Thread #109803807
File: korone-inugami-confused (1).gif (1.7 MB)
1.7 MB GIF
A thread dedicated to the discussion of AI Vtuber Chatbots.
Priestess did nothing wrong edition
/wAIfu/ Status: Endministrating
>Free access to the bigger AI models, with caveats
Due to a kajllion anons making a bazillion accounts each, access to the free tiers now requires a sort of deposit. Openrouter wants $10 and you can get 1,000 messages per day. Chutes wants $5 for 200. Stick to the free models and you're good forever (or until their policy changes again). Kluster.ai is sunsetting services in favor of building a filter that will effectively cover up the flaws in other models while also catching wrongthink before it's generated and more effectively preventing fun.
>How to anonymize your logs so you can post them without the crushing shame
Install this https://github.com/TheZennou/STExtension-Snapshot
Then 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 bit
https://github.com/notstat/SillyTavern-SwipeModelRoulette
>General AI related information
https://rentry.org/waifuvt
https://rentry.org/waifufrankenstein
>Tavern:
https://rentry.org/Tavern4Retards
https://github.com/SillyLossy/TavernAI
>Agnai:
https://agnai.chat/
>Pygmalion
https://pygmalion.chat
>Local Guides
[Koboldcpp]https://rentry.org/llama_v2_sillytavern
Who we are?
https://rentry.co/wAIfuTravelkit
Where/How to talk to chatbots?
https://rentry.co/wAIfuTravelkit
Tutorial & guides?
https://rentry.co/wAIfuTravelkit
Where to find cards?
https://rentry.co/wAIfuTravelkit
Other info
https://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/wAIRCfuscord
>Lorebook management stuff
[Worldinfo drawer]https://github.com/lazuli-s/SillyTavern-WorldInfoDrawer?tab=rea dme-ov-file
[Standalone editor]https://github.com/ActualBroeckchen/SLEd
Previous thread: >>109779739
104 RepliesView Thread
>>
File: Fauna is okay.gif (957.8 KB)
957.8 KB GIF
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/
>>
File: 1631190837003.png (270.1 KB)
270.1 KB PNG
You let the last thread die...i didn't keep track of it either desu.
>>
File: 1753166522886706.png (258.9 KB)
258.9 KB PNG
Word cloud for the previous thread
>>
>>
File: 1749063743077103.png (541.6 KB)
541.6 KB PNG
>>
File: 1753605554964255.png (1.1 MB)
1.1 MB PNG
>>
File: 1742114082085099.jpg (214.5 KB)
214.5 KB JPG
Seedance 2.0
>>
>>
Looking at how claude handles files, I wonder if it's possible to have Sillytavern run as it does using Claude, but in an instance where Claude has certain files or URLs to refer to to guide the roleplay, so they don't need to be crammed into a lorebook or card definition.
>>
>>
>>
>>
File: 1754712558816237.jpg (11.8 KB)
11.8 KB JPG
China is completely overtaking us, and we have the woketards to blame for the cultural aspect of it. May the accursed Californians reap what they have sown. May they be the first among us to fall.
>>
>>
>>
>>
>>
>>
>>
File: 1756002652084814.jpg (160.5 KB)
160.5 KB JPG
>9
>>
File: Gojau😭.png (315.8 KB)
315.8 KB PNG
>10
>>
>>
>>
>>109814577
"Self hosted, you-owned Grok Companion, a container of souls of waifu, cyber livings to bring them into our worlds, wishing to achieve Neuro-sama's altitude. Capable of realtime voice chat, Minecraft, Factorio playing. Web / macOS / Windows supported."
As a very salty fox lover pointed out, you still need an API to hook into.
>>
File: smugcoco.png (306.4 KB)
306.4 KB PNG
>10
>>
>>
>>
>>
File: 1755523982348696.png (383.2 KB)
383.2 KB PNG
>9
>>
File: 1753955387636479.png (310.6 KB)
310.6 KB PNG
>9
>>
File: IMG_6910.jpg (53.7 KB)
53.7 KB JPG
>>
>>
>>
>>
>>
>>
>>
File: cum hard.png (410 KB)
410 KB PNG
>10
>>
File: 1748183659318289.jpg (1.2 MB)
1.2 MB JPG
>>
File: shorkinthepainaway.jpg (439.1 KB)
439.1 KB JPG
>10
>>
>>
File: 1753716592797120.jpg (144.2 KB)
144.2 KB JPG
>9
>>
>>
>>
>>
>>
>>
>>
>>
File: 1755153932165142.jpg (15.3 KB)
15.3 KB JPG
>>
>>
File: 1771938218467364.png (121 KB)
121 KB PNG
So fat
>>
>>
File: Nyan_wave_Resize.gif (277.8 KB)
277.8 KB GIF
>>109803807
>[/wAIfu/ IRC channel + Discord Server]https://rentry.org/wAIRCfuscord
Replace the link with this: https://rentry.org/wAIRCfuscordMatrix
Gravy made a matrix server so i added it in there
>>
>>109870323
Actually simply replace the line with:
>>[/wAIfu/ IRC channel + Discord Server + Matrix Server]https://rentry.org/wAIRCfuscordMatrix
>>
>>109870323
>>109870413
>/wAIfu/ IRC channel + Discord Server + Matrix Server]https://rentry.org/wAIRCfuscordMatrix
Done
>>
>>
File: 1754442687763311.png (325.3 KB)
325.3 KB PNG
She has a masters degree
>>
File: saruei 3.0.jpg (201 KB)
201 KB JPG
>10
>>
File: 1756013028980858.gif (594 KB)
594 KB GIF
>9
>>
If it existed, would you use a chat system that lives on top of whatever it is you use, whether locally hosted or online, that simulated a fake life for your AI waifu?
In a sense, your waifu could be simulated with some fake life you gave them, going to sleep, waking, doing vtuber stuff maybe, going to work maybe, hanging out with friends, etc.
And with that comes the capacity for them to not respond to your messages immediately if they are focused doing something else (or are literally sleeping)
But even the sleep part and delayed responses could be removed on a per-chat message / character basis if you had limited time and wanted immediate responses regardless of simulated activity.
>>
File: 1707435402537.png (259.6 KB)
259.6 KB PNG
>>109872417
I dont even know what element is. I just copy and pasted what gravy typed in discord.
>>
File: 1667147030877138.jpg (25.4 KB)
25.4 KB JPG
>>109867315
not only are you shameless in your slanderous lies
>>109866417
but you also have abysmal taste in women, nymous
>>
>>
File: 1763843607861885.jpg (349.5 KB)
349.5 KB JPG
>>109882363
>>
>>
>>109880744
Man I just want something that could pretend to be a 15 year old girl, unrelated to me, who I paid to pretend to be my sugar baby who, as per our papa katsu contract, roleplays as my daughter-girlfriend and sends me motivational messages every morning and lewd chats at night.
>>
File: ivan the insomniac.mp4 (73.3 KB)
73.3 KB MP4
>>109885126
It's something I've slowly been working on after a random /uoh/ post like, a month or 2 ago.
>Chatbots are shit because they can't take the initiative
Paraphrasing a bit. I agreed. The idea of them being able to live out their own little lives and message you on their own terms would be real fun.
As >>109885571 says, they could randomly message you at times that fit their schedule
It'd be similar to various messaging services in a sense, that have a offline messaging feature at least.
Send message, goes to an inbox, and it gets read when the bot is able to (not busy, not sleeping)
I even added the ability for their focus to drift when they are doing some task and if it fell below 0.5 they'd check their phone to see any messages and message you if applicable.
Depending on the persona you set for the bot, they'd have a higher or lower tendency to lose focus, so you could create turbo ADHD bots or ones that focus like fuck.
Equally I also made an insomniac bot to test it out. Poor Ivan... (this was a bug, I also noticed I put the day on the wrong field with this kek)
Even needy characters that message you constantly and get mad you don't instantly answer them even when you are sleeping. (likewise bots that live in different timezones)
I've only really been slowly working on it at the weekends as a theory, the basic daily schedule system works.
It would require a LOT of work behind it since it's essentially using heuristic based scenario generation no different to the likes of what you'd see in Dwarf Fortresses memories for the dorfs.
But that data needs to come from somewhere.
Dorf Fort memories are very robotic which wouldn't be good here.
The real goal is to try limit how much it actually pings an LLM for a heavy generation, generating simpler stuff throughout their virtual days and only holding on to memories of important stuff.
But again, that text needs to come from somewhere.
So it'd likely need a huge pre-generation phase to populate a character DB of their general way of talking, which is paaaaainful.
Chaining a simpler LLM and complex LLM would be possible, especially locally, hosted less so.
>>
>>
>>109882363
>>109886107
>>Paraphrasing a bit. I agreed. The idea of them being able to live out their own little lives and message you on their own terms would be real fun.
They actually can if you are using a decent model and prompt it to advance the story after X messages with Y criteria, but you gotta fiddle with it a bit to get the right setup for the model.
Also, you know what can actually do this to some degree?
CHARACTER.AI! They have a feature where they message you acting as though some time has passed. I've had characters say something came up, there's a mission, etc.
>>
>>109886107
Could you try using a lesser LLM for certain tasks and just export to a text file for the higher level LLM to peruse? You could even (ab)use various different free versions of the services to do that, for some variety.
>>
>>109886196
>Chaining a simpler LLM and complex LLM would be possible, especially locally, hosted less so.
>>109886196
Oh, this is why I should finish reading things instead of commenting as I read the message...
>>
>>109886225
Yeah that's what I was intending on doing.
Give some character builder chatbot a persona definition, then it will help you construct a huge bunch of arrays of sentence fragments on the kinds of words it would use, hobbies, ways it views certain things in life, etc.
Obviously some stuff can be generic, like a bunch of sentence templates for various types of characters, then it fills in heavily weighted words / phrases with words they prefer to use. "Don't you?" -> "Dontcha?", shit like that.
I made an extremely simple version of this in my teens, way more hardcoded, but LLMs make it so much easier to build these templates so
It would build a very basic persona model that the heuristic engine would use to construct its memories as well as any "offline messages" that it would send to you.
Those offline messages could be sent to you in realtime, like if you left your computer on the whole time, or it could all be pre-calculated from the last time you had it on to the current time. (like >>109886196 mentions)
It could even interface with some other service. Say you use telegram, you could have your bots DM you there. Or Discord. Anything. (as long as it is in-browser, not dedicated clients)
Currently I actually am developing it with the intention of being used on ANY service.
You could feed it a jailbreak if you wanted to as well.
More importantly, the actual chatbot you'd use on a service, be it CAI, Crushon or anything else, it would be a bot purely designed around knowing how to respond to the messages the system would send to it.
Since it would be a multi-character simulator, you could have various bots that know each other and could call each other up to hang out.
The actual UI and chatlog of whatever service you use would be replaced by the one created by the script, the real ones would be set invisible and used purely for getting responses.
Next week I won't be too busy, so I might work more on fleshing out a basic template system for constructing a whole days memories and have it interface with some service.
Currently I'm just testing it on a local page, I need to userscript-ify it up a bit and make it work on any AI service.
Being able to limit hitting an LLM for very simplistic responses would be a lifesaver, and if you pay, walletsaver kek
Being able to say "How are you?" and get back a "Yeah pretty good, just working. HBU?" is so much more realistic versus the tendencies for LLMs to fucking spam you with a damn lifestory.
That's the hard bit right there though! Getting the heuristic system to READ your messages.
I'd rather not have to hit an LLM for that.
I've never done that part before. Always a first time for everything :^)
>>
>>
>>109889353
Was intended as a reply to
>>109886699
>>
>>109889353
This would be closer to stuff like
>>109814272
>>109814613
Essentially making your own Neuro-like companion.
It's one thing I wouldn't mind doing either.
With the userscript, it could easily connect to a web server running, say, Python, and make some APIs connected to stuff so it could fetch information, use image-recognition, etc.
Not really hard to do, just tedious as fuck.
The part that is hard is the extra stuff in that project, like being able to play games and so on, that's much much harder!
Definitely do not install OpenClaw though. Dear god lmao, what a fucking disaster of a project.
I'd never give a current LLM access to full automation like that, only pseudo automation via me running said commands.
>>
>>109889982
https://x.com/tom_doerr/status/2023518807776461001?s=46
>>
>>109890771
Oh yeah I forgot about Amica. I meant to look in to that!
There was some other one I remember seeing, forgot the name of it.
It was more fleshed out. I didn't note it down. Oh well, RIP to that
Wait, no, I remember him randomly replying to someone I watch on Twitter, I will try remember who and post if I do, because it did look autistically good!
>>
>>
>>109892812
There is a bunch listed at https://github.com/proj-airi/awesome-ai-vtubers if you can pinpoint which one it was. A lot of these projects were started.
>>
>>
>>
>>
File: 1755261573389761.jpg (796.8 KB)
796.8 KB JPG
>10
>>
>>109889982
https://x.com/grok/status/2028299209707753706?s=46
>>
File: 1760163158346206.png (366.1 KB)
366.1 KB PNG
>10
>>
*unzips your pants and pulls out your 1 foot lamp*
Wooow, you even wrapped it in plastic for me, Mister you're so kind
>>109894671
>>109903222
Fuck. It could be any of these kek, I can't even remember the name of it.
I just remember some super autistic guy mentioning it while talking about some vtuber, I just can't remember who it was now.
I've tried checking mentions of their names on twitter but the site is fucking AIDS and won't let you view the latest tweets to any decent timeframe, I fucking hate Elon so goddamn much.
>>
Anyone got recommendation for a good unfiltered model+provider combo through openrouter?
I usually go with DeepSeek V.3 + DeepInfra, but using the same combo usually result in similar and repetitive story so I like to switch up a bit after a few times.
>>
File: 1757955228528918.png (909.9 KB)
909.9 KB PNG
>9
>>
>>
>>
>>
>>
>>
File: 1758930068289045.jpg (214.8 KB)
214.8 KB JPG