Slightly-Less Barebones TavernAI Guide

How to Install - a 4 step guide

I'll be focusing this guide on how I personally installed it. There are other methods, but I can't help troubleshoot with them. I mean I probably can't help troubleshoot you with this either, but it's at least worth a shot. This is also aimed for a Windows using audience, but SillyTavern can also be installed on Mac, Linux and Android.
https://docs.sillytavern.app/installation/windows/ - Read here for alternate installation strategies.

Step 1.
Install NodeJS LTS and Git

Step 2.
Open an easy to find folder, such as your desktop or downloads folder, then type cmd in the address bar.

Step 3.
Once the command prompt has opened, copy and paste this string into the command prompt. git clone https://github.com/SillyTavern/SillyTavern -b release

Step 4.
Once that has finished, you can now close the command prompt and click Start inside the SillyTavern folder.

Extra.
You can update by either clicking UpdateAndStart or entering the command prompt again while in the SillyTavern folder and typing git pull

Connecting to a model.

While you now have Tavern open, you can not yet chat with any bots as you do not have a model connected.
There are two ways to connect an AI language model to Tavern,

1. Running a local model.
2. Connecting to an online model.
   This seems simple enough, but in these large categories there are many sub-categories. We will start with the second option first.

Remember that all online models are hosted on someone else's computer and your connection may not be entirely secure. Do not send sensitive information, such as your real name or location to these models. You may also want to use a VPN.

Horde

The easiest free way to connect to an online model is through Horde. Click the button that looks like a plug on the taskbar in Tavern. This will open up the API menu. In the top dropdown, click KoboldAI Horde. You can then choose a model from the list. These are all local models hosted by other Horde users. You can also optionally sign up to horde or rent out your own model, but we will not be discussing that in this guide.

Most of these will not be very good local models, and there may be long wait times. However you will almost always have access to them. After you have chosen a model, click the button that looks like multiple sliders. This is the API Response Config menu. In the dropdown, I usually default to Simple Proxy for Tavern. Then you should be ready to chat.

OpenRouter

This site gives you access to multiple different models where you pay for the amount you use, as compared to paying a flat fee per month. There are hundreds of models on the site, ranging from low quality to the very top of the line. Some, but not all of the models are moderated for content.
While most models are paid, there are some models that are free to use.

As of this Rentry update, Hermes 3 405B is no longer free but is still of decent quality and is mostly uncensored. I've also heard good things about Unslopnemo 13b which is much less expensive.

NovelAI

I haven't used this since a free trial years ago, but I do believe you can hook it up to SillyTavern. Their pricing plan is a flat monthly rate, starting at $10 and going up to $25 per month. Their models are completely unique to the site and you can't get them anywhere else. The text models are generally considered overpriced for the quality but their image generation model is often considered one of the best on the market. They advertise having an anti-censorship and privacy stance but you should always be careful anyway.

Also, as far as I know, they're the only AI company on this list that officially supports vtubers. They sponsored Pipkin Pippa once.

GPT and Claude API

Like OpenRouter you pay as you use, but they're specific to the GPT and Claude series of models and you may also get more features that you wouldn't get with OpenRouter. I believe the GPT API also gives you API access to DALL-E, which is an image generation model, for example. Do to my lack of personal experience I can't answer any questions about these ones.

Google AI Studio

Google's API is a bit different because you can use the API for free unlike GPT or Claude which only has free options within their frontend, I've heard the rate-limits are pretty bad though. Probably best to stock up on google accounts if you want to trial scum or do NSFW.

Proxies

If you ever see me posting with AIke, I'm probably using a proxy. Proxies are a way to allow multiple people to access one or more totally legally gained API key at once for free. These are usually GPT, Claude or Gemini API keys, but some proxies have other model types. Due to how quickly these can go down I can't link any specific ones, sorry. Some proxies are public, while others are invite only. You can usually find public proxies by looking in /aicg/ on /g/ but I wouldn't recommend spending too much time in that thread for the sake of your sanity.
If you see a seemingly random string of four words like for example "dog cute candy metal" then that's probably a link to a proxy hosted on cloudflare. You can "decode" the link like so: "https://dog-cute-candy-metal.trycloudflare.com/" This is just an example, it won't actually take you anywhere.

Local Models

More info on this later because while I have ran some models on my computer, my knowledge is very rudimentary because I can't into technology.
I've mainly been testing with LM Studio and Ollama recently. These are very simple to use for my dumb brain. Something you will notice with local models is that they'll typically have a number followed by a B in their name. Something like 8B, 13B, 70B, etc. Typically the higher the number, the better the model is but also more resource intensive they are to run.
For now guides like the \lmg\ Spoonfeed Guide are probably a better help than I can be with setting local models up.

Model Showdown

Open each image in an new tab so you can actually read them instead of them being the size of an ant. Note that this is not a rigorous test and does not accurately show the strengths and weaknesses of each model. There is no testing of such things as context length, which measures how much of the story it can remember, for example. The prompts and settings I used were also intended for GPT-4, and with different ones the other models could possibly shine a bit brighter. It's just a direct comparison of their responses in the same exact circumstances. (•ω•)

What model is the best really depends on what you want out of your AI but generally models like GPT, Claude and Gemini are much better then anything you can run on your computer. In my personal experience, GPT models are fairly smart, resistant to doing porn but good at it when you get there, and have writing styles that can be a bit dry if you don't put work into it. Claude models are great at writing porn and fairly creative even when SFW. Claude models are usually dumber then GPT models but that gap is closing. Gemini models are great at writing dialogue and not too hard to get to be uncensored but aren't very good at writing arousing porn. They also have pretty much a hard censor on lolisho. I haven't really seen much that impressed me with Gemini's smarts but it's at least not stupid.

Actually Chatting

Now that you've chosen a model you're probably ready to chat. Not all models are specifically for roleplay so you may need to tell them what exactly you want it to do. Some models, like the models in the GPT and Claude series are censored, this means that unless you trick the AI in some way you won't be able to get adult content. Sometimes all prompts are called jailbreaks, but jailbreaks are technically a more specific thing. Either way this rentry has a listing of a couple of jailbreaks and prompts you may want to look through.

Remember to have fun and plap plap those shotas!

FAQ

Q: How to export bots I made?
A: After choosing the bot you wish to export, click this button. Then click PNG, after that upload it to the site Catbox and use the link given to share your bots. Uploading the PNG directly to 4chan will not work, as data is removed.

Q: How to upload bots made by others?
A: On the main bot page, click this button. Then navigate to the bot you downloaded.

Q: I saw someone who had TTS and a Live2d model and Speech Recognition and an Emulator hooked up to their SillyTavern. How do I get that???
A: I'll make a guide on Extensions soon™

Bots hosted on this Rentry

Niji bots hosted on other sites:

https://www.chub.ai/characters?search=&topics=nijisanji
https://mega.nz/folder/cLkFBAqB#uPCwSIuIVECSogtW8acoaw

EN:
https://web.archive.org/web/20231218100113/https://botprompts.net/wp-content/uploads/2023/03/Aster-Arcadia-Nijisanji-EN-ILUNA-Yin.json
https://web.archive.org/web/20230201102543/https://botprompts.net/wp-content/uploads/2023/01/SonnyBrisko-Nijisanji-justanormiexxx.json

JP:
https://www.chub.ai/characters/nymous/higuchi-kaede
https://chub.ai/characters/pomudachi/emma-august-4074566f
https://www.chub.ai/characters/maliksusanto/shirayuki-tomoe
https://www.chub.ai/characters/nymous/ange-katarina-c514477d

ID:
https://www.chub.ai/characters/nymous/mika-melatika

Extras

Fanfic Translations I made using AI:
[Reader x Ike x Mob] [Pee] [Exhibitionism] [Ike being a cute slut]
https://rentry.org/8v943

Useful Links

Official Documentation
/aicg/ on /g/
/aicg/ on /vg/
/wAIfu/
Botmaking guides
Local Model guides