Gemini Guide

This guide will cover the following:

Getting a Gemini API Key

Proxy URL

Model Names

Sophia’s Lorebary + Custom Prompt

500 Errors

Getting a Gemini API Key

First go to:

https://aistudio.google.com/apikey

Then hit the button that says “Create API Key”

Make sure to save your API key somewhere you’ll remember.

Do not use your main Google account for the API key. I haven’t encountered consequences for using Gemini for NSFW but that doesn’t mean that Google won’t take actions against your Google account.

Open a chat and go to API settings. You can do this by hitting the little button that says “using janitor”. Select Proxy and hit the button that says “Add New Configuration”.

Scroll until you see “API Key” and a text box and paste your API key into that text box.

Proxy URL

Having an API key doesn’t mean anything if you don’t have a proxy URL.

I recommend using Sophia’s Lorebary because it makes things easier when it comes to having an uncensored roleplay when using Gemini.

Go to: https://lorebary.sophiamccarty.com/ THIS IS NOT THE URL YOU WILL PUT INTO THE PROXY URL FIELD

Hit the “Server Portal” button and “Setup Instructions” under Google AI Studio. You will find model names alongside the Proxy URL which is: https://lorebary.sophiamccarty.com/aistudio

Model Names

Here are 2 Gemini models that you can use for free.

Gemini 2.5 Pro

Model Name: gemini-2.5-pro

Number of free messages per day: 100

Context: 1,048,576 Tokens (Note that like any model its responses will slowly decline in quality as the amount of tokens it has to remember increases. It may begin to “forget” certain plot points and events. Plus Janitor only lets you go up to 128,000 Tokens.)

Gemini 2.5 Pro is slower than Gemini 2.5 Flash, but it produces better responses. When I switched from Flash to Pro I noticed that its responses were better. It slows down as the amount of messages goes up.

Gemini 2.5 Flash

Model Name: gemini-2.5-flash

Number of free messages per day: 250

Context: 1,048,576 Tokens (Note that like any model its responses will slowly decline in quality as the amount of tokens it has to remember increases. It may begin to “forget” certain plot points and events. Plus Janitor only lets you go up to 128,000 Tokens.)

Gemini 2.5 Flash is faster than Pro but I find that its responses are worse than that of Pro. I haven’t used Flash as much so I don’t have much else to say on it.

Sophia’s Lorebary

Before you start chatting with Gemini you need a custom prompt to put into API settings and to use Sophia’s Lorebary you need commands.

I personally recommend using Molek’s prompts, while they’re intended for Deepseek I find that they work well for Gemini. The link for these prompts is:

https://archive.ph/FNE8f

As for the commands for Sophia’s Lorebary you can find them by scrolling to the bottom. I recommend using the following but you do you:

<PREFILL_ON>

<JAILBREAK=ON>

<FORCETHINKING=ON>

<AUTOPLOT=ON>

<BETTERSPICE=ON>

These commands should be placed at the end of the custom prompt.

Sophia’s Lorebary is not an official Janitor AI extension and is a 3rd party tool. Use it at your own risk but you’re also using Gemini at the risk of whatever Google account you’re using if you use it for spicy stuff.

500 Errors

500 Errors are usually on the provider’s end, in this case, Google. If you get 500 errors you’ll just have to wait it out as there is nothing you do about it.