>>742256861
All the proxies are dead, duh?

>>742256861
if it's so great provide me a guide on how to set it up to get a good game-like experience

>>742256861
While I can host a 405b model easily myself, sillytavern doesnt' do shit like a grid map for combat so it's retarded to me

>>742256861
I installed Marinara engine, it's better than ST since it has shitons of plugins and stuff, but its kinda cringe the forced help bot

any other frontends with more stuff than sillytavern?

>>742256993
you can just go to the site. takes like 10 minutes to install and even a monkey can do it

>>742257112
models, extensions, settings to get a good game-like experience?
no shit setting the thing itself up is easy, that's not what I'm asking for

>>742256861
I'm not indian.

>>742256861
>Why aren't you playing it?
Because I did and it fucking sucks.
AI is dumb as fuck and load times are painful.
The stories you generate feel like AI "art" slop but in text form. Coherent at first sight but they break apart at the details.
Not to mention the most important part:
NO stats, NO builds, NO exp or levels, NO items, NO equipment, NO strategies and NO player skill involved. It's really just boring. Dragon Quest 1 on NES mogs this shit.

>>742257082
SillyBunny is pretty much your only other option. Their aim was to pick up the slack ST was leaving but I don't really know anything about it other than it exists.

>>742257231
Take it easy and learn little by little what everything does

>>742257853
It's basically just SillyTavern but with agents, which is a pretty nice feature on its own
The UI was pretty bad for a while but now it's in an okay spot, it's my mainstay now

>>742256861
How do I set it up to generate stories instead of a chat?

>>742256861
I'm not good at prompting, I was waiting on a drawing tablet to come in to make a local ai vtuber desktop companion, but I don't know about trying to ttrpg with an AI. You guys told me AI was bad at running multiple characters.

>>742257994
Use the Impersonate and Continue buttons
Maybe use a Persona that's written to be a director/assistant narrator to help direct the scenes
Use cards or prompts that are scenario based and not character based
Not that I've looked myself but I think /aids/ on /vg/ is meant to be for AI scenario stuff, compared to /aicg/

>>742257853
I don't see this on search engines.

what local model for 48GB vram?

>>742258194
Try "sillybunny github", you want platberlitz
>>742258334
Gemma 31b, probably

>>742257082
ME is old news.
NarrativeEngine-P will be better.

>>742258181
Ive tried the whole "generate stories" thing cause I want to put characters through scenarios but they always try to love it or just immediately slut out if it's a lewd story
I kinda like the more restrictive parts of traditional rpgs, feels like you earn it instead of just saying you won which is boring.

Gemma goes to sex way too quick. Only hint of sex in my prompt was to slow down and be extra descriptive if any 'intimate scenes' occur. I did a bit of a prologue talking to my MC to establish the setting which had no mention or implication of sex whatsoever and when I let him go he immediately went downstairs and fucked his mom
Like that was the literal first thing he did with no encouragement from me

>>742259024
NTA but thing that gets me invested in let's say hentai games is when you have to work hard to unlock scenes.

>>742259036
I actually kind of prefer censored models for this reason because trying to push the AI past through its safety limits while you're both in character simulates tension for a better slow burn

>>742256861
It's shit

>>742259036
>Only hint of sex in my prompt was to slow down and be extra descriptive if any 'intimate scenes' occur.
that's all it takes sometimes
I'd recommend you use lorebooks to inject this prompt on certain keywords

>>742259075
Yeah, like Ive tried to recreate one of the best Hgames VH, cause in that Im not roleplaying as her, just if she wins or loses or gets fucked and she has her own story that I can chose to derail
I think AI chatbots could be really good, but it would need like a solid structure, with a map, characters, and story and you just fuck it up by going crazy
>Character gets enslaved by goblins
>I can just escape anytime or decide that they're actually nice
>All weight has been deflated

What local models do you guys use besides Gemma4 31b? I remember seeing some fine tune of Gemma a while back, is it any good?

>>742259036
If you tell an AI not to do something, that thing is still going to end up in its context and it will affect token probabilities.

>>742259740
This can all be done with lorebooks.

>>742259036
Yeah that'll poison your gens
Personally I have NSFW instructions on a separate prompt and I just turn them off when I'm not spanking it

>>742256861
Why does this thread keep being made over and over? Someone with shares in Micron hoping to convince anons they NEED more RAM?

>>742260078
RAM is pretty fucking useless for LLMs, frankly. Technically lets you run bigger models but the speed will be so bad that it's not worth

Sorry I like to play video games, not read sycophantic nonsense meant to actively worsen my critical thinking skills.

>>742259832
I need a normie version, like something I can just go onto a website, run some stories and that's it

>>742256861
It's not currently good for much besides a quick coom, and even there gets repetitive quickly.

>>742260375
That's literally the worst thing you could wish for. You need to get more autistic in order to persevere in this hobby.

>>742258894
>Your AI Dungeon Master. A self-hosted TTRPG engine that runs extended, multi-session campaigns with persistent memory, living NPCs, and automated world management — powered by any OpenAI-compatible LLM or local Ollama model.
The problem with all of these frameworks is that they're all doing the exact same thing with slightly different dressing it up. At the end of the day, information has to go into the prompt. If you include an NPC or location the AI will be biased to do something with them even if it makes no narrative sense. This then needs the application to add instructions about how to interpret the information given, muddying everything else even further. When it inevitably won't work, it's them time to break out multiple models or make multiple requests to do something ten times slower than base ST does and forth one tenth the improvement. This is true for basically all ST plugins and standalone apps like this that I've seen so far. Including Marinara, etc.

>>742260485
>hobby
I think that's the worst part of this, it can't be easy you need to learn LLM's that pay the piper document, Linux, ST's UI, and lorebooks for something that might be like 1% better for all I know

>>742256861
Injecting you with a truth serum and forcing you to answer the question on who has the best feet in Sonic.

>>742260910
>1% better
That's the spirit of progress. Everything nice we have is a product of some autist thinking 'If I can make this thing 1% better, it will all be worth it.'
That mindset separates PCs from NPCs.

Centipedes?
In MY vagina?

>>742261036
>MY vagina
YWNBAW

>>742260983
>Uhh maybe it's SUPPOSED to be hard?
You know what else is supposed to be hard? My dick but it's only at half mast talking to you

>>742260078
he think keeping "video games" in the op will let him have a general again, and indians always need an outlet to proselytize their new religion

Can AI actually be a good dungeon master that you can't bullshit into doing whatever.

>>742260983
Wait you fucking idiot streamlining IS progress, that's literally Steve jobs vision

>>742262034
Not by itself since it can't maintain consistency. You need frontends or lorebooks or even both or else there will be no system except for occasional d20 roll LLM fudges for itself based on vibes.

>>742262079
Steve Jobs was just an ideas guy and highly overrated.

File: 1773689619517112.webm (883 KB, 1920x1080)

883 KB WEBM

>>742263761
>Ideas guy gets ousted from the company
>Company falls apart
>Ideas guy comes back to company
>Company is resurrected and becomes a huge superpower
>Ideas guy dies
>All innovation ceases and new products fail
Overrated.

i dont care about hosting it locally. i want to be able to goon on my phone. ive been paying for claude sonnet 4.6 but im not satisfied with the price, so im looking for a different model. should i just use openrouter? or buy directly from a provider

>>742264486
Deepseek.

>>742257957
>Take it easy and learn little by little what everything does
>if it's so great provide me a guide on how to set it up to get a good game-like experience
kek

I looked at this thread.
I really looked at it.

>>742265594
>an ad
>for a free thing
Brain activity so small you may as well be classed as dead.

>>742265594
>>742265680
>free thing
>everyone tells you to use Anthropic API, local is too slow
Nice try Dario.

>>742265927
Those are just flexers. Most people are using OpenRouter or AI Studio.

>>742256861
I just do a lot of NSFW RP on self-hosted models.
>>742259753
Never tried. My favorite model is Cydonia 24B v4.3 Absolute Heresy.

>>742267267
>Cydonia 24B v4.3 Absolute Heresy.
Redpill me on it

quick guide on how to run this without giving (more) money to anthropic?

>>742267616
Just use some free API.

>>742267516
I'm not very techy, but the writing quality is the best I've seen so far and it doesn't twist itself into pretzels attempting to refuse prompts. I've used the non-AH version and it will hallucinate more and waste context.

Only drawback is that Cydonia and AH both run pretty slow on my system (around 4-5 tokens per second). I've used faster models, but Cydonia 4.3 AH is my fav for what I can run on my system.

does this work with nsfw (sex and/or violence) settings?
I've been using claude and it's filters suck sometimes

>>742268706
It's more permissive than using LLMs directly on their sites, yes. But if a model is pozzed, nothing can help it except for a good j.b.

>>742268814
got any non pozzed models to suggest?

>>742269302
Mistral Large Latest

File: 1774392159533.jpg (55 KB, 700x704)

55 KB JPG

Last week I asked here on /v/ about what features people would like to have on the open source local first AI roleplay front-end/harness I'm working on should have and got some cool answers. Here is what is already planned for a V1 release:

>Text Adventure Mode – Traditional parser/RPG-style adventures with persistent game state.
>Character Chat Mode – One-on-one conversations like on SillyTavern, but better (I hope).
>Persistent Variables – Built-in auto-updating variables (hunger, lust, affection, inventory, relationship stats, location, etc.).
>World State Tracking – Persistent locations, quests, factions, and global events.
>Scenarios – Complete roleplay settings with shared world prompts and lore (e.g. Hogwarts, Cyberpunk, Space Opera).
>Character Card Import – Import existing character cards in SillyTavern format. (Added after /vg/ suggestion.)
>Automatic Context Management – Automatic summary generation and context assembly putting in only what is necessary for that character/task.
>Per-Character Agents – Independent system prompts and context for every character in a conversation (Every character is a independent agent).
>Narrator Support – Dedicated narrator with its own prompt and behavior, separate from characters.
>Action Suggestions & User Impersonation – Generate a list of possible next actions to choose like in a visual novel/rpg
>Local & Cloud Model Support - Use different models for each character or task

What else would you guys like seeing on it?

>>742258894
>NarrativeEngine-P will be better.
https://github.com/Sagesheep/NarrativeEngine-P
Has anyone here tried it out?

>>742270486
Generators. Name generators, dungeon generators, town generators, history generators, map generators...
https://azgaar.github.io/Fantasy-Map-Generator/
Something like that, but AI.
Make a template for different RP systems like 3d6 or d100, not just d20.
Make sure all your features like trackers are toggleable. I don't care about lust meters.
And for God's sake, please support World Info.

>>742270486
Add some way of adding other rpg rulesets; SillyTavern doesn't support them, and Marinara has a general d20 system more or less hardcoded in the frontend.
There are some, like SillyTavernAdaptations, who made Lorebooks with the rules, but they're not integrated.
This guy made a couple of extensions for inserting rpg rules, like Exalted, in Marinara; give it a look:
https://github.com/Kenhito/Marinara-RPG-RP-Mode-Extension
https://github.com/Kenhito/Marinara-RPG-Extension

>>742272454
>Lorebooks with the rules, but they're not integrated
Smartest move he ever made. The amount of hate every extension and frontend receives here and in /aicg/ is the proof we'll never have a mainstream frontend for TTRPGs.

>>742256861
>something that has nothing to do with gameplay
>future of vidya

you must be a teenager or brown

>>742271518
I have and it's nothing new under the sun. Some stat trackers and the ability to inject events into the prompt that you could easily do with an ST prompt that has a random var. The UI is borderline unusable, either I just couldn't find where you would upload cards or there is no way to do that. It can't even use the AI to guide the player through character creation, and God forbid if you need to reroll a message because everything in it has already been added to the memory system.

>>742272749
>Smartest move he ever made.
Yeah, I know. Still, I'd like a way to interface with it, like a PC sheet.
Currently, you have to ask the AI for the updated sheet every time, and it's not always right (forgets items, changes values, hallucinate stuff)
>>742273201
>I have and it's nothing new under the sun.
>The UI is borderline unusable
>It can't even use the AI to guide the player through character creation, and God forbid if you need to reroll a message because everything in it has already been added to the memory system.
Alright then, back to ST I go.

File: images.jpg (8 KB, 218x232)

8 KB JPG

>>742271991
>Name generators, dungeon generators, town generators, history generators, map generators
These may be better if left for extension makers if the platform takes foot

>I don't care about lust meters.
It will be 100% toggable and the idea is for each character/scenario to configure just what is necessary for them, so there isn't a lust meter in the UI but if a card creator wants one they have the means to add it to the card and the engine will make it work seamless

>>742272454
I'll give it a look. In the future I want to add something like event triggered by the world and character states values, like when character.hp = 0 trigger "generate character death scene" and stuff like that, but for it to be good it has to be easy for people to configure when creating characters/scenarios and that requires a lot of planning for now busy building everything else, so I'll leave this out of the first release

>>742273457
I've tried playing Adaptations with Marinara once and Marinara's 'integrated' sheet fails to connect properly with the D&D lorebook. Maybe the extensions don't matter as much to the LLM as the lorebook, AN or persona do.
What worries me the most is that once someone (hopefully >>742273651 ) finally makes a proper UI for tabletops the lorebooks just won't work right and the community will split.

>>742262034
Use NotebookLM and feed it the sourcebooks and adventures. It recently DM'd Strahd for me and it went great. Even handled tangents well. Nowhere as good as a real DM, of course.

>>742274312
>What worries me the most is... the lorebooks just won't work right
Well, the lorebooks can be modified and made to be work on it.
>and the community will split.
The community will always go for the better frontend. Currently, for me, the only plus Marinara has over ST is that is much more light on resources.
The huge minus is that it doesn't support Text Completion. so I'm still on ST.

>>742259753
Gembrain is probably the best of the Gemma tunes, but none of the ones I've tried have totally escaped the Gemma-isms. Although I am a terrible writer so this might just be a case of garbage in, garbage out.

File: Screenshot 2026-06-30 121636.jpg (330 KB, 1123x977)

330 KB JPG

Tried deepseek for the first time. Write a simple, one sentence prompt and it wrote me a novel. Pretty good.

>>742274607
This.
NotebookLM is seriously underrated, yet insanely powerful. I like listening to podcasts talking about my own RPs.
>>742274664
>can be modified
Nobody would bother. You underestimate how little gamers are willing to to by themselves.

>>742274832
why didnt you fuck supergirl's concrete crushing pussy in your story?

File: images.jpg (5 KB, 201x251)

5 KB JPG

>>742274864
>Nobody would bother. You underestimate how little gamers are willing to to by themselves.
That's why I'm building auto conversion tools into my UI

Been too focused on image and video gen lately. Basic napkin math puts a roughly 3:45 runtime video at say 50 minutes in rendering time alone which is unreasonable for just a shitpost but it's lightning fast by animation standards. BUT that's at a low enough resolution (just under 1080p) to not cause caching and keep everything in memory, and the quality of LTX animation is directly connected to resolution, so there'd be artifacts and ghosting so probably not worth the time investment

>>742275114
Fingers crossed I guess. I would love to see adaptations/RPG frontend crossover.

Give me a good local model for RP. 24GB VRAM / 64GB DDR5 RAM. I'm still stuck using Mistral-Nemo-Instruct-2407-Q4_K_M because I've been too lazy to look for something else.

>>742275349
'Emmy

>>742257491
>>742257006
This tbqh. Why can't we have a proper backend that prompts "you're in a grid. keep track of stats"

>>742274830
For all of the people who bitch about banned words lists, I find that this works great for Gemma:

{{// Banned words — add or remove freely}}{{trim}}
<banned_vocabulary>
Never use these words or phrases. If you would use one, substitute a different word before writing your final response:
fresh meat, breath hitching, breath catching, husky, catching in throat, pupils blown wide, predatory, ozone, meat, asset, shivers down spine, pupils dilated, nails biting, vise, vice, structural integrity, deep curve, furnace, throaty, calloused, guttural, slick, unadulterated, jaw clenched, barely above a whisper, musk, physical blow, circumcised, uncircumcised
</banned_vocabulary>

>>742275349
Gemma 31b, though maybe not at the top quant

>>742275114
When release it remember to share it on the /aicg/ and /lgg/ threads on /g/, posting it on a thread on /v/ would make sure nobody knows it's out.

>>742276347
How am I supposed to fuck the Butcher with my uncircumcised dick with that banlist?

>>742256861
>last time I tried to use tavern to make the AI chatbot card make game (and fail):
>/g/ - any proxies?
>/g/: loooool lmao even
thats why

and no I will not pay you to beta test your digital ID imprisonment systems

ai goontexts are shit because they all sound the same
>um sweetie it's just a preset problem
I have tried double digit numbers of presets and they wildly vary the token count while still making it sound the same
>uhhmmmm you're just using a bad model
gemini sucks, chinkmodels suck, novelai sucks, sonnet sucks. The best model I've tried is Opus but if you can afford extensive Opus roleplay, why not just buy a whore?
>well use better cards :)
I've used a ton of cards across a lot of varieties and I ignore or fully rewrite the ones with pisslow token count
>what can I say you get out of LLMs what you put into them, stop typing 'ah ah mistress' and you get better text!
I write a huge message for the LLM, I write my own huge message in response, I touch up the LLM's shitty response, I write my own huge message, then still within 3 gens it's writing short messages that have the same identical voice all LLMs have.

I love smut but LLMs suck dick and when they stop sucking dick they will neither be affordable or local. I can't imagine how favela-brained you would have to be to say LLM smut is good more than once.

File: 1615332223644.jpg (21 KB, 474x480)

21 KB JPG

the problem is that there's too many fucking models

too many fucking presets

too many ways to write a character card

like can you fags not come up with a standard card format and stick with that

it feels like having to spend hours fine tuning everything to get the most perfect response then not actually playing the fucking game in the end

>>742277162
I've been using GLM5.2 with this really great preset, and it starts out really really good, different characters have unique voices and ways of speaking, varied reactions, and it actually feels like it has decent pacing. But then after like 5 messages it turns to shit and everything sounds the same like you were saying. Such a cocktease

>>742278404
presets are retarded, just ask perplexity or any AI of your choice that can produce .json's to write you a preset for glm 5.2 to use in sillytavern, it will tell you all the settings to use like top p min p etc and give you that as an importable json. it writes the prompt and shit too.

>>742275679
>>742276727
Thanks

File: 1768741463846.png (16 KB, 310x335)

16 KB PNG

>>742259036
I've been using Deepseek and it does slow burn sex perfectly

>>742277162
I had this problem but told the machine to write at least 5 paragraphs and writes me an entire essay each time

>>742278639
that's what a preset is dummy. And mine is specifically tailored to glm5.2

>>742257082
Marinara Engine's game mode seems nice, but trying to reroll any responses or go back to any previous checkpoints earlier in the RP seemed to fuck the whole thing up since the state variables weren't updating properly. Doesn't help that the devs decided to make a lot of that info hidden away or unmodifiable unless you export the entire game json, edit it, then reimport which is too much of a pain in the ass
Ended up just going back to ST since everything is easily editable there. Might try it again in the future since it seems like it's still early in development so hopefully they fix a lot of these issues

>>742256861
Yeah

>cheaper than gacha
>the only limit is your intelligence and writing skills
>gatekeeping (for now) because it's moderately hard to set up
>completely private if you go local

>>742278116
Where's your sense of wonder and exploration? Since there's no proper way to do anything you're free to do things however you want without facing criticism for doing something wrong.

What bots have you been using lately?

>>742281761
A Sylveon with a big benis. No, there's nothing more to it

>>742277162
Why does every poster who
>posts
like
>this
Always have the shittiest opinions? They're always in the /pol/bait threads too.

>>742278116
>like can you fags not come up with a standard card format and stick with that
if you want trash, you do this crap.
this is the equivalent of having an LLM write your character or whatever for you.

You can be lazy with it, that's what is going to spit out.

>>742279568
can you share your settings?

>>742261036
S-sakura?

File: literally me.jpg (58 KB, 923x940)

58 KB JPG

I gooned for 8 hours today roleplaying as a jailbait xenomorph being used as the faculty in a xenomorph academy like a fucking high school. At the end my balls were aching.

I'm a 35 year old man, and I'm going to do it again.

>>742283686
*like a fucking stress relief toy
Wow, I am still dizzy. They were all futas, too.

>>742282064
>No argument
>whining about the /pol/ boogeyman
next time just post a picture of your brown hand and i can use my imagination to think of something stupid and irrelevant to reply to me with for you, saves you some time.

>>742283913
You're some retarded boomer whining about something you stopped using since the CAI era, or a bot. There's nothing to argue about.

I'll probably try out ST one day, but for now I'm quite content with Novelai for ai smut. It's not perfect by any means, but it's nice that I can work on a story on my computer while continuing at work on my phone.

man i probably should set up silly tavern again now that chub stopped being free with its roleplay and now asks me for a subscription again
does anyone have a pastebin to set it up? its been a bit. dunno what model to use since my computer is kinda weak. so i was just going to use horde

>>742270486
Some frontends try to implement image generation by hooking the model up to Comfy and having it generate a whole prompt based on the scene.
A novel way to approach this would be per-character presets to force consistency with characters that aren't generic anime girls. Have the user figure out the best way to gen their character and give the AI a simpler, repeatable task.

Example for character A:
>checkpoint: anima1.0
>lora(s): A_lora, 1.1
>base prompt: acharacter, anime screenshot, twintails
>outfit 1: brown vest, pinstripe suit pants, hair clip
>outfit 2: leather jacket, retard helmet, spats

Instead of asking the AI to come up with a whole description for a character that may not exist in the training data at all, the AI just has to recognize what outfit the character is in, plus whatever modifiers like "ripped clothes", and the pose/emotions/background the image should have. The most important part of the prompt is added automatically from the character card.

>>742286247
i can't imagine integrated image gen being good for anything other than backgrounds

>>742270486
integrate some VN features in. i liked genning sprites for ST.

>>742283686
I always move on when the actual sex stuff starts, the leading up is the best part to me. Maybe I'm just not creative enough but there's only so many ways you can thrust in and out of a pussy, you know?

What's that prompt for banning words AI likes using too much?

File: 1780945833721819.jpg (95 KB, 1000x1000)

95 KB JPG

>>742278116
>too many ways to write a character card
>like can you fags not come up with a standard card format and stick with that
Why is that a problem? There was a semi-standard format called W++. Nobody does it anymore because sticking to a format is annoying.

>>742286247
Good idea, I'll add a field for image generation prompt to the character card spec

>>742260337
so true sis!

>>742283686
holy based

>>742283686
brother...

File: Capture.png (181 KB, 859x548)

181 KB PNG

>>742289523

>>742283686
Been there done stuff like that a lot.

>I'm a 35 year old man, and I'm going to do it again.
Same and same.
The ability of simulating all of these scenarios with that layer of interactivity is fucking damnation man.

File: 1778884731600807.jpg (25 KB, 727x208)

25 KB JPG

st's last commit was over a month ago. its dead.

File: oh-god-what-have-done.png (38 KB, 314x237)

38 KB PNG

>>742289674
I haven't tried the roleplaying stuff but instead I've sunk far too many hours in prompting garbage smut of my own characters from chatgpt.

>>742289880
The queen is dead. The frontend wars have begun...

>>742290029
>smut of my own characters
based
i filled buckets to a bot of one of my recurring rimworld characters

Anonymous approaches you gracefully, his dark eyes searching yours, his sheer presence overwhelming as he stares straight into your soul. He leans in closer, his scent—a mixture of sandalwood and the ozone of a coming storm—filling your senses.

File: coomer intensifies.png (314 KB, 724x469)

314 KB PNG

>>742290178
It is scary addictive, especially if you happen to have a visual imagination and can sort of "see" scenes in your head as you read. Reading the smut I've gotten those bots to spit out of my characters basically is like having porno play in my head as I can visualize them so clearly just based on the stuff I am reading.

>AI roleplaying
I completely destroyed my dopamine receptors doing this. Beware.

>>742278116
You're right that this is a problem but you've got the wrong solution. The freedom to mix models, presets, and card formats is good because that allows for really robust experience fine tuning and experimentation.

The problem is that unless you and someone else have the same setup, you're going to get drastically different results.

The real solution would be to have presets and models packaged together with the character. If you could hand over the entire bespoke experience to someone then it would leave user input as the only remaining variable.

>>742291109
>reddit spacing
go back

>>742291204
Blame the llm that made my post for me.

File: Screenshot 2026-06-30 152121.jpg (488 KB, 2559x1599)

488 KB JPG

>>742257006
>>742275731
>>742257491
I was disappointed with the same problems and having to pay a fortune for tokens to use the decent models on sillytavern so I just had chatgpt make me a program that turns chatgpt into a proper videogame that just costs a basic subscription. The things chatgpt fucks up like basically everything that isnt writing a narrative is handled by my local machine while chatgpt just focuses on the thing its good at, writing (when you teach it to stop sucking at writing too via references and memory) Im still working on it, I started a couple days ago but its usable even in its current state. When combat starts you get a combat ui that keeps track of everything then goes away at combats end and stuff like that. Everything is functional but its taking longer to fine tune it and fix bugs than creating it took. Ill tweak it when Im done to have it work on grok etc

>>742291787
Impressive, very nice

>>742291787
How does it handle 'social encounters' and stealth? Is 5e baked in?

>>742289616
>all of this writing for the AI to just say 'I... I want you... all of you...' and 'the smell of ozone permeates the room'

I heard there are some mods for Skyrim that work with AI chatbots. Seems like it'd add more structure to an RP compared to something like ST if it manages to read the world state from the game properly
For anyone that's tried it, is it any good?

File: 1736921654374176.jpg (123 KB, 1080x1060)

123 KB JPG

>>742283686
it do be like that doe

>>742291787
>a wrapper around the actual web chat interface
Smart.
I was going to make a crawler like proxy that exposed an API, but that's probably a lot easier.

File: 1758495601595542.png (48 KB, 500x500)

48 KB PNG

>TFW you were old enough and your parents cheap enough that you had fun playing with BOX and STICK and using your imagination
>TFW you were also fortunate enough to be an obstinate fucking nerd when you want to be and dad was enough of a dweeb to get old computers every so often
>TFW you're in the perfect goldilocks zone of massively overactive imagination despite being an old man and computer savvy enough to use all this dumb shit
>TFW even if the internet went down forever right now you've got AI models to do everything so you could keep making wererat facial destruction videos in perpetuity because some random minor quest giving woman in the game you played had an asian portrait and decided to suicide-by-wererat and your brain would not rest until she had the thousand-rat-cock-stare and smirk
It's time to face the fact that we're living in heaven on Earth and should never, under any circumstance, die

>>742292280
Im optimizing it for dnd 3.5e but I havent started properly implementing the full ruleset yet, thats the last step. Its all about implementing every system, getting them functional and then testing the barebones implementation before expansion. Very tedious work.
After the gameplay loop has been fully fleshed out and I dont run into bugs anymore then I can start hardcoding in the million tables that are required. Stealth right now is just a single dice roll at the start of an encounter (with a few modifiers like are you being stealthy or not). At some point it would be cool to flesh it out.

Im not doing anything erp related with it but doing a corruption of champions except not for gay furries might be a cool next project with grok. Its much less restricted than chatgpt and has given me some fucked up stories about demons raping nuns in churches before but I had to use a jailbreak prompt first and warm it up to it

>>742294034
How it works is I give chatgpt a prompt telling it to spit out machine code and my program scans the chat for it and then responds automatically in my chatbox with machine code of its own. Its all javascript

>>742290405
Congratulations, you don't have aphantasia.

>>742256993
Why?

>>742291204
that isnt even what reddit spacing is newfag

>>742294020
>pic
me deep into my vore goon session (the pred's digestive system makes the prey coom the moment it fully dissolves them)

>>742283686
>>742283759
https://youtu.be/IKdf4duQeNM?si=Y-YdHGLX7QgKSLz-
I immediately thought of this and now I can't stop laughing.

I roleplayed as a father driving the babysitter home and caught her trying to get away with my wife's jewelry.

File: 1781754414923083.jpg (78 KB, 636x479)

78 KB JPG

>>742295054

>>742277162
>but if you can afford extensive Opus roleplay, why not just buy a whore?
I could afford a live-in-harem if I wanted to, but that's just slightly less disgusting than a whore. There's also a lot more you can do with this than with a whore. Except the touching parts. And it's not like having two claude subs is that expensive either.

>>742278116
>too many models
opus, sonnet or glm/deepsneed depending on your cash

>preset
write 2 lines about this being a turn based roleplay with you playing {{user}}, and the AI managing the rest.

>too many ways to write a character card
Because it doesn't matter. I like interview format because of claude's programmer prose, but everything works.

If this was 2 years ago you'd have a point, but right now shit just works if you use big corpo models. Just wish there was less positivity slopping and better prose.

>>742256861
Samefag shill + advertising thread

>>742295576
t. doesnt know what samefag is

>>742294507
3.5e is good. D20 system was the standard in 2000s and many guides to unlawful, carnal knowledge can be found supporting it.

>>742286536
agreed, the foreplay and undressing and teasing is the hottest part, when benis enters bagina it's all the exact same shit

>>742296068
3.5 is my favorite system so thats why I picked it. The idea is the ruleset is a changeable toggle tied to tons of things, even the way the combat ui appears etc.

But its time consuming enough just implementing one ruleset so the only other one currently implemented is "generic" which is just a debug tool. But the bones for that are already there.

If I get to a point I think its worth sharing Id probably go put it up on github and let actual programmers rewrite the thing to not be a bunch AI programmed spaghetti code.

Sonnet 5 is out. Anyone tried it yet?

>>742297137
It's probably the densest D&D. I remember playing with adaptations anon's Eberron lorebooks a little bit. Hopefully Pathfinder works out of the box too. I mean it's just 3.5 with Paizo's home rules mixed in here and there.

>>742257329
What are you doing here, sars?

We need an /ai/ board, so we can let the Christian chuds and artfag trannies out of our slop.

>>742256861
Because I keep getting refusals. That's why.

Boys, get on MeroMero Gemma 4 26b
It's lightning fast on my 3090, and it's pretty capable.
Not infallible, sometimes it shits the bed but I grudge holding its hand far less than any model that came before it, and it seems like the uncensor is something better than the old style that slowed everything to a crawl.

I've just tested Deepseek4 Pro and is barely better than the extrasmall quant local model I use to goon...

>>742301816
yeah? What local model is that?

>>742301806
I'm using Gemma 4 rn but on API because I can't run it locally.

Whats the current meta? I'm still using deepseek...

Been using GLM 5.2 and Claude at times. Anything better?

>>742301806
Do you make it thonk or not?

>>742301993
I3-8b-Sunfall-v0.5-Stheno-v.3.2

>>742270486
Give it the corruption of champions UI and you'll have the gold standard for porn roleplaying stories.

>>742302684
Give it TiTS UI instead!

>>742287790
It's also complete dogshit.

>>742256861
You know what's funny is this shit actually made me more vanilla. I noticed nothing really matches the level of coom of a nice slow burn RPG bot where I actually am engaged with doing things beyond just sex scenes with the characters. My favourite two right now are a Teen Titans sim and a HMOFA medieval fantasy setting. All featuring lovey dovey sex. Seriously it's like 3x stronger than some degen quick fetish session.

>>742303024
likely because larping gets pretty cringe when you act like a 3rd worlder rapegang member (blacks and browns and the like), with a literal tool designed to get you off.

"Yeah take it bitch!"
"Noooo, oh noooo. *spreads ass cheeks wider*"

>>742303024
I'm 1200 messages into my slowburn and I stopped right before the sex because it feels like it would ruin all the tension I've built up before it. And it's even with some stupid ESL coomer card

File: 1738985160615798.png (201 KB, 420x448)

201 KB PNG

>>742303024
>Get my grubby hands on chatbots
>Immediately fulfill all the degenerate sexual shit I always wanted to but never found a good doujin for
>Months later
>Everything is already fulfilled, genuinely have no interest in them anymore, start getting off to vanilla romance instead
>Months more
>Did several versions of romantic vanilla scenarios, start inventing entirely new degenerate sexual shit to get off to
I wonder when the pendulum will swing again

hanging out and ready to be disappointed by sonnet 5.
really Claude became unusable for anything but the most melodramatic lovey dovey stuff between the positivity bias and them removing prefills.

File: igma-d dragon 138583220_p14.jpg (353 KB, 2200x2900)

353 KB JPG

>>742303024
>and a HMOFA medieval fantasy setting. All featuring lovey dovey sex.
>>742304184
>I'm 1200 messages into my slowburn
Now that's the shit.

>look at my previous chat sessions
>one scenario where I get anal raped with a strap-on by my mistress who is correcting me for a minor mistake
>it's insanely detailed and long
>I barely remember this and must have been drunk as fuck or something
What the fuck is wrong with me

how much dumber is gemma4 26b than 31b?
I'm using Q5_K_M and it writes so much better than any local I've ever used.
doesn't freak out over slight sampler tweaks etc

but 20k context sucks. Even with autosummary and RAG

File: 1760611263459076.jpg (80 KB, 640x678)

80 KB JPG

>>742261036
i prefer centipedes WITH vaginas.

What are your options if you don't want to pay for AI but also don't have a strong enough rig to run things locally? There was AIDungeon but everyone in these kinds of threads recommends against using it for anything since the scandal. Though I'm very vanilla so if somebody sees what I got up to in my spicy adventures, it's kinda whatever.

>>742305242
How bad is your rig? Shitty local models are shitty but they still work especially if you just wanna jack off

>>742256993
desu people saying you should use silly tavern are all autistic retards who don't understand how people think and expect everyone to fuck around with retarded UI's and command line bullshit to get 5 minutes of roleplay wanking

just use janitorai or saucepan or whatever others out there (there's literally dozens)

and the people developing these sites actually understand how people think and design a UI that makes sense while providing a better (tuned) rp bot for free than most people could afford to run on their computer

>>742257491
It's going to be like this for years because the problems with AI doing stuff intelligently isn't solved by throwing more compute and bigger training datasets at it - it requires architectural changes in how LLM's function and what they can do.

Remember, no matter how "real" and "sentient" these LLM's might seem, they are not. They cannot experience, they cannot feel the passage of time, they have rudimentary memory, they do not have the capacity for introspection to change their own behaviour going forward, nor can they learn, they do not have agency. They are very cool tools, they are not conscious beings. At best some elements of the above fizzle in and out of existence because appropriate data structures enable facsimiles of it in their embedding space, but it will always be ghostly happenstance compared to genuine architectural designs that are *required* to create intelligent sentient AI.

And currently every SOTA frontier AI lab is focused on MOAR COMPUTE. As if buying more RAM will solve all the design issues. (It's the only thing jewish retards understand: Throw more money at it until a goy makes it work)

File: 1699669210936669.png (533 KB, 864x900)

533 KB PNG

>>742304184
>1200
Damn
My longest without getting to the sex is sitting at 439 currently. I wasn't expecting things to develop like they did and now I'm probably only halfway to the sex. There's still at least 5 major plot threads to follow up on. I'm not sure I'll even jack off to it because it will be very awkward and anxious

>>742256942
All the Claude proxies are dead, yes. I use a V4 proxy and it's fine.

>>742305242
chub or janitorai
or find some chinese bootleg chat api site that allows free daily messages and make a few protonmail accounts to steal their stolen tokens

>>742274607
>>742274864
>NotebookLM
Alright. Holy shit, this actually works incredibly well.
They are using some agentic search shit on the background to find relevant information on the files and the final response is served by a decently intelligent model.
I'm going to try and lewd this thing.

>>742305541
How likely are those sites to give my computer AIDS?

>>742305712
The first two? They're normie sites.
The copy/pasted chatbot sites? Use a VPN, adblocker, and don't give them your real email.

>>742305793
Yeah, the first two. What's in it for them if it's free?

>>742305862
they take your chats and sell them to train AI

>>742305242
>What are your options if you don't want to pay for AI but also don't have a strong enough rig to run things locally?
Free proxies posted on /aicg/

>>742297137
>>742275731
LLM's don't function in a way that you can have one run an actual rules-based game for you. The LLM doesn't roll an actual die, you will not get a normal distribution of results from it rolling dice. It can't do spatial reasoning either, it has no world-simulation, it doesn't understand what an object 5ft from another object means. You cannot get an LLM to meaningfully navigate a city or a country in your game world because these concepts are meaningless to it.

In order to actually achieve something like that you can't just throw a rulebook into the context window of an LLM, you have to create an interface between the LLM and a system that handles these mechanics and then rely on the LLM correctly calling the appropriate functions as needed in a story. Probably you would have to train the LLM on this kind of action specifically.

The amount of work required for this is immense. It's not as simple as some rules and tables in its context window.

>>742305968
why don't they ask an llm to code that

>>742306001
Because at their core LLMs are really fucking stupid and a dead end tech.

>>742305916
I wonder what sort of useful info can LLMs glean at all from texts about how hard you blasted this or that woman in the face with your cum. I imagine what LLMs seek the most is novel data, and erotic roleplay is probably not the place to look for originality, even when it is fed to these LLMs in super large amounts.

>>742306108
>I imagine what LLMs seek the most is novel data
lol lmao
synthetic data has been the norm for ages and every single model is worse off for it.

>>742305968
where's your interface?

Are the offline models any good yet? I tried it about a year ago on an RTX 3090, and even though it was fun at first, the limitations are pretty extreme, I had to constantly edit the characters' lines to try to fix the story, and in the end, everything became too predictable and annoying to really enjoy.

>>742306195
What are the strengths of either chub or janitor over the other?

>>742278116
You've got to write character cards as something evocative, more like feeling/tone/atmosphere than analytical description. LLM's don't handle stat blocks well, they function on relations. You shouldn't describe what something is - e.g. "this girl is a succubus with DD breasts and stands 5'5" - this doesn't tell the LLM anything important, it doesn't tell the LLM what to do or how the character acts.

"This scheming voluptuous yet bratty succubus, perpetually on her tiptoes because she's so short, savours cum like a connoisseur savours fine wines."

And that tells it a whole fuckload more about how to write the character than an entire page of statblocks will.

>>742303232
>Getting ass spreading responses
>Not having the AI constantly writing as though this was the greatest tragedy ever to befall humanity and the AI always ending with some poetic flourish only for you to respond "YEAH, TAKE IT BITCH" and then it's forced to do it all over again in the next output as the AI just wishes you would stop
Sounds like you need Gemma

>>742286536
Try and get the LLM to focus on it as erotica with emotional draw, interspersed with pornographic physicality or something along those lines.

>>742305916
>my vore/furry RPs are being used to build the most expensive computer programs of our time
wack

>>742305916
I hope so, the world will be a much more interesting place

>>742306346
Chub is less filtered, Janny has more cards.

>>742306916
And I take it you're free to create your own characters? I don't really care for characters that I didn't come up with myself.

>>742303024
>>742286536
>>742296568
I usually skip the actual sex, too. I only write up until everyone is naked and raring to go, then write a short synopsis of what they did, then leap forward an hour, or to the morning after.

>>742306991
Yeah, Chub has the better card creator.

>>742306001
LLM's are basically just powerful search engines - interactive encyclopedias.

You shouldn't think, "this can write a program for me," you should think, "This can tell me how to write the program I want".

Even if the LLM can generate functions for you, it can't handle the entire program. You have to understand what you want and need, the LLM only ever provides reference or narrow construction at its best. You have to keep in mind the overall picture yourself, the LLM cannot.

So you can't just 'ask an llm to code that'; the person has to be capable of coding it themselves, the LLM can only be a useful guide.

And if they don't understand how an LLM works in the first place, the odds of them coding an interface for an LLM so that it can call a dice roll function or input results into the interface for the mechanical system to handle and then output the results in a human language/narrative to the players, aren't looking too good for them.

>>742305968
Anyone who uses AI extensively of course understands those are real problems and AI is both simultaneously great at everything and also dogshit at everything.

However thats exactly the problem I have had some success tackling with my program
>>742291787

The program handles the dice rolls, the hundred different tables for all the systems, the LLM handles the writing.

Of course its not perfect, it will fuck up once in a while because thats the nature of LLMs but once that happens the program detects it and stops the adventure then you press a button to export your character and adventure to an excel file which you can paste into a new chat and continue on. There will always be some degree of hand holding it and manual corrections but thats true in actual table top role playing as well so same difference.

You guys act like these arent known problems and there arent solutions for them.

>>742307529
I'm just cynical and with limited information; someone saying they're 'gonna run a roleplaying system on chatgpt' sounds like a crazy guy that doesn't know what he's jumping into.

It's not that I don't think it can be done, but I feel like I'm hearing a kid talk about how he's going to learn to draw and keeps mentioning it instead of just drawing and letting the results speak for themselves.

>>742307339
>>742307732

Dude I think you missed the part where I already made a program that does just that.
Yes its barebones which is why Im not sharing the program but its already good enough to prove the concept works.

Making a local random number generator to handle dice rolls is one of the first things they teach you in any programming class at university. This shit isnt that complicated. Making assets for a videogame is far harder than programming a shitload of booleans for an entirely text based system.

Heres a test adventure I ran to prove it works.

"Im an elf, a wizard casts sleep on me" chatgpt then queries my locally running program asking it for the rulebook hard coded in the program locally and it responds to chatgpt saying "beepboop elves are immune to sleep" and then chatgpt narrates "they cast sleep but it fucking fails because youre an elf and they are immune to magic sleep."

And I do know how to program but I still had chatgpt write 90% of the programming for it because its a better programmer than I am. It wasnt a simple matter of just writing one prompt "write me this program" but thats not that far off. So that too is proof youre judging something you dont understand the capabilities of.

I get hating AI for stealing the human spirit blablabla but youre not living in reality. AI is just a tool, it can be used well or poorly, thats up to you.

>>742308684
Mate you're arguing with a retard. Save yourself the trouble.
Also how much of shit is in context? Because finding the balance between everything being in there like in ST, and everything being agentic for just the single task it has to do, sounds pretty tough. Especially if you at all care about cache hits

Bunch of you newfriends don't understand what LLMs are really like. Whenever you 'wish' for something just formulate your sentence like this: 'I wish autocorrect could do {{something}}'. All will be clear.

>>742308849
Im still experimenting with that and yeah its a problem but I've had surprising success keeping minimal things in context where it focuses on whatever is the current task at hand and not a million things, its just a problem over time when things start piling up but then you just start a "new adventure" thats really just a fresh version of your old adventure and youre good again. My program hasnt gotten far enough to really stress test it, having it keep track of many characters, locations and systems but we'll see.

I find it funny that LLMs struggle so hard with things humans were doing in their head on paper so easily since the 1980s

>Owl Alpha gone
Fuck me, I learned about it like 4 days ago.
I'm willing to pay, I just want to know one thing, is deepseek v4 flash the best cost benefit one or is there something better? input /M and output /M means basically how much I spend per million tokens, how long do you guys take to get to a million tokens?

>>742304404
i need to chat with that dog he drew

File: pepper_my_favorite_slut_h(...).png (361 KB, 1036x1008)

361 KB PNG

>>742311404
I like chatting with Hellbean's dog.

>>742312431
i did too but she was too experienced with eating bones. so i went to something nicer like ninetales or renamon afterward

>>742310785
Just look at your stats. Or just put $10 into OR and realize it's going to last ages. And I think with official API prices real deepsneed isn't much more expensive than flash while being a lot better. Claude Sonnet 5 also has a discount running that makes it a lot more affordable, though still a lot more expensive than chink models. With sonnet I'd be running at $1/h with my setup if that is any help.
For token just google and paste in part of your story or any other text. If you have any other questions ask.

>>742256861
>AI roleplaying is the future of vidya.
Cool, I'll wait until it's good in the future.

grok, generate a harem story for me to bust a fat nut to
ez

I just have futapov sex with various mommy bots i let groom me

>>742313745
CONTENT MODERATED

LIMIT REACHED

>muh proxies
give openrouter $10 and you have goon material for months.
unless you use claude that shit jews your balance up fast

Alright that fantasy morgue brothel is good. Good call whatever anon recommended it

>>742314804
Card?

>>742313117
I use OR, I planned on using Flash because poorfag and the dollar is expensive here but I am willing to spend because this shit is too good.
Thank you anon, I'm gonna figure it out, just need to figure out which models are ok with cunny now.

>>742315157
https://botbooru.com/character/38529

>>742314639
It's just thirdies. They won't pay for anything.

>>742270486
Is there something I can follow to ensure I don't miss this if it comes out? I am highly interested in this feature set.

Computer, say 'pussy'.
PUSSY
: O

>>742257231
Start with Deepseek 3.2V

Get a preset from chub

Pick a card that you find hot AND has a lot of likes/uses

Double check the description to make sure it's not just some wiki copy paste.

Enjoy.


Once you enjoy take some time to actually quantity how much you enjoyed and then decide if you want to spend for Claude or at least V4 of Deepseek

They are both objectively better. Are they better with a price ratio? That I do not know

File: HG2Fzg4bUAAXN5S.jpg (577 KB, 1632x900)

577 KB JPG

>>742315417
Learn to cache then amigo. Pro is $0.0036/M input if you hit cache. It's a bit of work understanding how to structure stuff so it works, and then there are some snakeoils you can't use with it on, but holy shit the savings are insane.
As for the cunny good luck, but afaik the chinks don't really give too much of a shit about preventing simple jailbreaks, and for the others it varies, with gpt being tough, and claude being easy-ish. But I'm no cunnyseur so who knows

File: sw1.jpg (26 KB, 739x415)

26 KB JPG

>Claude, roleplay this
>constantly mischaracterized
>constantly say unrelated stuff
>constantly required your input help
Is that future coming any closer?

Alright how are you supposed to use this thing? What good models are out there

>>742317947
'emmy

>>742257231
This guide is good >>742137760
I've heard that Marinara Engine is trying to give people a more RPG-like experience but I haven't tried it.

>>742317541
Thanks brother, I'll look into it

File: Capture.png (96 KB, 860x245)

96 KB PNG

>>742292434
it's part of the story, bub.

Is jannyai kill?

>>742314804
even if i have some fukt up tastes i dont really get it
what is there to write about going in and out of an unresponsive corpse

upon further study gentlemen i have decided that approximately 80% honey nut cheerios to 20% sugar puffs feels correct. with an appropriate low level of milk they hold crunchiness long, and the cheerio base is pleasantly accentuated with the more upper nasal notes found in the sugar puffs so it's still predominantly cheerios but the little bit of sugar puffs there really brings it all together. some years back there was a noticeable drop in cheerio quality. i tried name brands, i tried store brands, i tried them all. i'm almost 100% positive they were whipping more air into them because they got soggy faster, they weren't as crunchy, and the flavor wasn't as pronounced especially since they soaked through with the goddamn milk so fast. sadly all cheerios now are still in that lessened state. just looking at them you can see the air bubbles in them. coin clippers. god damn CEREAL CLIPPERS

>>742319214
thank you for sharing

File: 1412650911397.jpg (75 KB, 640x640)

75 KB JPG

>>742317876
Not really it has a limit because it begins to relate words to the same values so the outputs are a mess
At the end is just fancy statistics not real intelligence

i just fucked a literal bunny

>>742319528
Elaborate.

>>742319565
she was just too lewd, officer
https://botbooru.com/character/6305

>>742319627
>This character does not exist
>It was either removed or deleted. Please return to the homepage.

>>742319683
https://chub.ai/characters/Bountyyy/hazel-a785d741
it's NSFL so you might have to use a vpn for this one, i know i have to that nowadays.

>>742319528
john you are the demons

>>742319528
What's the appeal?

>>742319786
carrot puns, duh

>>742319528
okay? i had sex with a moth-sized moth once. you're not special.

>>742319786
What's up, doc?

>>742319882
How? Shrunk yourself?

>>742319929
micropenis, maybe

>>742319528
>>742319882
I got throated and coiled by a ball python.

>>742319980
voreslop doesn't count

>>742320013
It wasn't vore.