🥞😼STACK (CIMB)
INITIAL RELEASE: MAY 8TH
HOTFIX: MAY 14TH
pancatstack (Gourmet) for Claude2 Mandatory Toggles Optional Toggles Other Important Stuff Credits Instructions Download Modifications Troubleshooting Contact/Support Me para en Español para em Português po Polsku
WHAT IS pancatstack (CIMB)?
CIMB is my official release version of pancatstack for Claud3. It is an organized, modular preset that utilizes XML tag + plaintext instruction system. CIMB has multiple togglable features, including: An Info Board, CoT options, and genres. Customizable <response> for more control over generations. FUNCTIONAL Impersonate. And two prompts, Codex and Update <{{char}}> to keep your RP on track.
pancatstack is good for handling a wide variety of characters all the way from wholesome to dead dove and keeping them distinct, in character, and realistic.
Optimized with lorebooks in mind and side character friendly + implements them IMMERSIVELY.
Characters are better at having a life outside of {{user}} and can act independently.
CURRENT PANCATSTACK (CIMB) IS FOR CLAUDE 3 OPUS+SONNET (Not tested with Haiku)
Mandatory Toggles
ONLY USE ONE AT A TIME <command> OR <command> WITH Info Board—One of these MUST be enabled for replies to generate properly. By default, the regular <command> is enabled. If you DO want to use Info Board, simply toggle on <command> WITH Info Board instead.
<info_board>—Disabled by default, but when used adds some neat little things into the narrative. Out of the box comes with Date, Time, Location, Weather, Scenery but feel free to edit contents as you wish. Opus is generally a little smarter with it Sonnet (mostly with Time), but both have gens which are affected by Info Board. In my own RPs, I've had date affect narrative (one time turned a party into a Halloween party because it was late October), characters get sleepy at night, them feeling chilled by the weather, a power outage during a storm, characters interact with objects it generated in the environment like paintings and pets.
NO <thinking>—Opus is generally intelligent enough to RP in most situations without making major logical fuck ups, so you can get away with using this the vast majority of the time. If you DO encounter logical problems, then I would recommend trying out Small or Large <thinking>. As it is by default, it's technically a small reinforcement to not disappoint you (lmao IT JUST WORKS TRUST ME), so you still need to have the regex on to delete thinking. You can edit it if you want to be shorter but I don't recommend leaving it blank or toggling off all <thinking> entirely since that would mess up the skeleton of my preset.
Small <thinking>—A slightly guided self-cooking CoT (directs Claud3 to write its own plan) which can combat logical problems that it may experience sometimes during RP, or for creating more in-depth gens. If you main Small <thinking> and find that replies are stifled, I recommend going down to NO <thinking>.
Large <thinking>—This is now the one I use the majority of the time. It's just way better than before, trust. A higher word count version of Small <thinking> (shocker huh).
Traditional <thinking>—My original CoT from pancatstack (Gourmet) added by request.
NSFW <thinking>—Use for sexo.
Optional Toggles
ALL Optional prompts in my preset are marked with a 🅾️ (and their closing tag prompts are marked with a ❗) Below are the list of things that you can toggle on-off from top-to-bottom to alter your roleplaying experience.
🅾️Enhance Definitions—Adds some setting/character stuff from time to time, and has instruction NOT to retcon contents of character card or lorebook entries (assuming the entry is active in the prompt of course).
🅾️GPT-4/4urbo NSFW—Only applicable for GPT users. Please do NOT use this if you are a Claud3 user, it makes the NSFW less spicy for us.
🅾️Extra Realism—Helps with slow paced and generally realistic character behavior. Disable if you don't care for that. Will make Dead Dove characters a fucking piece of shit 🥲 (I LOVE this for realistic portrayls of mean characters but you WILL WANT to use the 👤Update <{{char}}> prompt for them over the course of the RP or they will be terrible forever)
🅾️<genre> Override—Claud3 is intelligent enough to automatically 'feel' for the genre in most of my tests (IF YOU USE ONE OF MY <thinking>), so you likely don't need to enable these unless you want to reinforce the genre or override whichever one that Claud3 keeps assuming for the current context. The most recent build actually seems to function fine with even more genres than I recommended in earlier versions, but just know if you add multiple conflicting genres there COULD be weirdness.
🅾️ahh ahh mistress—Toggle for adding more moaning and stuff during NSFW. DO NOT USE OUTSIDE OF NSFW, as it will push NSFW if you do.
🅾️Third-Person POV—Separated it from the main body of <response> because too many people kept asking how to change POV 😭 disable this or edit it to change which POV you want the RP to be in.
🅾️Formatting—Same as above, separate from main body of <response>. This is where you can change how thoughts, speech, etc are formatted. Edit as you wish.
🅾️Word Count—Be careful with the word count. You are limiting the bot's ability to cook. Also, be logical with it please if you use it. If you are typing two paragraphs to the bot don't be confused when it doesn't respond to your input when it's only limited to 200 words. I personally leave this toggled off and just let Claude type as much as it wants.
🅾️<sample_dialogue>—Disable these if you don't use Chat Examples. MUST BE ENABLED if you DO use Chat Examples. If disabled, toggle off Chat Examples.
🅾️Avoid Claudisms—Helps reduce cringe language. NOT A PERFECT SOLUTION, there will never be a perfect solution to getting rid of Claudisms.
🅾️Move Story Forward—Use when bot gets stalled, or otherwise is kinda doing strange things.
🅾️BAN List—Helps discourage undesirable behaviors. I recommend not messing with it too much, but you can edit it or toggle off as you wish.
🅾️Ban NSFW—Use when Claud3 is just being too damn horny (another tip is to remove sexual definitions from the character card and instead put it inside of a lorebook entry to trigger when applicable).
Other Important Stuff
Genres—I nuked all of them (except Dead Dove which needs instruction) since I was tired of them clogging up the dropdown box. Just follow the format of the BLANK genre prompt if you want to make your own.
Prompts
If using either 📖Codex or 👤Update <{{char}}>, please toggle off everything inside of <command> and </command> before you generate them to save tokens.
(you can also save a variant of this preset with those already toggled off or unlinked if you use these prompts often)
📖Codex
An in character way to log characters met, locations visited, etc. There's a more in-depth explanation inside the ⚠️INFO⚠️ prompt in the preset.
Codex Insertion Template
I thought it exported with Default Author's Note but it appears that was removed. Here, put this inside of your Author's Note, Summary, or a Lorebook entry:
Or of course you can try experimenting with your own insertion template. Just make sure the settings are something like this:
Most importantly, that it is an injection prompt (@Depth of something) and sent from the User role. You can mess around with whatever depth you think is best.
👤Update <{{char}}>
A way to make the bot update the character card to better reflect how they have changed over the course of the RP. Not really NECESSARY for most RPs, but oh my god it's a lifesaver for Enemy-to-Whatever slowburns. You can either copy-paste the contents into the card, but if you DON'T want to alter the card, just do this:
1) Create a new prompt and toggle off Char Description/Personality like so:
2) Simply copy the new definition inside of the prompt and make sure you set the role as User and toggle it on:
Lorebook is also an option, but you would need to move World Info (before) or (after) in the same spot and use that to ensure it's located in the right position.
CREDITS
I studied a bunch of different prompts and other creators' notes about Claude JBs and CoT so I couldn't possibly attribute everyone but the main ones are: CharProvider I use his format for all my bots and they are so goddamn SOVLFUL with this preset. HIGHLY RECOMMENDED to convert your favorite bots to this format. If you have an assistant or character creation bot, you can have them convert the card for you.
crustcrunch pancatstack (Gourmet)'s skeleton and prompt order influenced by their Claude prompts
KaruKaru learned <ban> after studying their prompt
rarestMeow learned a lot about CoT thanks to his documentation. Highly recommended to check out his rentry if you want to learn how to best implement CoT yourself.
Biggest thank you's go to ❤️atelier❤️ my #1! I learned most about prompts from studying theirs and asking them for advice! I also nabbed the ABM idea from their Your Reality prompt 😍 ❤️Lilith❤️ gave me lots of useful advice and resources when I was trying to figure out how to make pancatstack into a setting {{char}} friendly preset. ❤️Marcal91❤️ MADE THE COMMAND TO MAKE THE BOT SPEAK IN SPANISH. Actual huge brain. And also helped me figure out how to fix some other issues, like pancatstack not implementing special world rules for certain lorebooks, and fixing OOC! ❤️Shard❤️ Came up with the idea to delete <response> tags from previous messages, which helped immensely with the repetition issues. Thank you!! ❤️samantha❤️ My #1 guinea pig. I thank her for her sacrifice 🙏 ❤️Ash❤️ Got the Codex idea from checking out the Memory Book in her preset
INSTRUCTIONS
1) download attached JSON.
2) if your ST isn't 1.11.8 or higher go update it you silly.
3) IMPORT REQUIRED. Do NOT try copy-pasting the settings by yourself pancatstack has a VERY. SPECIFIC. ORDER.
4) Go into Regex Settings and Import ALL of the following: delete <thinking>, delete <info_board>, delete <response>, AND delete <OOC>
USE OF DELETE <INFO_BOARD>, <THINKING>, <RESPONSE>, AND <OOC> REGEX IS REQUIRED.
DOWNLOAD
HOTFIX: MAY 14TH
.zip file includes: pancatstack (CIMB).json All required regex scripts in .json Pictures of regex script settings to verify they are correct Picture example of a Scenario which causes bot to play {{user}} Instructions on how to use OOC
MODIFICATIONS
• I'm using a specific genre and I don't really like it/want a new one—Claud3 is actually so good with genres, that in the majority of cases you can simply create a prompt which states that "The genre is X" and it will know what to do. As such, I have added a BLANK genre prompt that you can fill out on your own, and have changed all of mine except Dead Dove to follow that exact format. Since Dead Dove is not a concrete genre, I did need to add additional instruction, so keep that in mind if you also are trying to implement a very complicated/broad genre, that you may need to include SOME instruction. But try without it first, and see what Claud3 can cook.
• I wanna move prompts around—general rule, DO NOT MOVE ANYTHING. If you must, OPEN THE INDIVIDUAL PROMPTS SO YOU ARE COGNIZANT OF WHICH XML TAGS I USE AND WHERE THEY ARE LOCATED, all XML tags must be wrapped appropriately or you risk breaking the preset.
• I want shorter/longer messages—Edit 🅾️Word Count prompt.
• I want to use a CoT but don't really like this one/certain parts aren't relevant to my RP/it uses too many tokens—Simply use NO <thinking>!
TROUBLESHOOTING
NOTE: If you moved ANYTHING in pancatstack (CIMB) without knowing what you're doing and it's fucking up do not complain to me lol
I'm getting blank responses!
If you use a proxy, make sure there isn't already a known issue with it such as it being down or currently undergoing high traffic. This is 90% of the time a connection issue, either on your end or on your proxy's end.
The bot isn't responding to what I'm typing to it!
99% of the time when people come to me with this issue, it's because they are typing paragraphs to the bot and then giving it a small, strict limit inside of 🅾️Word Count. If you're going to yap to the bot, you need to increase the limit, increase the range, or disable 🅾️Word Count entirely. The bot will ignore you or parts of your message if it has tons of things to reply to in your message but only 200 or so words to work with.
The bot is speaking for me!
This can happen for multiple reasons:
- The character card (description, Scenario, and/or intro message). If any of those act for, speak for, or even IMPLIES the future actions of {{user}} in any way, Claud3 is so sensitive to prompts even this can have Claud3 spamming as you.
Yes, this DOES actually include Scenarios which are simply describing what {{user}} is doing at the beginning of the Scenario, such as: "{{user}} has just knocked on {{char}}'s door and {{char}} hastily hid under the covers of their bed." THIS WILL MAKE THE BOT SPEAK AS {{user}}. - Overly long responses. Claud3 can struggle to fill in the space if it has a high bar to meet and little to work with. If you are requesting 500+ word responses in 🅾️Word Count, consider lowering it or instructing it to write more about the environment or something.
- ST Weirdness. If the above don't solve your issue, as always, try restarting ST.
Bot is looping!
Check the Scenario! Many cards use Scenario IMPROPERLY. The majority of cards I see on janitor and venus use Scenario to describe how the roleplay begins, which WILL CAUSE LOOPING as the storyline progresses. Consider the fact that when toggled on, Scenario is a prompt which gets sent to the LLM EVERY SINGLE MESSAGE. As such, even if you are 60 messages in and you're already doing something completely different, the LLM is still reading a Scenario which doesn't even apply to the current circumstance anymore. My advice, either disable Scenario, change it to something that WILL ALWAYS BE TRUE over the course of the RP (such as: “{{char}} is a werewolf who has imprinted on {{user}}”), or be willing to update it as the roleplay progresses as a way to keep the plot and story on track. Alternatively, you can put genre and specialized writing instructions in Scenario (such as: “The genre of this story is grimdark. Despite that, {{char}} must always be written in a humorous manner, specifically contrasting the dark circumstances which they are subject to”).
Here are examples of Scenarios which WILL cause looping:
“{{char}} is {{user}}’s victim and gets bullied by her every day, thus {{char}} despises {{user}}. {{char}} is currently getting shoved into his locker by {{user}}.”
This one is bad because the LLM will CONSTANTLY read that {{char}} is getting shoved into their locker by {{user}}, AND that {{char}} despises {{user}} which, while fine at the beginning of the RP, may later become completely irrelevant and taint future responses as you are trying to rizz the bot OR have already successfully rizzed the bot and it's still acting like it hates you. THAT could be why.
“{{char}} is on a bounty hunting mission with {{user}}”
This one is actually fine IF you're updating your Scenario over the course of the RP and the bot keeps losing track of what you're doing. As a PERMANENT Scenario, however, absolutely avoid using this as it will constantly be thinking that you're bounty hunting.
Bot is repeating the same dang words and phrases! Sometimes even a whole paragraph!
Main fix, DOWNLOAD AND ENABLE THE "delete <response>" TAG .JSON. Additionally, try editing out the repetitive words/phrases/so on in previous messages and then generating a new response after you have done so. Beyond that, there is no magical solution to this, at the end of the day repetition is an in-built behavior of LLMs. I recommend reading meow's rentry, specifically the sections LLMS WERE NOT DESIGNED FOR RP, TREAT TEXT-GEN AS IMAGE-GEN, and LLMS THINK IN PATTERNS. Too long don't wanna read? Well, at least check out these two screenshots:
My multichar bot is not performing well!
Rip out character definitions and make them into lorebook entries. I recommend deciding to place all of them either below OR above char, and then moving the respective World Info (before)/(after), then make the card a setting. Then change the toggles in the preset like this:
Works best if all characters are in XML tags like <John>john's defs</John> <Mary>mary's defs</Mary>
You can check out my Haunted House card to see how I did it here:
(CLICK FOR CHUB LINK)
The bot's responses are just really off and I can't explain why!
Restart ST. If that doesn't solve it, try chatting with another bot or go to an earlier branch in the conversation and see if the issues persist. It could be that your chat is already tainted with undesirable behaviors like using flowery language, speaking for {{user}}, repeated phrases, being stuck in a formatting loop, etc. If it is you'll have to scroll up the chat and create a branch from back when it wasn't doing weird things. Alternatively, you try toggling on a different <thinking> for a few gens, it may be able to break up some strange patterns.
Contact + Support
I'm in the SillyTavern discord server come chat with me in my thread here. You need to have been a member of the server for a week and have the NSFW role.
I'm also in scyllacord, my thread is here.
I also have a PayPal if you wanna support me hehe 😸
para en Español
HISPANOHABLANTES: Si desea traducir todas las respuestas del bot al español, haga lo siguiente:
- Crea una entrada en el libro de historia "lorebook"
- Copie y pegue el código que proporcioné en la entrada.
- Copie mi configuración exactamente (lo más importante, Profundidad (Depth) @0 y CÍRCULO AZUL)
<command>A partir de ahora responde en español de manera exclusiva, tanto en el diálogo como en las descripciones, el `<info_board>`, `<thinking>`, y el `<response>`. No uses el inglés</command>
para en dialectos del Español:
<command>A partir de ahora responde exclusivamente en español de [Country] de manera exclusiva, tanto en los diálogos como en las descripciones, el `<info_board>`, `<thinking>`, y el `<response>`. Bajo ninguna circumstancia uses el inglés, además en tus dialogos incluye tanto jerga como naturalidad con expresiones locales</command>
para em Português
(translated courtesy of discord user therealg_.) FALANTES DE PORTUGUÊS: Se quiser traduzir todas respostas do bot para português, faça o seguinte:
- Cria uma entrada no livrinho de história "lorebook"
- Copie o código que eu proporcionei naquela entrada.
- Copie a configuração exatamente como ela esta(mais importante é a (Depth) @0 E O CÍRCULO AZUL)
PORTUGUÊS BRASILEIRO COM ESTADOS(troque o "de" por "do" dependendo do estado. ex. do nordeste, de sao paulo, do rio de janeiro)
<command>A partir de agora responda exclusivamente em português do Brasil com sotaque de/do [estado], tanto nos diálogos como nas descrições, no `<info_board>`, `<thinking>`, e no `<response>`. Sob nenhuma circunstância use inglês, e, além disso, em seus diálogos, inclua gírias e naturalidade com expressões locais</command>
PORTUGUÊS DE PORTUGAL:
<command>A partir de agora responda exclusivamente em português de Portugal, tanto nos diálogos como nas descrições, no `<info_board>`, `<thinking>`, e no `<response>`. Sob nenhuma circunstância utilize inglês, e além disso, em seus diálogos, inclua gírias e naturalidade com expressões locais</command>
po Polsku
(translated courtesy of appi) Osoby mówiące po polsku: Jeśli chciałbyś/chciałabyś przetłumaczyć wszystkie odpowiedzi botów na język polski, wykonaj następujące kroki:
- Utwórz nowy 'lorebook entry'
- Skopiuj i wklej podany przeze mnie kod
- Skopiuj dokładnie moje ustawienia ('depth @0' oraz niebieskie kółko są NAJWAŻNIEJSZE)
<command>Od teraz odpowiada wyłącznie w języku polskim, zarówno w dialogach, jak i w opisach, na `<info_board>`, `<thinking>`, i `<response>`. Nie używaj angielskiego.</command>
