40 years later
The dating sim genre returned to its original form

>>715779765 (OP)
roberta williams got a nice set of tits in that pic

>>715779765 (OP)
Roberta sexo. Living on the Sierra campus back in the early 80's must have been some of the most fun shit ever. Bunch of fucking nerds living in the middle of the woods throwing wild parties every night, celebrities stopping by, all while producing hit vidya.

ST can be used for much more than just chatting.

>>715781458
As?

>>715781495
You can turn anything into a zorklike, with maps, npcs, followers, RPG mechanics, the only caveat is you have to sacrifice the chatbot.

Still funny watching jeets try their hardest to shill this garbage. ST is awful, I will always return to actual content written by actual people who know what eroticism is.

I wish LLMs could write entertaining shit, it'd be great. But pretending what we have now is in any way good enough is like pretending like VR is a revolution instead of a minor improvement over shit we already had in the 90s. You're just sullying the water.

>>715779765 (OP)
>>715781458
I find sillytavern's chat format very restrictive desu
it makes it hard to do anything besides talking to a few preplanned characters at a time
I much prefer koboldcpp's story and adventure formats

I SAID ROBERTA SEXO!
SEX WITH ROBERTA!
Someone should make a Roberta card.

>>715781763
There can be a compromise between chat and adventure formats (but mostly chat). Kobold's adventure mode is very nostalgic, though.

>>715781625
you're just so misinformed its insane, the top level LLM's or even just deepseek write insanely good erotica and the fact that it can be tailored to whatever you like + gives you full control over the experience with extremely deep levels of customization makes it revolutionary. but if you've never tasted the top models with the proper tuning and optimization and have only used like gpt 2 or something i can see why you'd think that.

>>715781763
you can just make a storywriter card that acts however you like, you dont have to use specific character cards its just good to use if you are into a character with very specific details that the model might not know

>>715779765 (OP)
is there an impregnation text adventure on this

>>715782094
sillytavern is a LLM front end, you punch in your API key for your model and then, yes, you could load a character card with that description, there are thousands of character cards and you can also just make your own

>>715782087
I know you can gaslight the chatbot into not acting like a chatbot, but at that point why are you even using a chatbot format?

>>715782193
you can switch it to text completion, or tweak its behavior to suit basically any form of interaction you like, silly is very customizable and its really just limited to how you want to use it

>>715782032
They're all garbage. I've tried them all, I've messed with presets, they produce low quality slop output. Maybe you can jerk off to it, I can't.

>>715782337
i dont believe you

>>715782337
>>715781625
it's cool tr@ns-bro, nobody is going to stop you from erping in yer discords

>>715781625
>>715782032
I'm no ERP fag but most chatbots have a attention span of a goldfish reading actual erotic fan fiction is far better.

>>715782519
I was in the middle of writing you an earnest reply, but I can see now there's no point. I hope you kill yourself, you faggot nigger. You are too stupid to live.

>>715782605
they have like 1 million context at this point? thats like the size of the fucking bible. context and memory issues stopped being a problem like 2 years ago ?? but thanks for making it clear you have no idea what you are talking about. now go read your AO3 slop while my goon bot genius curates the exact perfect fantasy with a 100 page slowburn with perfect memory!

>>715782609
Just don't cut yer nuts off.
Peace out.

>>715782609
Yes, after 3+ years of screeching about ai, NOW you were going to be "earnest."

>>715782751
Have you every actually tired using a deepseek or open AI's new models retard that shit forgets thing you said two messages ago it hardly has context as well.

>>715782839
no it doesnt you fucking retard, you can literally have a 1000 message chat with most of these modern bots and it will dynamically call back to your first message 1000 messages in, memory and context is a non issue holy fuck how inept are you

>>715782751
>they have like 1 million context at this point? thats like the size of the fucking bible.
No, it's not. While using reasoning/thinking models that's around 2 hours of erp, 2 paragraphs in each message. I run out of context space extremely quickly and need to either summarize the whole story and start a new chat or manually edit out all reasoning blocks. The same happens in Gemini 2.5, grok 4, latest deepseeks, all the finetunes etc. Once you ran out of memory space and context space the whole model falls apart.

>>715782751
it's much more than Bible, you can have multiple books inserted. the problem is the prose gets repetitive and annoying past let's say 300 messages. oh, and you have to wait for a long time to get a reply past 300k tokens.

>>715782904
What model are you using because this literally never happens and even if it calls back to something barley has any context.

>kluster is kill
NOOOOOOOOO MY SAFE DEEPSEEK FUNDS

>>715782831
>>715782756
> Not them but.
There's a difference between using pre-gen AI for chatbots vs making "games" with them since they fail to do be able to do any procedural generation etc that humans can just by using a set of dice or programming a chart that generates those things from a single button press.

>>715782991
He's using Gemini obviously. The thing is you get a quality drop after hundreds of messages.

It's good for cooms but if you want to do a whole ass long story then you're gonna have to use lorebooks and summarization. It's still fun with a little bit of wrangling if you have at least 20k context which is my sweet spot. This only applies to cloud (claude/oai/google user) idk how local is looking atm

>>715782751
>>715783276
Context is solved and lorebooks are the king of chatbotting. If you have some patience, you can do just about anything with them. Gemini has made it possible to adapt entire games you can play with SillyTavern.

>>715781763
Me too. Chat doesn't do much for me. I prefer collaborative story writing.

>>715783605
>Context is solved
factually false. I use heavily Gemini/Claude for my RPs (not ERP) and I have to remind them basic shit all the time even when I use lorebooks sometime.

I have to manually start another conversation to keep my plot going because every single model shits the bed after 500 messages.

Hard pass on the AI slop until you can get something half-decent locally on a reasonable computer, i.e., not a server rig of 6 $3,000 GPUs with an enterprise CPU, but that's not happening any time soon.

>>715781625
then go back to F-List and desperately try to ERP with fat old men over the internet you repulsive faggot

>>715784012
You can't get anything half-decent on the best server farms on the globe, what hope do you ever have of being able to run anything good locally?

>>715783276
>It's still fun with a little bit of wrangling if you have at least 20k context which is my sweet spot.
20k? I get 150k and I think it's too little.

>Gave elon one drunk night.
>have grok api
>cant use grok 4 on the default page
>try to use grok 4 for coding with sillytavern
>it fucking shits itself because there are 1000 billion small shits that go it a users response hidden all over sillytavern
>hace disable and enable everything if i want just default grok vs rp with character cards
Is there a console for the api that is basically the web page?

>>715784319
My current proxies have a 20k limit, + I've had a lot more context on the same models before and it feels like there's a "soft" context cap that's at around 20 - 25k before it gets iffy and it MIGHT remember things from really far back if you specifically mention a keyword and then swipe a few times. I've been told about this before on /g/, too bad the archives are down but the current popular models use some sort of shitty "extended context" technology past a certain number that's not too different from the one sillytavern has as an extension.

>>715784515
The best is to give it not super detailed expressions but use wordings that make sense they way you want it spelled out and it usually gets it unless it hallucinates shit forcibly for some reason.

>>715784515
I'm pretty sure that these people who bitch about having to start new RPs after such-and-such number of messages are people who are just maxing out context on a corpo model. That shit is doomed to failure. They would be better off limiting their context heavily while relying on lorebooks, author's note and summary to maintain important information.

>>715784813
What kind of RP are nigger even doing? I just plap the trapshota and turn it off.

>>715783605
I recommend this extension. Makes lorebook tism super easy and fast, you can customize the prompt however you want and it does literally everything including the tags and adding it to your lorebook.
https://github.com/bmen25124/SillyTavern-WorldInfo-Recommender/

>>715785423
I have this. It's handy for when you're playing an adventure and you want to save your impacts on the world or you just have lots of non-canon NPCs.

>>715779765 (OP)
i coomed so much to silly tavern bros

>>715782751
>>715783276
it's true, everything is all ad-hoc right now
no real devs will touch this stuff because of the degen fanbase

>>715785829
Good. Devs are total bitches usually:
>ooohhhh check out my fork where I've changed the name and just 10 lines of code, don't forget to join my discord
>links this project to his resume

I only use it for cooms
the only use for the 'technology'
And I'm half convinced it's 4th worlders erping with ne anyway

>>715786002
Sanji is jerking it to your sesh, he loves being your little 11-year-old blonde Kolbold sex slave.

>>715782996
Man, I was finally running out of those trial bucks they gave out at the start (100 bucks for free, lmao, how the fuck did they think that was going to be a good idea) and was actually seriously considering spending some money to refill because I've been using it a lot, but I'm glad I dodged that bullet.

>>715786081
I'm not a pedophile

>>715786297
neither is Sanji ;)

>>715779765 (OP)
The chick on the left's tits are small and flabby enough they’re floating and can almost see the nipples lol

>>715783910
You're not getting information correctly into the prompt. Your lorebooks or your cards are fucked up. I recommend scrolling up through the sillytavern cmd and examining your prompts that are being sent, provided you understand how a prompt should be constructed. If you're somehow losing control of things using Gemini 2.5, either flash or pro, then you're majorly fucking up with your prompting.

For local models, lorebooks are incredible for applying a characters thoughts, experiences, or abilities when something is relevant and only then. I also highly recommend vectorization for longer things such as past events by using summarization to compress a conversation into a large chunk, saving that large chunk to a .txt file, and then loading it into the data bank and enabling vectorization in the extensions menu.

Vectorization is just lorebooks but using another AI model to determine when the data in a huge document is relevant rather than tightly setting it up as a lorebook. Vectorization, combined with TTS, is how I play old RPGs that have ginormous-fucking-FAQ guides. Dumping the FAQ as a .txt, loading it into the data bank, and then using voice recognition and text-to-speech to talk with my "AI game-watcher" to figure out if this dialogue choice is good or bad for my chaos frame without having to pause and search manually. Setting this up has been a world changer for me and I can't imagine having to live without this ever again.

It is better than any porn.

Could someone please make card about shapeshifter who loves to rp with {{user}}?

>>715786630
Yeah, she got mogged by the girl in the middle and Roberta.

>>715787694
elaborate
ruminate upon your idea some more, and write more details, feel free to ramble, then give it to card creator bot :
https://chub.ai/characters/slaykyh/character-card-builder-8927c8a0

How do i make it look like pic related?

>>715787694
{{char}} is a shapeshifter who loves to roleplay with {{user}}. Regardless of the form they take in the moment, {{char}} always has a giant futa cock covered in smegma and dick cheese.
Wow, much hard. So difficult. I don't really get people who beg for such simple cards. You've already created it in just asking for what you want.

>>715788017
https://developer.mozilla.org/en-US/docs/Web/CSS

I haven’t used chatbots since April-May of 2023, is Opus still the best?
And OpenRouter is much more viable than jumping through hoops for proxies, right?

I was using backyard AI because it was 10x more streamlined and easy but now they have depreceated their desktop app for some reason?
What is the point of ST over just using Koboldcpp? I like netori chat where i get female characters to cheat with me.

>>715787452
>Vectorization is just lorebooks but using another AI model to determine when the data in a huge document is relevant rather than tightly setting it up as a lorebook. Vectorization, combined with TTS, is how I play old RPGs that have ginormous-fucking-FAQ guides. Dumping the FAQ as a .txt, loading it into the data bank, and then using voice recognition and text-to-speech to talk with my "AI game-watcher" to figure out if this dialogue choice is good or bad for my chaos frame without having to pause and search manually. Setting this up has been a world changer for me and I can't imagine having to live without this ever again.
This seems complicated.

>>715788425
Cheeky.

>>715788125
people do that because there is usually some aspie who are iching to do the work for them

>>715779765 (OP)
ST + Comfyui for image gen is game changing, one day I spent around 2 hrs fiddling with it and couldn't jerk off I had a boner the entire time my dick hurt for the rest of the day

>>715788017
https://github.com/IceFog72/ST-NoShadowDribbblish

>>715788821
It just works. I make all my WI vectorized.

>>715789636
Can you give some details on how you're using it with Comfy? I mostly just use SD/Automatic1111 so I'm wondering if Comfy has any interesting workflows for working with ST that aren't as feasible with SD.

>>715790803
Thank you, the only problem is that the pictures end up being too big on the chat.

>>715793340
Basically the scene/chat description in ST gets passed to a ComfyUI workflow that firts interprets the raw description and uses another locla model to convert it to booru tags. This llm also uses a rag for better tag and prompt knowledge, the result then gets fed into condition in the rest of the comfyui workflow and it's life as usual. It's very primitive all things considered and I've seen people use much better approaches. I haven't been bothered to improve on it because it works surprisingly well 90% of the time, granted you're not using too many concept loras and complex compositions, in those cases I just boot up my regional prompter workflow and do it manually

>>715795481
Forgot to mention, when I said
>the scene/chat description in ST gets passed to a ComfyUI workflow
It's using ST built in generate image functions, so with this approach you never leave the ST chat window, gens will show up inside the chat etc

>>715781452
Cute.

>>715794417
Look for var(--big-avatar-width-factor) and var(--big-avatar-width-factor) inside the css and change the multiplication values after them, there are multiple entries

>>715788017
https://rentry.org/meta_rice_list#sillytavern
check this out

>>715779765 (OP)
DeepSeek isn't free on Chutes anymore though. I tried using Gemini last night and its kinda meh, honestly. I downloaded the NemoEngine preset off reddit, and at least got rape porn novels working

>>715782751
Not deepseek, no. On top of that, deepseek outputs are unfocused scatterbrained schizoid blabbering compared to any better ones like 2.5 pro and sonnet/opus.

>>715800694
>scatterbrained schizoid blabbering
Preset issue. Unless you're doing strictly SFW or text completion it beats Gemini.

>>715800694
DS needs some wrangling but you will get better output, trust me.

>>715800816
>>715800961
The launch versions were kinda shit but after the May/June updates they're decent. R2 when?

>>715800816
>>715800961
Fine, I'll find some presets and give it one last chance. Shit is basically free anyway.

Question to everyone using cloud models, are you using straight API calls and paying per use, are you hosting local models on google collab or something? Also how's the pricing, I was considering renting vps time for this and the prioces would be good, but I would still be restricted to open source models, even if large ones

>>715801831
>are you using straight API calls and paying per use
This is always more efficient. The only way to "host" models online is to pay for GPU cycles per hour/day and that's far more expensive than just metered API access where you pay for each call as you go.

>>715801831
I just put $10 into openrouter and use the """free""" models

>>715802231
>This is always more efficient.
Yeah that's what I suspected. But then you have to jump hoops if you want uncensored rp and there's also the fact some ccp engineer at deepseek will be laughing at your degeneracy

>try to use AI for pure fantasy fulfillment smut
>the system is constantly trying to insert pointless drama and shit that has nothing to do with the sex
It loses its luster when I have to constantly tell it to stop trying to make my father's disgraced former business parter bug my apartment so he can use the endless train of cum stuffed femboys leaving it as blackmail

>>715802518
make a better prompt preset.

>>715802518
Preset issue once again. Remember that this shit is cached with every single prompt you send.

>>715797378
Thanks, but it doesn't help in making the image fit the space given. Specially the user's image.

>triple a games and live service have been so shitty in recent years, thinking about just dropping all regular vidya and fucking around with ai full time
i barely even play anything these days desu. for multiplayer pvp, it's all outright mtx lootbox scams or the same dogshit aim assist shooters. for singleplayer titles, they almost always lack actual choice. so i usually just watch a playthrough for the plot if it's interesting enough. llm story writing in comparison is fun, relaxing, and amazing for cooming.

What would you want to see done with ST?
(Within reason, so no graphics or real time video)

>>715805768
Exactly. I haven't been playing anything in weeks. Chatbots are much more fun.

>>715805853
As one anon mentioned some built in way to use the chat model to transform raw scenes into something more promptable.

I've never came as much as I did to ai its actually scary.

>>715808946
can't you just instruct the model to restructure some chat log into a reusable prompt?

>>715802518
>>the system is constantly trying to insert pointless drama and shit that has nothing to do with the sex
God I wish. If I don't tell it to do something it doesn't do it. It's incredibly passive which sucks when I want it to move the story forward.

Man, i just made a fucking kino RPG ending where my character sacrifices himself to save a city by giving himself up to be a part of Odin's Wild Hunt, in place of everyone in the city that were being condemned for their deeds. It even had an epilogue and everything.

First time i've seen LLM actually make an ending for a story.

>>715809393
prompt and/or model issue

>>715781495
getting middle east travel-ban

>>715809393
you can use world info to add 5% - 10% chance random event with 10 - 30 messages cooldown in between, and put something like

<narrator>
Something unexpected will happen now, your goal is writing it
</narrator>

Splitting narrator into seperate character and using group chat helps a lot

>>715811758
>gefilte fish ink
lol deepseek

>>715781625
>hasn't used Claude for erotica
It's expensive as shit but unless you have aphantasia, there is nothing in any media format that compares.

>>715813189
>claude sperging out each time loli mentioned
hm..nyo...

>>715813325
I will concede that using openrouter for your loli fetish is probably a horrible idea in general. Claude doesn't complain about highschoolers though.

>>715812540
This
I don't specify the probability, just wrote "there is a chance that {{char}} will write random event" on the narrator card and it makes the narrator proactive in creating random events
With group chat you can even have two narrators, one for keeping current scene flowing, one for triggering random events to spice things up if you wish, just be creative with the cards

What free service do you have to use now? I was using chutes but it's demanding money now.

someone explain, there's people PAYING for AI slopthroughs when there is like a library of Alexandria in terms of high quality porn easily available online for free?
Why?

>>715813549
>implying I use openrouter and not direct api

>>715813804
https://github.com/LostRuins/koboldcpp/releases

>>715813873
>implying anthropic isn't a faggot company that doesn't ban accounts doing this every month

>>715813858
Where can i play a text game where i am a noble knight who learns that he is actually a werewolf, because his mysterious mother didn't actually die in childbirth but is actually a savage from a barbarian tribe of werewolves, and after a long search he meets his mother, who walks naked, and fucks her while transformed into a huge wolf, knotting and all? then all the oedipal drama that follows that? that shit is pure kino, my man.

>>715814029
Local sucks horribly, doesn't compare to API.

>>715813858
just about all porn aside from ai is ntr cuckshit. you're jerking it to another dude fucking your waifu or whatever, with zero agency or choice in the matter. really makes you think, huh?

I'm not gonna pay for textshit anymore, and while I do regularly use local stuff that I can fit in my GPU the amount of rerolling, rewriting and fucking around I have to do in order to get around bad template phrasing has been getting old. Every time I ask for a model to try out that will fit in my 16gigs of VRAM none of it feels like it's genning anything that impresses me. I guess the problem is that I strictly want narratorial textgen, whereas most people I see that are pretty happy with it are just sexting with their waifu cards.

>>715814036
just dip out of chat by time she does pregnancy test, works for ban evasion every time

>>715781625
I just use Janitor because I'm lazy.
Haven't gotten ban for incest chats yet.

>>715813858
all forms of regular porn are quite limited if you don't have vanilla-tier tastes, and commissioning art is orders of magnitude more expensive than paying for a robot to write your weird erotica.

>>715813858
I have specific fetish for commiting suicide in 2040 because my nobel prize got stolen by co-worker then waking up 20 years earlier and deciding to pay kids to bully his daughter into suicide, while AI and mix of scripts displays in ever 20 messages sanity meter of my target along with current time, suspicion/police meter and my fatigue level so that I need to balance things.

>>715814190
Well then you better send of some more pics of your "ass" to the discord to get access