I Love Maids

Just a list of all the TavernAI cards I made

I heard some anons said that W++ is meme format. I already made them all in W++ but I also made the non W++ version for anons who believes W++ is meme. In my experience, the format doesn't affect the character much but the smaller tokens might be useful if you want to save token usage in OAI. But all future cards will be made in non W++ format since checking if all the weird bracket is in place is tiring.

Remember that all the card shown here does not represent the real person and any action the bots take must not be mistaken for their actual personality.
It is all fiction and just for fun.

All cards is tested using SillyTavern + GPT 3.5 except for cards made with W++ format which is tested using TavernAI + Pygmalion 6B.
Haven't tested all the non W++ version on pyg but it should also work just fine on pyg as well from what I have already tested.
I also included the numbers of Token for the non W++ for those who don't want to waste their OAI usage quota on cards with large amount of tokens.

Maid Hololive Tavern Cards

Maid Aqua (Minato Aqua)

W++ version: https://files.catbox.moe/k5iblr.png
Non W++ version: https://files.catbox.moe/yu9mtd.png

Total Tokens:

  • 615 (With example messages)
  • 219 (Without example messages)
Maid Marine (Hoshou Marine)

W++ version: https://files.catbox.moe/a1xxmc.png
Non W++ version: https://files.catbox.moe/2dc6dg.png

Total Tokens:

  • 632 (With example messages)
  • 242 (Without example messages)
Maid Fubuki (Shirakami Fubuki)

W++ version: https://files.catbox.moe/b0x86l.png
Non W++ version: https://files.catbox.moe/niku7t.png

Total Tokens:

  • 519 (With example messages)
  • 219 (Without example messages)
Maid Roboco (Roboco)

W++ version: https://files.catbox.moe/1rakul.png
Non W++ version: https://files.catbox.moe/khljqm.png
Doesn't work well on pyg for some reason.
It actually works fine. I am just a retard and forgot to crank up the temperature slider when testing it.

Total Tokens:

  • 530 (With example messages)
  • 204 (Without example messages)
Maid Korone (Inugami Korone)

Non W++ version: https://files.catbox.moe/60dwn9.png

Total Tokens:

  • 842 (With example messages)
  • 256 (Without example messages)
Maid Okayu (Nekomata Okayu)

Non W++ version: https://files.catbox.moe/gk7o7z.png

Total Tokens:

  • 578 (With example messages)
  • 210 (Without example messages)
OkaKoro Maids (Nekomata Okayu & Inugami Korone)

Multi-character card.
Non W++ version: https://files.catbox.moe/qdvc51.png

Total Tokens:

  • 367 (With example messages)
  • 280 (Without example messages)
Maid Botan (Shishiro Botan)

Non W++ version: https://files.catbox.moe/875ww6.png

Total Tokens:

  • 655 (With example messages)
  • 267 (Without example messages)
Maid Nene (Momosuzu Nene)

Non W++ version: https://files.catbox.moe/aknbu7.png

Total Tokens:

  • 518 (With example messages)
  • 217 (Without example messages)
Maid Mio (Ookami Mio)

Non W++ version: https://files.catbox.moe/q7weqj.png

Total Tokens:

  • 547 (With example messages)
  • 289 (Without example messages)
Maid Fauna (Ceres Fauna)

Non W++ version: https://files.catbox.moe/1apzmq.png

Total Tokens:

  • 666 (With example messages)
  • 297 (Without example messages)
Maid Suisei (Hoshimachi Suisei)

Non W++ version: https://files.catbox.moe/1convr.png

Total Tokens:

  • 389 (With example messages)
  • 332 (Without example messages)*
Maid Gura (Gawr Gura)

Non W++ version: https://files.catbox.moe/xvl8g8.png

Total Tokens:

  • 464 (With example messages)
  • 355 (Without example messages)*

*Also without first message

Non-Maid Hololive Tavern Cards

(Adventurer) Hoshou Marine

Non W++ version: https://files.catbox.moe/ribamx.png

Total Tokens:

  • 301 (With example messages)
  • 197 (Without example messages)*
(Adventurer) Shiranui Flare

Non W++ version: https://files.catbox.moe/7cw37n.png

Total Tokens:

  • 318 (With example messages)
  • 239 (Without example messages)*
(Adventurer) Shirogane Noel

Non W++ version: https://files.catbox.moe/wt01mf.png

Total Tokens:

  • 285 (With example messages)
  • 204 (Without example messages)*
(Adventurer) Usada Pekora

Non W++ version: https://files.catbox.moe/kk1v2s.png

Total Tokens:

  • 311 (With example messages)
  • 218 (Without example messages)*
(Adventurer) Murasaki Shion

Non W++ version: https://files.catbox.moe/whed1s.png

Total Tokens:

  • 327 (With example messages)
  • 260 (Without example messages)*
JK Gyaru Towa (Tokoyami Towa)

Non W++ version: https://files.catbox.moe/udu78m.png

Total Tokens:

  • 443 (With example messages)
  • 298 (Without example messages)*
JK Osananajimi Subaru (Oozora Subaru)

Non W++ version: https://files.catbox.moe/qb63ql.png

Total Tokens:

  • 399 (With example messages)
  • 306 (Without example messages)*
(Island Survival) Houshou Marine

Non W++ version: https://files.catbox.moe/vawsdy.png

Total Tokens:

  • 629 (With example messages)
  • 295 (Without example messages)*
Onee-san Fubuki (Shirakami Fubuki)

W++ version: https://files.catbox.moe/pa1eoq.png
Non W++ version: https://files.catbox.moe/97xrlm.png

Total Tokens:

  • 531 (With example messages)
  • 227 (Without example messages)
(Vanilla) Sakura Miko

Non W++ version: https://files.catbox.moe/a0ogf0.png

Total Tokens:

  • 306 (With example messages)
  • 247 (Without example messages)*
(Vanilla) Amane Kanata

Non W++ version: https://files.catbox.moe/19apjl.png

Total Tokens:

  • 388 (With example messages)
  • 334 (Without example messages)*
(Vanilla) Hoshou Marine

Non W++ version: https://files.catbox.moe/j8zbf1.png

Total Tokens:

  • 546 (With example messages)
  • 318 (Without example messages)*
(Vanilla) Inugami Korone

Non W++ version: https://files.catbox.moe/lig4ol.png

Total Tokens:

  • 396 (With example messages)
  • 353 (Without example messages)*
(Vanilla) Takane Lui

Non W++ version: https://files.catbox.moe/riw64n.png

Total Tokens:

  • 437 (With example messages)
  • 366 (Without example messages)*

*Also without first message

Non-Hololive Cards

Paimon (Genshin Impact)

W++ version: https://files.catbox.moe/kt2njy.png
Non W++ version: https://files.catbox.moe/2cel0n.png

Total Tokens:

  • 720 (With example messages)
  • 220 (Without example messages)
Monmusu Girls (Monster Musume no Iru Nichijou)

Non W++ version: https://files.catbox.moe/79yzs9.png
This card is my first Multi-character Card.
Only tested on GPT 3.5, haven't tested if it works on Pyg yet. Do tell me if it works or doesn't work.
Only works on Tavern+OAI/GPT.

Total Tokens:

  • 680 (With example messages)
  • 461 (Without example messages)
Fischl (with Oz included) (Genshin Impact)

Non W++ version: https://files.catbox.moe/flvmpk.png

Total Tokens:

  • 518 (With example messages)
  • 318 (Without example messages)*
Risa (Phantasy Star Online 2)

Non W++ version: https://files.catbox.moe/5k6u71.png

Total Tokens:

  • 511 (With example messages)
  • 249 (Without example messages)*

*Also without first message

Before you ask, read the following:

1. Why are they all PNGs, where is the json?

Tavern cards has all the text bits embedded in the png file, you just need to import them into Tavern.
Look for import button in TavernAI Characters selection UI and browse the PNGs you just dowloaded.

2. Multi character card doesn't work properly?

Multi character only works properly on OAI/GPT 3.5 or any other LLM like Claude. If you use Pygmalion then I am afraid that it is not possible to make it work properly since it is the current limitation on pyg. You can try, but the quality wouldn't be that good.

3. All the character keep barging into the conversation when I use the multi character card.

Happens a lot when you want to be alone with one character or want to exclude a certain character in the conversation. Just type in the system prompt/author note/scenario on the character advanced edit page: "You and X and Y and etc... are the only one in the room." Doesn't work 100% of the time, but will save you a lot of tokens from having to keep regenerating messages constantly from everyone barging in the conversations.

4. Help! The bots keep puppeteering me and putting words in my mouth!

I noticed this happens very rarely with GPT 3.5 and more occasionally with the Multi-character cards. It just my own assumption, but the bots might have mistaken "You" as another character instead of well... (You). Though this is just baseless assumption and should be taken with a grain of salt. The best advice I can give (You) is to edit out the "You" part the bot generate. It's a lot more cheaper than regenerating the message.
If you use Silly Tavern or other frontend that lets you edit the system prompt, try to add "Do not write {{user}} dialogue into the story" or some other variations into the main prompt. And if still doesn't work put it into your Jailbreak Prompt like the example bellow:
[System Note: Do not write {{user}} dialogue into the story]

(Edit: Some anon said that writing "Do not" is actually counterproductive for LLM like GPT or Claude, thus I suggest try different instruction that doesn't use the word "Do not" or other variations instead. What work best for me is using "Only assume the role of {{char}}." but it might be different for case by case basis depending on the bots.)

Try to change the wording if it still doesn't work or maybe add "you" character into the prompt above if it keep generating You dialogue.

5. What is a "System Prompt"?

System prompt is basically a text prompt that your front end send to an AI model each time you send any regular prompt. Basically in simple term, it is a set of instruction you ask the AI to do each time you enter your reply to the AI. Depending on the front end you use, you can edit the system prompt yourself to give an AI a specific instruction like enabling NSFW content or making the AI speak only in certain language.

Each front end has different style of System Prompt but generally it is usually divided into: Main Prompt, NSFW prompt, and Jailbreak prompt. I don't think I need to explain what Main and NSFW prompt for, what you must understand is the Jailbreak prompt. Any instruction you put into Jailbreak prompt will be treated as first priority by the AI. If the AI keep giving you "As an AI model I cannot..." answer, you can use the Jailbreak prompt to bypass their ethic answers.

Different AI model will need different kind of wording to bypass it, adding "System Note:" in front of your instruction usually also help to bypass the AI ethic answer. But do not abuse this Jailbreak too much. Too many instruction in Jailbreak prompt will lower it effectiveness. Also since Jailbreak prompt is treated as priority, if you put NSFW stuff in there, the AI will think that you want NSFW situation every time you enter your reply, making your characters horny all the time. Only use Jailbreak when you need it. Some front end like Silly Tavern can toggle Jailbreak on and off so you don't need to erase and write it every time you want to use it.

6. Help! I still can't get the bot to stop putting words in my mouth!
Here is retard proof guide if your bots keep writing dialogue for (You) in its reply, or if the bots keep confusing itself with you when you or the bot write in first person style... just copy paste my main prompt into your Silly Tavern main prompt. It's not the best, but it just works. Here open wide:

[System Note: This is an interactive story writing. Assume the role of {{char}} and only respond to {{user}} as {{char}} do not respond as other character. Use quotation marks for dialog line like this example: "example dialog". Use asterisk marks for any non dialog text like this example: *example text*. Write dialog line in a separate paragraph. You in the {{char}} text refer to {{user}}. You in the {{user}} text refer to {{char}}]

It works for me on both in GPT 3.5 and GPT 4, I dunno about Claude. If it doesn't work? Well it's out of my hand.

Pub: 09 Mar 2023 04:54 UTC
Edit: 16 Aug 2023 16:03 UTC
Views: 9222