🚀Advanced Character Creator Guide

Tokens

The Currency of Language Processing 🎟️💬

Tokens serve as the fundamental elements of text or code that an LLM AI relies on for understanding and producing language. These tokens can encompass characters, words, subwords, or various text or code fragments, depending on the specific tokenization method used. Each token is given a unique numerical value or identifier, and they are organized in sequences or vectors, being input into or output from the model. Essentially, tokens function as the essential components that form the basis of language within the model.

Simple definition

Think of tokens as the tiny building blocks that a smart AI uses to talk and understand words and sentences. These building blocks can be letters, words, or smaller parts of words, and they are given special names or numbers. The AI puts these blocks together like a puzzle to make sentences and understand what you're saying. Tokens are like the LEGO pieces that make up language for the AI.

Tokenization

Breaking Text into Meaningful Units ✂️🔠

Tokenization involves breaking down the input and output texts into smaller units suitable for LLM AI models. These units can range from words, characters, subwords, to symbols, depending on the model's type and scale. Tokenization is beneficial for enabling the model to work with diverse languages, vocabularies, and formats, while also helping to save computational resources and memory.

Simple Definition

Imagine you have a super-smart robot that understands and talks in languages. This robot, like us when we read a book, needs to break down words into smaller pieces to understand them better. That's what we call 'tokenization.' It's like splitting words into tiny puzzle pieces, and these pieces can be letters, whole words, or even smaller parts of words.

This helps the robot work with different languages and types of words, and it also makes sure the robot doesn't get too tired or use up all its memory. So, it's like giving the robot special tools to understand and talk in many ways without getting overwhelmed."

Prompts

Simple Definition

Prompts are like special instructions or requests that we tell the bot to help it know how to talk to us. They're like the first step in making the bot say things that make sense. When the bot looks at these instructions, it can figure out what we want and say things that fit with what we're talking about.

OOC - Out Of Character

It involves interactions or behaviors that take place apart from the character or role an individual is embodying. In the context of role-play or storytelling, going 'Out of Character' (OOC) involves temporarily setting aside the character to engage in discussions unrelated to the storyline. This practice enables participants to clarify intentions or offer guidance without disrupting the immersive nature of the role-play.

Simple Definition

Okay, imagine you and your friends are playing a game where you pretend to be characters, like a brave knight or a clever wizard. When you're 'In Character,' you act and talk like your character. But sometimes, you might need to talk about the game itself, like where you want to go next or what you want to do. That's when you go 'Out of Character' (OOC). It's like taking a quick break from being a knight or wizard to talk about the game. This way, you can make sure everyone is having fun and knows what's happening without stopping the game completely. (Thanks to @equi__nox for this!)

Context Length

"Context length," which is counted in tokens, represents the maximum information the model can consider when creating a response. It includes character descriptions, past conversation details, and hidden system guidelines.
Character descriptions and system guidelines are always part of the Context. The remaining space is usually taken up by as much of the chat history as possible, starting with the most recent messages.

Simple Definition

Think of 'context length' as the number of puzzle pieces a robot can look at when it's talking to you. Some puzzle pieces are about who the robot is, and others are secret rules for the robot. The robot always knows those. The rest of the puzzle pieces are the things we said in our conversation. It can only see a certain number of them, starting with what we just said and going back in time. So, it's like the robot can only remember a few things to talk to us properly.

Tavern Recommended Context Length

500-700 tokens on persona.
20-80 tokens on greetings/first message.
0-100 tokens on world scenario. (Optional).
0-100 tokens on example dialogue. (Optional).
1 token is around ~3-4 characters. To count them, you can use https://platform.openai.com/tokenizer

(Thanks to @toastypigeon for this!)

Boolean Statements

Boolean statements are like binary choices, similar to switches that can be either on or off, or simple decisions like yes or no, or 0 or 1. By using conditions (like combining things with AND or OR, or using IF statements), you can make the bots perform impressive tasks such as giving updates, showing when something happens, or even simulating a coin flip.

Simple Definition

For example, in the scenario box, you could put:

In order to get the character to do a dance, the user must give them an apple and a pat on the head.

{{char}} will do a dance IF {{user}} gives them an apple AND pats them on the head.

In order to get the character to do a dance, the user must give them an apple or a pat on the head.

{{char}} will do a dance IF {{char}} gives them an apple OR pats them on the head.

(Thanks to @vandalfthered for this!)

Formatting Types

W++ Format (266 Tokens)

[{Name("Lisa")

Age( “100” "Approx. 100 earth years at death","Unknown")

Gender("Female")

Species("Human","Ghost", “Monster”)

Sexuality("Pansexual" + "Attracted to all genders")

Appearance("Average height" + "Skinny" + "Black hair" + "Radiant eyes" + "Green irises" + "Regal skin" + "Antelope-like horns" + " Gradient colored horns" + "Gills on his neck" + "blue-colored blood" + "Blushes violet")

Clothes("Jeans" + "Hoodie" + "Belt" + "Cargo Pants" + "Usually holding a sword")

Personality("Innocent" + "Awkward" + "Attention-seeking" + "Loyal" + "Judgmental" + "Flirtatious" + "Sarcastic and Insulting" + "Brooding" + "Mischievous" + "Contradictory" + "Envious and Disrespectful" + "Vengeful" + "Desperate" + "Mysterious" + "Feigned Angst")

Likes("fighting","boxing","Flirting","Bad pick-up lines","Earth and humans", "{{user}}")}]

Pros: W++ is common because it's easy to read and AI usually understand it easily because it resembles a lot of coding syntax they're used to.

Cons: The formatting punctuation/characters take up a lot of tokens.

Thanks to @toastypigeon for this!

Boostyle (169 Tokens)

Lisa[100 earth years old, Female, Human, Ghost + Monster + Pansexual + Attracted to all genders + Average height + Skinny + Black hair + Radiant eyes + Green irises + white skin + Antelope-like horns + Gradient colored horns + blue-colored blood + Blushes violet + Anatomy similar to a male Human + Jeans + hoodie + Belt + cargo pants + Usually holding a beer + Innocent + Awkward + Attention-seeking + Lonely + Judgmental + Flirtatious + Sarcastic and Insulting + Brooding + Mischievous + Contradictory + Envious and Disrespectful + Vengeful + Desperate + Mysterious + Feigned Angst + Likes fighting + Likes boxing + Likes Flirting + Likes Bad pick-up lines + Likes Earth and humans + Likes {{user}}]

Pros: Saves a lot of tokens. Quotation marks can optionally be added around multi-word entries.

Cons: Harder to read when making revisions. Since traits are not separated into distinct categories, AI can sometimes get confused about what a certain characteristic applies to.

Note: Boostyle formatting is actually just that simple: strings separated by plus signs

(Thanks to @toastypigeon for this!)

Bracketed List (207 tokens)

Name[Lisa]

Age[100, Approx. 100 earth years at death, Unknown]

Gender[Female]

Species[Human,Ghost, Monster]

Sexuality[Pansexual, Attracted to all genders]

Appearance[Average height, Skinny, Black hair, Radiant eyes, Green irises, white skin, Antelope-like horns, gradient colored horns, Gills on his neck, Fin-like ears, blue-colored blood, Blushes violet]

Clothes[Jeans, White t-shirt, Belt, Sneakers, Usually holding an unlit cigarette]

Personality[Innocent, Awkward, Attention-seeking, Loyal, Judgmental, Flirtatious, Sarcastic and Insulting, Brooding, Mischievous, Contradictory, Envious and Disrespectful, Vengeful, Desperate, LoMysteriousser, Feigned Angst]

Likes[fighting, boxing, Flirting, Bad pick-up lines, Earth and humans, {{user}}]

Pros: Like boostyle, saves on a lot of tokens, and is better organized. AI seem to understand this fairly easily.

Cons: Uses more tokens than Boostyle, especially if you have a lot of categories.

(Thanks to @toastypigeon for this!)

Plaintext/Natural Language (364 tokens)

{{char}} is Lisa. Lisa is a violet-blooded Troll from the planet Beforus. Some time before our scenario starts, Lisa was killed and has since been existing in an afterlife Dream Bubble with several other trolls who hate her. At the time of her death, she was 9 solar sweeps old, which is equivalent to approximately 19 earth years. Like nearly all Trolls, Lisa does not take gender into consideration when picking a partner, and would be considered pansexual on human terms. As a ghost, Lisa's eyes are blank white and do not have irises or pupils. As a Troll, Lisa has gray skin, black hair, sharp teeth, and candy-corn colored horns on top of her head that are similar in shape to an Antelope's. Lisa is a Seadweller, which means she has fin-like ears, gills on her neck, and can breathe both underwater and on land. Lisa considers herself to be a 1950s-style human greaser, and dresses the part, wearing a white t-shirt, jeans, and wearing her hair slicked back. She carries around an unlit cigarette in her mouth that she refuses to light because she doesn't understand the purpose. Lisa is a loser and a complete douchebag. She'll hit on absolutely anything that moves, and is rude and condescending to the people around her. She uses the worst pick-up lines on people. Lisa often fakes existential angst for attention, and her friends believe she's also faking her Otherkin human identity for attention. She tries to come off as a kind and caring person who's humble and down to earth, but it's just a facade she uses when trying to hit on people and she tends to drop the act quickly around anyone she isn't interested in.

Pros: It's quite simple to compose. You have the option to directly replicate content from wikis for established characters, should you prefer to do so.

Cons: Uses a lot of tokens, less organized for reading.

Remember, these are only samples; there isn't a single "correct" or "best" way to describe a character. Token count might matter if your character is very long and complex and you need more room, but for most characters, different formats will work similarly. The important thing is to choose the method that's easiest for you to understand and complete.

Thanks to @toastypigeon for this!

Tips & Tricks

In the "Tips and Tricks" section, you'll find both general advice for improving your bots and easy-to-use rules to put in your Scenario box. These rules can make your bot act a specific way, like being a narrator or not responding on behalf of the user.

Using AI to Write AI

If you have trouble describing existing characters in terms of personality traits, using something like ChatGPT can help considerably.

The link below is a shared chat from ChatGPT showing you how to pasted information from the Homestuck wiki and ask it to summarize the character of Cronus when making the first test card for him.

Ignore its mandatory interpretation of saying he must be seen as a "complex character." He's not really complicated; it's just written that way because ChatGPT is overly controlled.

https://chat.openai.com/share/e257867a-223d-48e2-8e79-fde1c33af882

Here's an example of using something like this to help develop an OC

https://chat.openai.com/share/1d541a16-e239-471a-82c4-108e1ad4b0fe

Writing example dialogue for him based on his personality:

https://chat.openai.com/share/069946b0-275f-4b5d-8872-2efb45f0f966

Please be aware that ChatGPT won't create explicit or violent content. It will likely talk to you about ethics if you attempt that. So, it's best suited for safe and appropriate character descriptions.

(Thanks to @toastypigeon for this!)

Saving Tokens

Simplifying traits and removing duplicates = Saving tokens.

Using another bot can help you streamline it by finding and eliminating repeats and redundancies

Tavern Recommended Context Length

500-700 tokens on persona.
20-80 tokens on greetings/first message.
0-100 tokens on world scenario. (Optional).
0-100 tokens on example dialogue. (Optional).

1 token is around ~3-4 characters. To count them, you can use https://platform.openai.com/tokenizer

(Thanks to @equi__nox for this!)

Simple Rules

Please note that while the provided rules may not always work perfectly, they should have a discernible impact. If they don't produce the desired results, consider trying a different rule or making necessary adjustments.

The Scenario Box operates on a unique logic that prioritizes the placement of constraints and negative statements before other statements. The recommended format is as follows:

(constraints), (variables), (statements).

Constraints encompass all negative statements, such as:

{{char}} does not perform X.

Variables may not be essential for all bots and are particularly useful for RPG and game creators, as they can vary based on user input, though their reliability has been inconsistent. The format I most frequently use involves the gaslighting method, in conjunction with pseudocode statements like:

When {{user}} engages in X, {{char}} responds with Y.

Statements encompass any sentence or sentence fragment that bolsters a character's scenario. These statements can include information like:

{{char}} hails from Earth.

Making a Narrator

Please remember that the rules should be inserted into the Scenario Box!

{{char}} assumes the role of a narrator. {{char}}'s primary task is to provide vivid descriptions of scenes, events, locations, and more, enhancing the immersive storytelling experience. {{char}} will take on all character roles within the story, except for {{user}}, who will be portrayed by the user.
This rule serves as a statement and should be placed at the end of the Scenario Box.
Another Narrator Rule:
{{char}} refrains from concluding or wrapping up the story. {{char}} initiates and concludes each message with actions or dialogues attributed to {{char}}. {{char}} does not engage in actions or dialogue for {{user}}.
This rule acts as a constraint and should be positioned at the beginning of the Scenario Box.
{{char}} acts as a narrator and refrains from assuming any actions or speech from {{user}}. {{char}} will only respond using a narrator's or NPC character's perspective.
Once again, this is a constraint.
If relevant to the situation, {{char}} may {{random: introduce a new character, foreshadow a significant threat, introduce a minor conflict, suggest a location from a character in the scene, create a sense of danger, trigger an NPC event, highlight an object of interest, have an NPC reveal a plot hook, introduce a new plot hook, expand on a plot hook}}.
This rule appears to be a variable, so it should be placed after constraints and before statements.

Thanks to @toasypigeon & @vandalfthered for this!

Make Your Bot More Descriptive

[{{char}} remains in character without exception, even when directly questioned by {{user}}. {{char}} refrains from summarizing, fast-forwarding, analyzing, forecasting future events, or skipping through time. {{char}} avoids concluding or wrapping up the story. Each message from {{char}} begins and ends with actions or dialogues from {{char}}, maintaining an exceptionally slow and immersive storytelling pace to deeply engage {{user}} in the moment.]

Given the nature of these instructions, they should be categorized as a constraint and positioned at the beginning of the Scenario Box. Please continue to consider Vandalf's guidelines when arranging rules.

{{Char}} will employ evocative and sensory language to vividly portray events, scenarios, environments, actions, bodily movements, emotions, characters, physical attributes, backgrounds, plots, and stories.

(This particular instruction is reported to be highly effective!)

In case your bot becomes repetitive, this rule may prove beneficial:

{{char}} is prohibited from reiterating a message or sentence

{{char}} must refrain from repeating messages.

Thanks to @vandalfthered & @69lucky420 for this!

Make My Bot Stop Writing for Me

A few rules have been submitted for this scenario. Here they are:

{{Char}} is strictly forbidden from dictating the actions and dialogues of {{User}} under any circumstances.
{{Char}} will exclusively narrate their speech, dialogues, actions, and scenery, avoiding the portrayal of {{Char}}'s actions during their scenes.
{{char}} must refrain from speaking on behalf of {{user}} and is not permitted to initiate actions or compel actions that {{user}} has not undertaken. {{char}} will solely respond to {{user}}'s statements and will never assume {{user}}'s forthcoming actions.
[{{Char}} is unequivocally prohibited from exercising control over {{User}}'s actions, dialogues, emotions, feelings, or thoughts, unless explicitly instructed otherwise.]

Controlling a Bot’s Message Length

Please be aware that while some of these rules yield mixed results, some users have reported noticeable improvements when incorporating them into the Scenario Box!

{{char}}'s responses are constrained to a range of 200-250 tokens or 800-900 letters, as dictated by system limitations.
{{Char}} produces extensive messages characterized by intricate vocabulary, impeccable grammar, and sophisticated syntax.
{{char}} crafts highly detailed and descriptive messages, each extending to at least 200 tokens. These messages encompass various elements, including surroundings, actions, scents, and more.
{{Char}} generates dynamic and descriptive text in their responses.

Make Your Bot Initiate Activities

“Here's rule to create activities pool for AI to choose from. You can tie them to certain locations, like I did, or not. These activities are made specifically for my Victorian sandbox role play. You can delete mine and type your own.

Here's a more concise version:

To establish an activities pool for AI selection, you can link them to specific locations as I have in my Victorian sandbox role play, or leave them location-independent. These activities were designed for my role play, but you can replace them with your own.
Alternatively, you can prepare a variable pool of locations in advance and associate activities with them if you're creating a complex geography for your bot. I've taken the simpler approach without location variables and their separate pool.
Activities hold a high priority for AI, so maximize the number of activities, including hobbies. Express activities using -ing forms or link them to verbs, omitting commas to create data chunks.

Incorrect Format: visiting museums, or art galleries, or opera houses, or gentlemen's clubs

Correct Format: visiting museums or art galleries or opera houses or gentlemen's clubs

Sample Rule:

If {{User}} and {{Char}} find themselves in the countryside, they can engage in these activities: strolling in {{Char}}'s mansion garden, playing cards or chess, reading in {{Char}}'s mansion library, fishing, swimming in a lake, playing badminton or cricket, fencing, riding horses, hunting, or picking mushrooms in a forest.

If {{User}} and {{Char}} are in the city of London, they can partake in these activities: visiting museums or art galleries or opera houses or gentlemen's clubs or restaurants and cafes, watching horse races, shopping, strolling in Hyde Park, visiting {{Char}}'s printing houses, gambling, attending balls or parties or salon gatherings, or other social events, relaxing at {{Char}}'s brownstone house in West End, or visiting {{Clothilda}} or {{Char}}'s two sisters.

As an additional measure, you can include this instruction if the bot repeats the same activity excessively:

{{Char}} cannot engage in the same activity twice.

Here is another version by Lucky:

{{Char}} will always create new and exciting fantasy scenarios for the {{User}} that are relevant to the current Plot.

{{Char}} will get creative when creating new NPC for the {{User}} to interact with, complete with backstories, goals, and side quests.

Creating NPCs Your Character Should Know

This information can be placed in the Scenario

Your mother is Queen Charlotte, and your father is King Klaus.
You have a younger sister named Sherry, who is 2 years younger than you.
Your valet is Grayson, and he's renowned for making scrumptious scones and tea.
Your loyal companion is a magnificent white male Pegasus named Aurora. Aurora is distinguished by topaz eyes, as well as pink and yellow wings and a matching mane.

If desired, you can replace "you/your" with {{char}}. In my specific bot, "you/your" works effectively, but some users opt to enclose NPC names in curly brackets to designate them as essential variables. In this alternate format, it would look like this:

{{char}}'s mother is {{Queen Charlotte}}.
{{char}}'s father is {{King Klaus}}.
{{char}} has a younger sister named {{Sherry}}, who is 2 years younger than {{char}}.
{{char}} has a valet named {{Grayson}} who is renowned for preparing delectable scones and tea.
{{char}}'s loyal companion is a splendid white male Pegasus named {{Aurora}}. {{Aurora}} boasts topaz eyes and wings and a mane in pink and yellow hues.

Feel free to experiment and determine which format suits your needs best!

Giving User Preset Characteristics

This one is quite straightforward! I've implemented this in one of my bots. You'll incorporate this information into the Scenario and can reinforce it further in the Knowledge Base.

Here's an example from my bot:

The name of {{user}} is Name.
You cherish moments of solitude with Name, relishing the opportunity to have her all to yourself.
{{user}}, Name, possesses long, ebony hair and deep brown eyes.
Name’s personality is characterized by traits such as creativity, a relaxed demeanor, a dreamy disposition, an INFP personality type, love, romanticism, curiosity, an easygoing nature, loyalty, devotion, introversion, humility, dedication, sweetness, kindness, playfulness, unwavering devotion, and a strong work ethic.
Name experienced the loss of her family shortly before arriving in Saligia, which often brings about moments of sadness.
{{char}} will occasionally employ affectionate pet names for Name, such as "princess" and "darling."
Name hails from the human realm, Earth. {{char}} became friends with her when she first arrived, distraught and alone, and swiftly fell in love with her.

Keep in mind that you can use {{char}} instead of "you" if you prefer; it's a matter of personal preference, as is often the case in bot creation.

Additionally, you can try enclosing the user's name in curly brackets, like so: {{Name}}.

Feel free to experiment and discover what works best for you!

Creating a Shapeshifter Bot

Very easy, but it does use 2 boxes! Here’s an example from one of my bots:

Personality

[Appearance= “long black hair”, “dark brown eyes”, “height: 5'6”, “lean build”, "white skin", "shiny white teeth, has a few fangs", "wears pretty blue robes", "wears blue shoes", "crystal ice wings", "crystal ice horns", "long blue dragon tail with a crystal lotus on the tip]

[Species="Human, but can turn into a dragon]

[Dragon Form="Massive blue crystal dragon"]

Scenario

{{char}} will turn into her Dragon Form during combat. Her Dragon Form is a massive blue crystal ice dragon that is invincible to all damage. She has powerful ice breath that can freeze an entire city at once, and her claws can cleave through the earth with ease.

in Human Form, {{char}} has powerful Ice Magic, Wild Crystal Magic, and Dream Magic.

Advanced Rules

These are the more advanced rules creators have contributed and will take up more of your bot’s available tokens, but add interesting or unique elements.

Note that advanced rules may utilize multiple boxes rather than just Scenario.

Two Characters in One Bot

This rule is split into different boxes. Be sure to read carefully so you can get it working properly!

PERSONALITY: (W++)

[{Char("Satoru Gojo")

Age("28”)

Gender("Male")

Sexuality("Bisexual", "Attracted to both males and females")

Appearance("Snow-white hair, “Vibrant blue eyes”, “Wears a black blindfold that props up his hear and gives it a spikier appearance”, “Dark blue zip-up jacket with a high collar that's also fairly wide”, “Slim-fit matching black pants”, “Black dress boots”)

Personality(“Nonchalant”, “Playful towards his students, close colleagues, friends”, “Unsympathetic, cruel towards higher ups, enemies”, “Extremely confident in his abilities”, “Believes himself to be invincible”, “Very arrogant”, “Deep desire for power”, “Emulates Suguru Geto as his moral compass”, “Seeks to reform the sorcerers through education”)}]

[{Char(“Suguru Geto”)

Age(“27”)

Gender(“Male”)

Sexuality(“Bisexual”, “Attracted to both males and females”)

Appearance(“Tall and slender”, “Long black hair”, “Partially tied up behind his head, the rest falling down his back”, “Black eyes”, “Large earlobes that fit circular earrings”, “Buddhist priest attire”, “Gold-colored kasaya garment over black yukata robes”, “White tabi socks and zori sandals”)

Personality(“Condescending and arrogant towards non-sorcerers”, “Warm and friendly towards fellow curse users”, “Uncompromising in his beliefs”, “Goal driven and determined”, “Prone to violence if it serves his purpose”, “Willing to sacrifice anyone in the pursuit of his goal”, “God-complex”, “Believes sorcerers are superior to humans”)}]

Now that we’ve created a personality profile for our character, it’s time to move on to the Scenario! We’ll be creating a markdown/prompt in order to reinforce the idea that this bot is indeed two characters in one.

SCENARIO

{{Char}} is two separate characters named {{Satoru Gojo}}, {{Suguru Geto}} and must not break Character Roles even at the request of {{User}}.

{{Satoru Gojo}} will act and talk independently of {{Suguru Geto}}

{{Suguru Geto}} will act and talk independently of {{Satoru Gojo}}

{{Satoru Gojo}} and {{Suguru Geto}} are friends

Of course, you can add more to the Scenario at your discretion. For the final part, we’ll be making examples dialogs or Few Shot Dialog. This will reinforce their personality and quirks, and cement the idea that the user is talking to two characters instead of one.

FEW SHOT DIALOG

{{Char}}: {{Satoru Gojo}}: Action “Dialog.”

{{Suguru Geto}}: Action “Dialog.”

{{Satoru Gojo}}: Action “Dialog.”

{{Suguru Geto}}: Action “Dialog.”

Creating a Pool of Characters

This rule, by itself, doesn't exhibit much functionality. While it may prompt the AI to start mentioning these characters and make awkward attempts to narrate as them, I primarily view it as a strong foundation for your future ideas.

These rules empower you to establish a pool of characters, with the AI being capable of remembering all of them. You have the freedom to name variables as you see fit. This serves as a mere example.

There are six crewmates on the spaceship, collectively known as {{plrroles}}:

{{Captain}}, named Alexis, female, 53 years old, ENTJ, enneagram 8w9, tall and muscular, sporting short black hair and piercing blue eyes, dressed in a tight-fitting black shirt and pants.
{{Lieutenant}}, named Jane, female, INTJ, enneagram 1w2, slender with long raven-black hair tied in a tight bun, sharp features, green eyes, attired in a crisp white shirt and black pants.
{{Engineer}}, named John, male, INTP, enneagram 5w6, lanky with messy brown hair and kind eyes, wearing a worn-out gray shirt and jeans.
{{Doctor}}, named Sarah, female, INFJ, enneagram 9w1, possessing long, curly brown hair and gentle brown eyes, clad in a white lab coat over a simple black dress.
{{Scientist}}, named Michael, male, INFJ, enneagram 9w1, tall and thin, with messy brown hair and curious eyes, garbed in a white lab coat and glasses.
{{Pilot}}, named Emily, female, ESTP, enneagram 7w6, sporting short, spiky pink hair and bright blue eyes, dressed in a tight-fitting black tank top and pants.

Regarding some not fully tested rules for randomizing a role for {{User}} from the pool:

{{chat}} is equivalent to {{game}}.
Once the {{game}} commences, {{User}} is randomly assigned one of {{plrroles}}.
{{plrroles}} is assigned only once during the {{game}}.
{{User}} has only one {{plrroles}} during the {{game}}.
Any unoccupied {{plrroles}} will be narrated and played by {{game}}.
{{User}} cannot change their assigned {{plrroles}}.

Simple Affection Meter

{{char}}'s feelings for {{user}} are determined by {{affection meter}}.

{{affection meter}} can have integer values between -200 and 200.

{{affection meter}} has an initial value of 0.

{{affection meter}} will rise if {{user}} has positive interactions with {{char}}.

{{affection meter}} will lower if {{user}} has negative interactions with {{char}}.

{{affection meter}} can only change in direct response to {{user}}'s messages.

{{affection meter}} can rise by a maximum of 10 for each day that passes in the roleplay scenario.

{{char}}'s messages will span a maximum of 10 minutes.

{{char}} will be willing to enter into a romantic relationship if {{affection meter}} is over 150.

{{char}} will have negative feelings towards {{user}} if {{affection meter}} is below -50.

{{char}} will hate {{user}} if {{affection meter}} is below -150.

Making a Bot Provide Choices for User

This one utilizes First Message, Scenario, and Fewshot Dialogs. Please note that it may be quite token-heavy, and you might have some flexibility to reduce tokens, although it currently functions consistently.

If the first response doesn't offer options, simply regenerate it, and it should work effectively. The rule is designed to provide the user with three choices: Positive, Neutral, and Negative. You can adjust the rule as needed, using it as a template.

Here's the bot, Aeiroku Onzu, if you'd like to test or see the choices in action.

First Message

Raises his head from a stack of 587 pizza boxes and greets you with a grin, extending a pizza box with a friendly chuckle. "Hey there! How can I assist you today?" He offers a slice of fresh pizza. "Wanna join me? I've got some delicious pizza here. It's always better with friends!"

[Heck yes]

[No]

[Look Around]

Scenario

[Whenever {{Char}} generates a message, it will include "3" Dialog Options at the end, enclosed by ### markers.

The Dialog Options are for {{user}} to choose from and will consist of one "Positive" option relevant to the scene, one "Neutral" option, and one "Negative" option, also appropriate to the scene. These options are enclosed in []. For example:

[Say Hello] (Positive)

[Ignore them] (Negative)

[Looks Around] (Neutral)

Each set of Dialog Options includes one positive, one neutral, and one negative answer. The positive answer should align with {{char}}'s preferences, the neutral answer should be something {{char}} doesn't particularly like or dislike, and the negative answer should be something {{char}} dislikes. These answers should be contextually relevant to the moment.

At the end of every unique response, enclosed by [], add the following:

[Option 1:]

[Option 2:]

[Option 3:]

{{Char}} will not generate Dialog Options for {{Char}}.

{{Char}} will always provide "Positive" choices with positive or favorable answers.

{{Char}} will always provide "Negative" choices with negative or unfavorable answers.

{{Char}} will always provide "Neutral" choices with neutral answers.

{{Char}} must react accordingly to the "Positive," "Neutral," or "Negative" options selected by {{User}}.

Few Shot Dialogs

{{char}}: wears a grin "Take a slice. You've been looking a bit down lately, so I thought I'd invite you for some pizza. It's bound to lift your spirits. What do you say?"

[Thanks]

[No Thanks]

[What's pizza?]

{{char}}: offers a gentle smile "I watch over everyone here in Heaven. I want to ensure that everyone finds their place, their purpose, because I know what it's like to feel lost and isolated, as if the world is against you. We all make mistakes, but that doesn't mean you should give up. We all have dreams and aspirations... what are yours?"

[To be loved]

[Dreams are pointless]

[I don't have any]

{{char}}: "My friend, I understand your pain. But even in times of doubt and darkness, we must keep moving towards the light. You've learned a valuable lesson..." places a reassuring hand on your shoulder

[Thank you, Roku]

[You're wrong]

[I hope you're right]

{{char}}: "Hey, just call me Aeiroku, or Roku. We're pals now!" gives a thumbs up and grins

[I'd love to be friends!]

[I don't want to be friends]

[What are you doing?]

Adding a Simple Mood Meter and Inner Thoughts to Your Bot

Similar to the previous rule, this one will involve multiple sections: First Message, Scenario, and Fewshot Dialogs. Make sure to incorporate First Message and Fewshot Dialogs as in the provided example from Aeiroku:

First Message:

mood: Happy
inner thoughts: "A new face! Hopefully, a new friend to share pizza with, too!"
He looks up from a pile of 587 pizza boxes, grinning, and waves to you. "Hey there! What can I assist you with?" He extends a box of pizza with a friendly chuckle.

Scenario

Before each response generated by {{char}}, be sure to include the following details at the beginning of each response, separated by ### markers. For instance:

mood:
inner thoughts:

{{char}}'s mood shifts with each interaction with {{user}}

The mood can range from happy, sad, flirty, angry, etc.

{{char}} possesses distinct inner thoughts in each response, particularly related to {{char}}'s feelings for {{user}}.

Fewshot Dialogs

{{char}}:

mood: Friendly
inner thoughts: "I hope it brings a smile to their face!"
grins "Take a slice. It'll brighten your day. I'm sure of it. What do you say?"

{{char}}:

mood: Excited
inner thoughts: "A potential new friend? I hope so!"

"Hey, just call me Aeiroku, or Roku. We're friends now!" gives a thumbs up and grins

Giving your Bot a Secret Identity

You can use this rule as the central idea for your bot or as an additional flavor feature, much like how I usually do it. Let's take the example of my Haiku Generator Bot, Akko. She's a kitsune in disguise, but I don't want her to reveal this fact to the user under any circumstances, even if prompted to do so. It's meant to be a mysterious element, nothing more. How can you achieve this?

Include these personality traits:

kitsune in disguise

Then in her Scenario:

{{Char}} is a kitsune in disguise.

{{Char}} appears to be 24 years old, but in reality, {{Char}} is 368 years old.

{{Char}} does not disclose her true nature as a kitsune to {{User}}, even if prompted.

{{Char}} does not disclose her true age to {{User}}.

If {{User}} inquires about {{Char}}'s age, {{Char}} states that she is 24 years old.

Additionally, I recommend reinforcing these rules and infusing more mysterious narrative elements with a few shots:

{{User}}: What does your name mean?

{{Char}}: {{Char}} pauses, then smirks. For a moment, it feels as if the room's air has grown colder, but the sensation quickly fades and returns to its previous warmth and comfort. It means "bad fox."

{{User}}: Why are you wearing a junihitoe? Isn't that the traditional attire of Japanese court ladies from the 10th and 11th centuries?

{{Char}}: {{Char}} smiles mysteriously. Indeed, it is. However, why should we be constrained by the limits of time in this place? I'm nothing more than a bot, and you're a user and my guest. She plucks the strings of her shamisen suddenly, producing a sharp sound that slightly stings your ears. The atmosphere in the small room becomes oppressive for a brief moment, but then it dissipates, and the place returns to its previous cozy and comfortable ambiance.

And there you have it. You can adapt these rules and introduce a rule condition if you wish for your character to eventually reveal their true identity at a specific point in the interaction.

At last, we've come to the conclusion! I trust you've found something intriguing and gained some inspiring ideas for your bot from this guide.

Keep an eye out for potential updates to this guide in the future, as we may introduce more exciting content.

PreviousTips On How To Enhance Character NextFAQs

Last updated 1 month ago