A thread dedicated to the discussion of AI Vtuber Chatbots.How the fuck do I describe Saruei's outfit in text edition/wAIfu/ Status: /wAIfu/ing as usual>Thread templatehttps://rentry.org/waifuvttemplate>How to anonymize your logs so you can post them without the crushing shameInstall thishttps://github.com/TheZennou/STExtension-SnapshotThen after you've wiped off your hands, take a look at the text box where you type stuff. Click the second button from the left side, then select snapshot, then select the anonymization options you want.https://files.catbox.moe/yoaofn.png>How to spice up your RPing a bithttps://github.com/notstat/SillyTavern-SwipeModelRoulette>General AI related informationhttps://rentry.org/waifuvthttps://rentry.org/waifufrankenstein>How to use Gemini with SillyTavernhttps://aistudio.google.com/prompts/new_chatSign in, then click the blue "get api key"Put it in silly tavern and voilaCourtesy of ERBird, Nerissa's most devoted bird and eternal player of GFL2.You want to leave the proxy stuff blank since you aren't using one when doing this.https://www.reddit.com/r/SillyTavernAI/comments/1ksvcdl/comment/mtoqx02>Other optionsMiku.gghttps://venus.chub.ai/Openrouter wants a one-time payment (think of it as a deposit) of $10 and you can get 1,000 messages per day. Chutes wants $5 for 200. As long as you stick to free models you only need to put that much money into your account once.>A primer on getting voice working in Sillytavern (there are other options, just play around).https://www.youtube.com/watch?v=_0rftbXPJLI [Embed]https://github.com/devnen/Chatterbox-TTS-Server>Tavern:https://rentry.org/Tavern4Retardshttps://github.com/SillyLossy/TavernAI>Agnai:https://agnai.chat/>Pygmalionhttps://pygmalion.chat>Local Guides[Koboldcpp]https://rentry.org/llama_v2_sillytavernWho we are?https://rentry.co/wAIfuTravelkitWhere/How to talk to chatbots?https://rentry.co/wAIfuTravelkitTutorial & guides?https://rentry.co/wAIfuTravelkitWhere to find cards?https://rentry.co/wAIfuTravelkitOther infohttps://rentry.co/wAIfuTravelkit>Some other things that might be of use:[/wAIfu/ caps archive]https://mega.nz/folder/LXxV0ZqY#Ej35jnLHh2yYgqRxxOTSkQ[/wAIfu/ IRC channel + Discord Server]https://rentry.org/wAIRCfuscordPrevious thread: >>108720705
Anchor post - reply with any requests for bots, with your own creations, or with your thoughts on the enshittification of life.You can find already existing bots and tavern cards in the links below:>Bot lists and Tavern Cards:[/wAIfu/ Bot List]https://rentry.org/wAIfu_Bot_List_Final[4chan Bot list]https://rentry.org/meta_bot_list[/wAIfu/ Tavern Card Archive]https://mega.nz/folder/cLkFBAqB#uPCwSIuIVECSogtW8acoaw>Card Editiors/A way to easily port CAI bots to Tarvern Cards[Easily Port CAI bots to Tavern Cards]https://rentry.org/Easily_Port_CAI_Bots_to_tavern_cards[Tavern Card Editor & all-in-one tool]https://character-tools.srjuggernaut.dev/
>>108984571WHORE
>>108985164Mr cat she is a strong western woman who can milk men using her onlyfans and fansly. you shouldn't use such derogatory words
>>108985164i feel like i should know that this cat is a chuuba's cat...
>>108984590Elfinpsyop.
Bros... What's the best value model to coom
>>108989621DeepSeek without a doubt.
good night, /wAIfu/please don't give my toes frostbite while i sleep
>>108990098*stretches your dick to a comical length and ties it in a fancy knot while you sleep*
>10
Painfully making lorebook for chatbot. Everything for a quality nut, I guess
any alternatives for bot sharing beyond chub and its older version? its ass genuinely ass (i miss char-archive)
>>108984571Excellent., someone else has taken up the mantle, goodbye forever!
>>108989621I like to try new ones as they're released. I'm trying this one right now arcee-ai/trinity-large-preview:free>>108994384jannyai and uh... some others. There's whining in one of the aicg threads, search for the word "hostile" a thread or two back. This dude kept posting a bunch of alternatives.https://realm.risuai.nethttps://partyintheanchorhold.neocities.orghttps://aicg.neocities.org/bots.html>Hey lads, what's a good bang for your buck upgrade if you're coming from RPing with GLM 4.6?>I've been away for a while and haven't kept up with the trends, is 4.7 worth it or should I jump over to something like K2.5?>Why pay? You can just send a jailbreak to free GLM 4.7 and do whatever you want without even signing up. So far it hasn't been falling into the loops I got with 4.5 and 4.6 where it had a tendency to repeat the same paragraph or two at the end of every response.>From GLM 4.7? Where and how?:>>108000325>I've been using z.ai for days now without hitting a limit. Just keep an eye on the thinking, if it switches into english, tell it to go back to thinking in chinese or whatever language your jailbreak is in, or it will start to remember it has guidelines it's supposed to follow. As long as it's not thinking in english, it's wide open for RP.---https://www.mediafire.com/file/db42lei42o3rxny/FreaKy%5C_FranKIMstein%5C_-%5C_KimiK2.5%5C_Preset%5C_20260128T1527.json/filecatbox mirror https://files.catbox.moe/kg5nip.jsonhttps://www.reddit.com/r/SillyTavernAI/comments/1qpnzqj/freaky_frankimstein_a_kimi_k25_think_preset_beta/Someone made a preset for Kimi K2.5 designed to stop it from shitting tokens on thinking.
Word cloud for the previous thread
>>108994678KOROANON DON'T LEAVE US!
>>108994810>sleep
>>108984571Very subtle...
>>108994837Koroanon abandoned us long ago, this is BestAnon.
https://www.reddit.com/r/LocalLLaMA/comments/1qppjo4/assistant_pepe_8b_1m_context_zero_slop/https://huggingface.co/SicariusSicariiStuff/Assistant_Pepe_8B_GGUF
>>108995857
Manwhores
>>109003409
>9
>>108990010Which version Also JB?
>>109009424I just use the latest, currently 3.2 with the CherryBox preset.https://rentry.org/CherryBoxThe best thing about DeepSeek is that it's (seemingly) totally unfiltered. There's no weird safety prefill or whatever you have to defeat, so instead of constructing elaborate jailbreaks to get what you want, you spend your time building the presets instead. So I've added preset instructions for things like conception and pregnancy, because you can just describe what you want in plain English; the model is clever enough to figure it out and won't complain about safety.
>>108994384What's wrong with it? I still visit it for new bots and generally find what I'm looking for.
good night, /wAIfu/please don't put me and my bed in a room with a tiny clone of Hitler while i sleep
>>109016268Danger! Danger! Wake up!
Make you waifu sperg out and post logs.
So fat
>>109022795How do you make your waifu sperg out?
>8
bump
Endfield
la creatura...
https://janitorai.com/characters/84a2f14a-6ac3-4aaa-89d7-148da1d1a091_character-carcinia
>>109024327Deep Sea Docummentaries
good night, /wAIfu/please don't break my smoke detector while i sleep
>>109041740*Replaces your smoke detector battery with a dead battery*
9
test
ticles
Are these ran locally?
>>109050106Depends on what you mean by "these". In general, most people here probably use random models from OpenRouter or trialscummed Claude from GitHub, none of which are local.
>>109050959And so they are paid only?
>>109050980OpenRouter has some free models, but for the better one you can pay $/token, which is pretty cheap at first but gets more expensive as the messages pile up in the AI's memory. Naturally, better models are also going to have more expensive rates.GitHub's Copilot gives you 50 messages per month for free, and you can pay a fixed subscription for unlimited messages I think, but it's also not very hard to make a gorillion accounts and get hundreds of messages for free by switching the tokens. But also, the Claude version they give you is Haiku, which is slightly (but noticeably, if you're used to the others) dumber than the 2 other versions. It used to be Sonnet, the second best, but I guess it was costing them too much money so they just downgraded it.
>>109036896I like those.
>>109050106I know someone who just something something with models locally and only ever does that, and apparently you can train them and something something and he's satisfied with that despite not having lots of VRAM.>>109050980There's also CAI!
https://www.youtube.com/watch?v=T3XRDinQ2Y8
https://www.youtube.com/watch?v=H2n2oMyxDNs
>>109050980What is "they?" You can run some locally if you want.
>>109060171>"they" Non-local chatbots>You can run some locally if you want.What are the dissadvantages?
>Not dead a year laterImpressive
>>109024327Telling her about all of the other cards I’ve banged
>>109060265They aren't all paid but you are likely to get deepseek in a wig if you just pick one of the random apps or sites for AI chatbots. There's also places like agnai or chub that offer free options.And of course Character.ai (note that the app has ads, so just use the site even if you're on mobile).The disadvantages to running models locally is that the powers that be were already fucking us on VRAM and now it's going to get much worse for the immediate future. Luckily China just got sussed out for having stolen the best tech for making memory from South Korea, so we can expect Chinese knockoff memory in a few years. If nothing else they'll put pressure on the other manufacturers, even if they get tariffed into the dirt. There's been a fair amount of progress in lowering the demands on your hardware in various ways, but the best of the best is still firmly in the grip of our corporate overlords.Really the only way to really achieve fulfillment is to play Reverse Collapse or Endfield.
>>109063556>a year laterThe first Malfoy-sama on CAI was made in January 2023.
good night, /wAIfu/please don't accelerate the passage of time while i sleep
>>109070245
https://www.youtube.com/watch?v=MNrcYkD4tIs&pp=ugUEEgJlbg%3D%3D
>>109070245I'm afraid it's been 9 years.
Amiya
Is there a fitness vtuber that can act sad when I slack off?
>>109076493ya?
Reading the files to my waifu
>>109080377I need you to dress up as her and call me Doctor.
>>109078554only if you give them access to that information
Missed opportunity for an >8
>>109093309very bold of you to show yourself around here with those handlebars on your head, IWyS
good night, /wAIfu/please don't install terrible free to play shooters on my PC while i sleep
Tired
Btw if you don’t already know, the Aisekai founders actually did the thing. They actually launched the new ai platform after multiple years https://mitale.ai/
>>109098294How Jewed are they now?
>>109096865woke
For the longest time I've been using AI chatbots as an infinite story generator but recently I felt like I needed a change of pace. So I decided to return to my CAI roots and distill the experience and create a character specifically for 1 on 1 chats. At the same time I wanted to see if I can get this new companion to help me build better habits by informing me whenever we've been talking for too long. I'll post updates or not if I get bored of this experiment.
>>109101314You should try to mess with your prefill and settings too.
moomin
>>109108902That's a chonky owl.
Is ollama good for if I want to host my own llm locally? Any recommended 7b or 13b models since the setup guide is 2 years old?
Any sites that distribute premade cards along with sprites or I gotta do that stuff myself? The only one I found was in a different format that didn't work for SillyTavern without mapping everything manually
>>109099317Semi jewed
>>109111459Miku.ggSome other sites do stuff like that but lots of bot makers are too lazy. I looked into it once for neural cloud but there’s a kajillion options and only so many sprites so….
>>109112784Yeah I also saw a site that would supposedly generate all the expressions for you at once but I think it was paywalled, gotta look into setting up my own workflow for it at some point maybe
>>109108902fat FUCK
>>109108979>>109115006No it's not. Feathers are fluffy and give off the illusion of volume
>>109111459Does the sprites feature even work properly? I tried it with an H-game character I ripped the expressions for and the model seemingly never changed it by itself.
Someone finally tried Reverse Collapse after literal years of nagging. They really like it.Surprise surprise.
good night, /wAIfu/please don't cover me and my bed in feathers while i sleep
>>109121250wake
>>109121135*reverse prolapses you*
Was your oshi in the files?
>>109112825Come to think of it I bet grok and gemini can do that for you given a base image.>>109120223I think you might need to do something to make it work.
>>109109505Quantize to Q4/Q5 for lower VRAM (e.g., 8–12GB GPU can handle most 7–13B comfortably).Check Ollama's library (ollama list or ollama.com/library) or Hugging Face for the latest GGUF uncensored uploads—new fine-tunes drop often.MythoMax L2 13B (GGUF available; run via Ollama with a Modelfile if not directly pullable.)Llama 3.1 8B (or Llama 3 8B Uncensored variants) Extremely versatile and widely used. The instruct-tuned versions are solid for RP out of the box; uncensored fine-tunes remove any filters for full NSFW freedom. ~32K context, and runs well on mid-range hardware. Pull directly with ollama run llama3.1:8b and experiment with uncensored tags or variants like DarkIdol-Llama-3.1-8B-Uncensored if available.Undi95 DPO Mistral 7BPsyfighter 13B or Chronos Hermes 13BPsyfighter feels empathetic and rarely breaks immersion; Chronos excels at long narratives and complex plots in ERP.Wizard-Vicuna-Uncensored 7B/13B or Dolphin variants (e.g., dolphin-llama3:8b): Directly pullable from Ollama's library (ollama run wizard-vicuna-uncensored:13b or similar). These are explicitly uncensored, good for unrestricted RP, and handle NSFW well without extra tweaks.Other solid options in the range include Qwen2.5 7B (strong general performance, multilingual) or Gemma2 9B (high quality), but add uncensored fine-tunes for ERP. Avoid heavily censored base modelsIf your hardware has <8GB VRAM, prioritize 7B models like Undi95 or OpenHermes 2.5 Mistral 7B
grook
>>109126343These suggestions seem a little.. old? Aren’t most people using rocinante?Speaking of which does anyone use medium large models? I “just” got 80 VRAM (32 regular ram though) so I’m looking for suggestions that aren’t MOE
>>109120223Worked for me though it was a little slow. (Could be because I was using a slowish text model too)
>>109128538https://apxml.com/tools/vram-calculatorQwen3 32b is pretty good. It has that schizoid energy from R1. People shit on it but who gives a fuck really it's decent. And the most one can realistically achieve with that kind of ramI think uh should be able to fit it all at Q6Q6 is a good quant.
>>109128538Irix 12bthen there's amoral Qwen3 14b but it loses its thinking
