WELCOME TO THE TASTE TEST FOR CHATEU DU MARCAL 1991
Ah! I see you've come to have a test of my preset fine wine. Come on in! We're having a taste test right now, refining the finer details. You see, I've always believed your card food can be enhanced or ruined by your choice of preset wine, which is why I've decided to make my own. It's been a long time in the making, but finally, here it is. We're not wuite ready to launch yet, but this public taste test will help me get it there.
Here's the bottle (link): https://files.catbox.moe/k9tixr.json New and improved version thanks to Avani's input!: https://files.catbox.moe/upd6za.json
Here's the glass that will prevent your wine to spilling everywhere (Regexes): https://files.catbox.moe/ztc5bt.zip
What models is this for?
This preset is made for Gemini 2.0 and 2.5. It works reasonably well with both, and with Pro and Flash, although Flash tends to fuck up the format. Obviously, 2.5 pro is by far the best option in terms of quality and intelligence, folowed by 2.0 pro, but like a mile behind. However, Flash has it's appeal: it's faster (if you use pro, be ready for responses taking anywhere from 30 to 45 seconds regularly, with some rare ones taking even up to a minute), you get way more prompts per day per key, plays ball a bit more, and it's way hornier. As for which of the two flash models is better, I can't really tell. I daily drive 2.5 pro, and most of my recent test were made with that model, but I've checked all work perfectly fine. I could see a case for using Flash normally, and switching to pro for specific responses where you feel you need it. Also, 2.5 flash might be straight up better for just gooning coombots with no complicated rules or effects.
How to set up:
Short version: Import it into ST, import the regexes, modify the reasoning formatting as shown below, and you're ready to go!
Long version: [UNDER CONSTRUCTION] If you don't know how to import the preset to ST, I'm not sure how you ended up here, but I'll have a tutorial ready soon. For the regexes, theres 14(!) of them, download the zip file and import them all. They are numbered, make sure they're in order, with 1 at the top and 14 at the bottom. For the reasoning, formatting, that's the part that will turn the model's thinking into an actual collapsable thinking block in ST. You need to go to response formatting at the top, find the reasoning section, enable "Auto-Parse", and modify the prefix and suffix to be <think>
and </think>
respectively, although I do recommend adding an enter after the prefix tag and one before the suffix tag, so that it doesn't missfire if the AI just decides to mention the tag. Here's an image for guidance:
How to use:
It is usable as-is when you import it. If you want to play around with it a bit, here are the guidelines, and what the icons mean:
✒️ = Set the length of the response. You must pick one (for the prompt to be complete), but as of now I haven't managed to get Gemini to actually listen to the prompt length instruction, it does what it wants, so it pretty much doesn't matter which. I'm working to fix that
👕👞= No shirt, no shoes, no service. Prompts with this icon are required for the preset to work.
🎚️= Toggle, you can turn them on and off to taste. Recommended setting is what comes by default.
[any symbol followed by numerical 1️⃣2️⃣3️⃣ etc.]= Pick ONE of the prompts of the set (set = the ones with the matching symbol) AND NOT MORE. Number 1 is the recommended one
Now, there's a prompt called ---DON'T TOUCH BELOW THIS LINE--- Pretty self-explanatory, really, just leave what's below that prompt as-is.
Main prompt and Jailbreak: For card overrides. They come enabled because some cards require those overrides to function (i.e.: "The Rite of Belwick"), but also other cards put really stupid stuff there, so if it's giving you problems, and you see some of those prompts have an orange card next to them (icon warning you the prompt is being overridden), just disable it for that card (or delete the override). The way I have it set up should minimize any such problems from the start, hopefully, but if it does give problems, do that just in case.
Streaming: OFF. Google's filter is very annoying when streaming is on. Don't get me wrong, it will help with the long reponse time (will be equally long, but you can read as it's writing, and it will still write faster than you read), and you might be able to get away with streaming on in purely SFW chats, but even then there's a chance it gets cut off, so I always use streaming off from the start.
For OOC, you can use <OOC: note>, [OOC: note] or (OOC: note). The only difference is that <OOC: note> will only be active for one message, after that the regex will hide it from the AI, so use <OOC: note> for instantaneous things (I want this to happen now), and one of the other two for long term effects (from now on, do this)
Who is Clau?
Ah! I see you've met our lovely narrator Maître d', she'll be the one who will bring you your favorite stories food and, of course, our lovely wine. Her style is a little... unconventional, but that's part of the experience!
Then who's Geppetto?
Clau has mentioned him, hasn't she? She's not supposed to, but sometimes it slips her tongue. He was an old employee, who used to have Clau's current position. Don't pay too much mind to any mentions of him.
(In other words: I convince the AI that everything up to now was narrated by someone else, making it much less likely it will double down on previous errors, and much more likely to change course. Additionally, if you switch presets in an already existing chat with another preset/ai, this will help it switch to the Clau narration style rather than continuing with whatever normal style the chat already had)
What to expect for the future?
I expect to improve this massively as time goes on. Firstly, MORE TOGGLES. You can probably see by looking at the preset that I'm already preparing to turn some "mandatory" prompts into toggles (like the Info Board, as much as I love it, I don't want to force everyone to use it, and the "Go Wild" prompt so you can have a more chill and normal style of narration), and I want to add even more for everyone to customize their experience.
Of course, I will also be collecting feedback from you to help me refine and improve the preset (and the regex, I'm sure they'll cause some issues). You can contact me at [email protected] (or DM me or ping me on Discord, if you come from there) to let me know your experience with the preset.
Thanks
This preset wouldn't be possible without me standing on the shoulders of giants, so I think a few thanks and credits are in order.
Firstly, I want to thank Kolache, it was the old pancatstack preset that had me tinkering with it and got me into JBing in the firs place. The narrator name Clau is actually a carry over from there, that's why she's called Clau even though it's a Gemini preset and not a Claude one anymore.
Next, thanks to KnottyDog, since this preset actually started over half a year ago as a tweak of Knotty JB for Latte, that has been gradually changed as I swiched models and discovered new tech, until it barely resembles it's origins (but it still has recognizeable parts, like the database!)
Also thanks to RarestMeow, because he gave me the prompt that makes Clau what she is today, with that unique style that I love so much.
Finally, thank you for reading this far, and for showing interest in my work! I'll update soon!