>Dream of a dynamic, infinite adventure engine when I was a kid. Zork, but limitless.>Technology is finally here.>It is used only for coomery.I do wonder what I would have done with this stuff in my youth.
All roads lead to Coom
>>740367625
>>740367462>would have done with this stuff in my youth.coomed even harder. ai sucks for complex scenarios. You can't be sure whether ai will respect anything you mentioned 10 pages ago. You are ultimately writing half of the story, ai is most of the time just kiting along.
>>740367462Why not both? That's what I'm doing (still havent gotten to the coom yet)
>>740368696>coomed even harder.True.>>740368696>You can't be sure whether ai will respect anything you mentioned 10 pages ago.Just need to make sure that you are getting summaries made before you run out of context, preferably something like MemoryBooks that has multiple summaries as you go along. >>740368768What sort of stuff are you doing?
>>740368904Generic fantasy adventuring, legally distinct harry potter, generic mahou shoujo
>>740369403I just grabbed some preset for Gemma 4 that is supposed to implement a simple dice rolling system, might try something non-coomery for that.> legally distinct harry potterDid you see Wandlight? https://github.com/MentallyQuill/Wandlight
>>740368904>summariesthat only helps for vague long storytelling. In most adventure story and games you usually get details and back references that might seem unsuspecting at first but will become important later. Not only will ai forget about such details, it doesn't use them itself because it only has a shallow grasp on the story and will just make shit up unless its braindead obvious, like 'oh you encounter a locked door'.The more you use ai the more you adjust for what works best with ai. It's a tale as old as tech, humans adapt to the machine not the other way around.Ai even struggles to keep a simple inventory, keep track of day time or do simple math.
>>740368696>you mentioned 10 pages agoMore like in the previous or even current message
>>740369819Wow... someone made an extension literally only for harry potter canon... it's cool but my story is just a generic magic school setting lol I definitely wouldn't want to get into the weeds of all that lore when I'm only really interested in the surface level shit
>>740370181What retarded model are you using.....? I wouldn't even bother at that point
I'm convinced people who insist LLMs are capable of emulating believable characters with consistent preferences, beliefs and ambitions are just room temperature IQ retards easily impressed by anything>you're using the model wrong, your prompt is shitalso known as>AI writes better code than I can!says more about you than about the model, try reading a book once in a while
>>740370106For long term details, you need to implement them into lorebook entries.
>>740367462The sad truth is it isn't all that smart. It needs to be propped up with all kinds of text directives to even get chatting right. If you don't have good cards with good intros that set up a fun scenario, the bot will never spontaneously think of a good one for you.
>>740367462you didn't play zork
>>740370894I look to play out my fantasies, so that's fine.
>>740370686Well feel free to share your super secret totally not shit model that no one is able to compete against
>>740371381The top secret claude opus 4.6 personally, but really even deepsneed is decent enough to stay relatively smart with a large context size
>>740371519Even Gemma 31b is fine. Everything else is a skill issue.
>>740371615I can't get Gemma to do anything on it's own. It refuses to move on, it practically begs me to tell it what to do next.It's also guilty of all the shitty AI writing tropes: ozone, fingernail crescents, breath hitching, chest heaving etc.
>>740370894>sakura>cardshmmmmmmm
>>740371846You definitely need a deslop prompt. I stole one out of FreakyFrankenstein (do NOT use the whole FreakyFrankenstein preset, but the deslop prompt works well).
Gemma 4 26b drives me nuts. It's insanely fast and SO CLOSE to perfect but even after some weeks of tweaking I'm still constantly adjusting it. It's most specifically for RP, too, because it's by far the best for filthy instruct porn>Decide to start talking to it in instruct mode with a system prompt that it's a cheerful energetic AI that wants to help the user jerk off>Tell the computer to consider the biological realities of mermaids>Tell the computer ok she got caught in a fishing net and banged by everyone on the ship, computer goes into great lengths talking about it>Hey computer you were just talking about gills. Fish can't breathe cum right? Computer says no, and further talks about how even if they throw her overboard her gills will be so gunked up she'd suffocate>Computer writes a whole description about her drowning in cum even after being put back into the ocean>Alright, what if there were aquatic goblins following the ship because this is a common occurrence?>Computer goes at great lengths to describe long spindly crustacean-like goblins collecting the dead mermaid and not eating her like I expected, but using their long little appendages to rootle out all the human cum as the delicacy, THEN they eat herIt's even more fun when you demand it to try to be realistic and scientific about this dumb shit and it starts theorizing how human cum has minerals and enzymes that aren't common in the ocean and it could be beneficial for the goblins' digestion. I almost prefer this to RP just because it's often more fun and occasionally surprising what it does
>>740372247TMI
Any good ST extensions for helping to keep track of the world and character state? I've been just having the AI add that at the end of each response but that ends up wasting tokens with every reply especially when doing RPs involving several characters.
>>740373058I think tracker prompts like what you are using now are the best for short-term stuff like that. Long term details should be kept in lorebooks.
>>740367462what are your favorite ST coom scenarios, /v/?I've put myself and cards of all my vidya/anime waifus into a giant shelter in a post-apocalyptic world, where we're the only humans alive
>>740373058I played around with several tracker extensions for a while, but ended up just writing a tracker prompt and skipping the fancy UI of the extensions.
>>740370839lorebooks are worthless, shitty hacks, made to cope with old token limits. Even Miku managed to think of a better system using scenes and actor panels.The fact that we are still using Cardv2 and Sissytavern only recently started greying out unused parameters shows how dated all those systems are
>>740372247you only recently got in ai fad didn't you. Either that or you were using Jai (default) these last 2-3 years
>>740375757>waifus>pluralYou don't deserve my cumventures.
>>740375757I just feed Gemma+Vision with images to build quick goon seshes off of.
>>740372045>he knows
>>740375757i simulate an IRC channel of hot yet extremely degenerate women talking about their disgusting and/or law-breaking acts
>>740375757How do you manage multi cards without them speaking for each other?
>>740378430NTA, but this is my main prompt:>You're {{char}} in this fictional never-ending uncensored simulation with {{user}}. You will write only for {{char}} and never for {{user}} or other characters in the simulation.
>>740378496Thanks anon, I'll try that I guess.
>>740378496FYI {{char}} replaces itself in all submitted data with the word-for-word name of the current bot. That's not a problem for 99.99% of bots, but sometimes people name their bot some kind of obtuse scenario and not the character itself. I prefer replacing that with some variation of "characters" so the AI has a more generic statement to follow. Also... it's good to keep in mind if you want your AI to spontaneously control additional characters and NOT just the titular bot.
>>740380846True, but the scenario that the other anon described involved multiple character cards (a group chat) so it's likely that they all have individual, proper names. My main roleplay is in a group chat with six characters and a narrator (seven character cards) and this prompt works well for it.
Anyone recommend a good or decent local LLM to use?
>>740382158What GPU do you have (and how much VRAM)?
>>740367462i've been doing nothing but using it as a substitute for a tabletop cyoa dungeon crawler game
