>>14098 I only use ST at this point. Used to use tavern ai but it was too simple. ST has all latest features and is much better overall, is a nice UI too. Plus it has way more control over local models. How is Kayra or whatever that 13B model from NAI? Any good? https://venus.chub.ai/characters/BookWormXtreme/lavender-80ef1adc This is one of my cards, with my venus profile having more. I don't have all of them uploaded, as a lot I do is take interesting ideas from CAI and export them using CAI tools plugin, then improve and modify till I'm happy.

I mostly use gpt4 and evade bans so I’m interested in setting this up on my machine. Did you follow any guides by chance?

>>14142 I'd say it's on par with CAI and GPT3.5 for the purposes of roleplaying. It hallucinates more often than GPT3.5, but it's also less "wooden" with it's responses. The other benefit over GPT is the context size is double (on the top tier plan), and it's a single monthly price so you don't have to worry about how many times you're regenerating/swiping. It does take a little bit of tweaking to get it to behave as a chat bot, I used these presets as a starting point: https://www.reddit.com/r/SillyTavernAI/comments/16ihh2v/comment/k0npx3i/?utm_source=share&utm_medium=web2x&context=3

>>14165 Do you mean using GPT4 on the OpenAI site? If you mean the API, then that must be costing you a fortune! This is a really good getting started guide if you've never used SillyTavern before. https://www.reddit.com/r/CharacterAi_NSFW/comments/14lat1f/sillytavern_the_nerd_guide/?utm_source=share&utm_medium=web2x&context=3

I am a bit of a SillyTavern enthusiast and I also like poke fat bitches... yeah just leaving these characters https://www.chub.ai/characters/notaxel12/hisoka-f1a55f30 https://www.chub.ai/characters/notaxel12/amara-f785baef these are the only fat girls bots that I have on my profile lol, whatever, in SillyTavern you can modify any character, feel free to modify them and add Slob or change their Sizes if you want

I used to mess with T.ai before ST became a thing, I recall some really good plays coming from it. Tried to keep up, but after the release of ST, I just haven't been able to find a reliable proxy to run with. It is kind of a shame, because I was working on a bunch of cards all the while. Mostly stuff I thought was missing on other sites. Not that I can really do that at this point, though, less so when I can't even go and test them out.

>>17541 You were using Proxys? of what model? GPT? Claude? if you know how to set up the configs and what is the prefill maybe I can find a proxy for you

>>17541 The one I used most recently was Ooba, but back when I was using Tavern.ai proper, I used Kobold. Nowadays, I don't particularly care. As long as it's reliable enough and can test out the cards, I'm willing to settle for about anything. And speaking of the cards, just in case anyone thinks I'm fibbing. Mostly done, but still needing refinement.

>>17547 >>17573 Holy misquote.

>>17573 Oh yeah, local models pretty much anything works for testing, if you only want to do that, the important part of making characters is making an organized prompt the character card you uploaded is organized, is a good job in that way, and if your character card is well done, that means that it will work pretty good in most IA models oh yeah, the "thinking" box of some characters, that thing can be very inconsistent, even in powerful models, so it can just appear and dissapear between messages, it's not certainly a "problem" because there isn't a real solving to that, but just keep that in mind when you are doing tests, don't try to force it and make the character card an instruction of the think box, just let it be

>>17591 >pretty much anything works for testing I don't doubt it, the problem is I don't have much of anything. The Ooba proxy I was using went offline months ago, and none of the stuff I was messing around with during the T.ai days seems to even apply anymore. The thinking box is something I was actually thinking to remove. I saw it in some other characters and figured it'd be really neat, but it proved to be nothing but inconsistent when I tried it, and the few times it actually popped up, the prompt generally wasn't better than it would've been without it. Of course, the proxy went offline before I could actually get to that, so there it has stayed.

>>17593 there is this thing... it's called mythomax or something, I think it's a model that you can execute in google colab, I heard that is a good model but I haven't tried it, you can try to find some more info about it and see if it works for you

>>17622 I feel like I've heard the name come up once or twice, reckon it should work well enough. Any pointers as to documentation I should start with? A git, repo, rentry, anything along those lines? While not always true, going into a new model raw is a recipe for painkillers down the line.

>>17626 Found this: https://colab.research.google.com/drive/1ZsRJCH4H6ZNlNoU3AMngR8MHmuZnQu2T?usp=sharing&fbclid=IwAR2-zdC0OjbWeUu2qoLGjxineIvamVqOuLtTpV7d0h5X1h4q0Yvc2Sec4NQ#scrollTo=LGQ8BiMuXMDG it seems it has more models than mythomax, and also comes with some kind of guide, give it a try

>>17630 Checks out. I'll get back to work and see what other cards I can post in due time. I've got a couple things in the works. Probably a couple more to add into the list. The priority is going to be a drunk, out of shape Yoruichi in heat, though. I've been owing someone that card for a few months now. I'll share it here too, though.

Any of you have very good fat bots for silly tavern And some tips for creating bots?

Axolot