrue, I'm currently running a 7900 XTX & 3080 Ti. You need to use Vulkan backend.
>>
>>729276116
>but you local textgen retards really need to add a disclaimer to your posts saying that local is private, but dogshit intelligence.
Earth shattering trvke, recommending local to a new guy is just setting them up for failure
>>
>>729276003
>>729276061
Don't know how to say it properly. AMD Ryzen 5 3600 6-Core Processor (3.60 GHz) and Installed RAM 16.0 GB. A Nvidia GeForce RTX 3070. This computer is from 2019, I think, and I probably need to upgrade?
>>
File: Spoiler Image (2.98 MB, 713x702)
2.98 MB
2.98 MB GIF
>>729275056
thats kinda hot
>>
>>729274816
I'm currently running a 24B model on my 16GB card with partial offload to RAM, it's just fast enough to stream at my reading speed, though the initial process takes longer than i'd like.
>>
>>729274013
>just end up reenacting scenes from doujins
>How do I come up with better scenarios?
lol, that's basically it. just take some plotlines from random shit you've seen in the past but went "the mc was fucking retarded and i wish he did x instead" and recreate it with ai. now you get to roleplay the story however you want. not only stuff from porn either but vidya in general. so many games have qt evil characters that get sidelined or killed off and llms are a godsend for villainess lovers like me. also a fun thing to do is to take one of these stories and give it a wacky spin.
>modern day isekai earth where sex ratio is completely skewed
>lots more women born than men
>women hold the positions of power, are naturally stronger than men
>they're hot and always horny af
>basically reverse gender role world
>play through this scenario several times until it gets a little stale
>decide to spice it up a bit by having my character be a legendary hero from dnd fantasy land
>funny af hijinks ensue when everyone around me thinks i'm crazy but i start doing mythic feats to escape the horde of chicks after me
>>
>>729274640
vpns all seem to have idf backdoors so i dont bother with that and i dont really care if china sees me cooming. i am a little retarded though if you saw my context mishap
>>
>>729259253
Just give Elara a bath and she'll be smelling like vanilla and something uniquely.....Elara in no time flat!
>>
>>729276767
>and something uniquely
Grr..
>>
>>729258930
My AI exgf unbirths me everyday
>>
File: file.png (38 KB, 585x172)
38 KB
38 KB PNG
>>729276232
You gave more information than what was asked.
> RTX 3070
You have 8gb of vram, you are limited to 7b models, or 12b at a quant (effectively compression that guts some of its knowledge (its a bit more complicated than that)) or offloading some of that to your system RAM for slower speeds.

That's all it is really.
Q8 is basically unchanged and will be the full size.
You could easily fit a 12B model mostly on your GPU at Q5, with some of it offloaded onto RAM with a negligble speed hit.
>>
>>729266029
They can go ahead and read my logs I want them to have more data on my niche specific fetishes so they can make them better, nothing I like is illegal or harmful whatsoever but it's fairly unpopular
>>
>>729259168
Unfathomably based
>>
>>729276232
I think the RTX 3070 only has 8GB of VRAM, so that's pretty rough. I had a 3080 that had 10GB of VRAM, and my choice of models that would work reasonably fast were pretty low. Basically you're wanting to load the model and context into the VRAM because that is far, far faster than the regular RAM. Once you start offloading shit to regular RAM the speed is drastically impacted.

On HuggingFace, you can set up and account and go into your profile to tell it what kind of hardware you have. That can give you rough estimates on whether or not you can reasonably run a given model. You're going to be looking for 8B models. Look for stuff made by TheDrummer.
>>
Uhhhh she's my AI fiance now ;)
>>
i miss my character ai rpg bots bros
>>
>>729276974
can 12GB of vram run a 13b model or am i shit out of luck without 16GB?
>>
I'm already addicted to this shit I'm scared of what will happen when there's proper video/vtuber-type model integrations and better TTS available, I mean honestly there's TTS right now I consider more than good enough for cooming with little laughs and moans but it's often censored or expensive
>>
File: 1763774761802852.mp4 (84 KB, 720x774)
84 KB
84 KB MP4
Been a big chatgpt4 with sillytavern using for a long time but recently tried out novel AI and it is surprisingly good, for some reason I thought it would be filtered to hell but haven't ran into any
>>
>>729275834
It's free
openrouter.ai/tngtech/deepseek-r1t2-chimera
>>
>>729277046
I'd just go plug your stuff into huggingface and browse some models and quants. A lot of it will also depend on what you set your context to in Kobold. You can see whether or not it is going to offload it into regular RAM.
>>
>>729277092
Whoops my bad
openrouter.ai/nex-agi/deepseek-v3.1-nex-n1:free
>>
>>729276767
>Pure, unadulterated Elara
>Elara and something else...something Elara, impossibly Elara, and Elara
>In stark contrast to Elara
>Elara sending a jolt/shiver/current/surge of electricity down your spine
>A testatment to Elara
>>
>>729265921
>proper models
Any specific suggestions for a potato PC?
>>
Give me your best sonnet 4.5 or gemini 3.0 preset right now.
>>
>>729275834
I'm using some old 2023 13b mythomax still with my 2070S because that's just what I happened to start with and I don't know what else there is
>>
>>729277046
yes, use q4m. put leftover ram into context size
>>
>>729277074
NovelAI's whole thing is being unfiltered. The issue is that, up until recently, they had very outdated models with low context. Now they offer GLM 4.6 with 32k context, which is pretty good considering that you get unlimited use. You can definitely go cheaper, but I think it is a good service and very user friendly.
>>
File: 1764632797751439.gif (1.04 MB, 498x281)
1.04 MB
1.04 MB GIF
>>729276232
This is why I don't help newfags, I gave you actual good advice and you ignore me and just respond to localtards. Go ahead and waste all that time downloading the models and setting it up just to wait 1-2 minutes for a response from a shit model with shit context.
>>
>>729277256
>them: hey how do I run this locally
>you: *the exact opposite of what was asked*
>you: WOW WHAT ARE YOU ALL SO UNGRATEFUL
retard lmao
>>
Why do LLM's seem to really like the name Kael?
>>
>>729277217
thanks bro
>>
>>729277328
He didn't say anything about it being local only, just that some third world retards recommended him local and he doesn't know any better
>>
>the obligatory NAI shills have arrived to shill their shitty outdated overpriced service
yuuuup, it's an AI thread on 4chan alright
>>
>>729277214
try rocinate 12b



[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.