Beta. Persona Created with Tarot Cards SFW

Please report any abnormal outputs or confusing points in the comments. You can simply write down the episode/model used and the problematic situation if you know them.

We've minimized tokens as much as possible so it can work even with models that don't have high reasoning capabilities (approx. 3,000 tokens). However, this required writing the lorebook tightly, which made the pre-use instructions somewhat lengthy. We ask for your understanding. We've also made it as usable as possible for non-Korean speakers.

Choose the desired item and input the cards yourself, or ask the AI to draw them. Even if the AI draws them, we've ensured that your preferences and settings do not influence the card selection. However, the AI's draws are not actual random draws and can be influenced by context, especially with lower-end models. If randomness is important, we highly recommend drawing and inputting cards yourself using physical cards or an external tarot website. (I personally use the canee tarot website.)

If automatic drawing or rule adherence falters, or if there's confusion with previous settings, please restart in a new chat.

You can choose between Episode 1, which creates settings, Episode 2, which only reads card meanings, and the LIGHT VER. Episode 3, which allows mode selection. If it's your first time, you can just pick one module you're interested in. There is a command to create a compressed version of the completed profile.

Q. I don't understand what you're saying even after reading the explanation.
A. Try switching to a cheaper model and asking, 'I want to do this and that, can you tell me how to use you?'... Would that work?

Q. I still don't understand.
A. You can also copy the spreadsheet at the bottom and ask GPT or Claude to read it for you...

Additionally, in prioritizing the naturalness of the character, we've observed a tendency to only recommend black, blonde, or brown hair colors. If you desire colorful hair colors, please specify in your instructions that you are considering hair colors that do not exist in reality... (The same applies to eye color...) If you want settings that go beyond our natural perception, meaning if you want a dramatic character, you must strongly appeal your preferences to the bot.

This bot is a dedicated bot for creating original characters using tarot readings. Do not use it for fortune-telling on real-world issues such as psychological diagnosis of real people, future prediction, love fortune, or medical, legal, and financial judgments.

While specialized for persona creation, it can also assist to some extent in AI character creation.

PC — Persona to be created

NPC — AI character to create a story with the PC

The bot prioritizes information provided directly by the user.

It is recommended to write detailed information about the main world setting, NPC information, confirmed PC details, and personal preferences in the User Notes or Persona Information section.

Settings generated in Episode 1 will be reflected in subsequent PC drafts and integrated versions. Please clearly state any content you dislike as follows:

It is recommended to organize settings to be discarded before the final output.

Interpretations and preference recommendations from Episode 2 are not automatically included in confirmed settings. If you wish to use them, please specify 'Adopt this setting' or 'Reflect as created setting'.

Episode 1 and Episode 2 have different result generation methods.

If you want detailed character settings, choose Episode 1; if you only want to check the meanings and possibilities of the cards, choose Episode 2.

You can also draw and input cards yourself. You can use physical tarot cards or a tarot drawing website, and the bot will prioritize the cards you provide directly for interpretation.

By default, both upright and reversed meanings are read. If you only want upright interpretations, please request it separately.

Requesting upright-only mode will apply it to subsequent readings. To use both upright and reversed again, input as follows:

When inputting cards you drew yourself, please include the question number, card name, and orientation.

If multiple modules are loaded, it's safer to include the module number.

Omitting the orientation can lead to misinterpretation, so specifying upright/reversed is recommended.

To save tokens, the question list has been placed in the lorebook. You must input the desired number, including the brackets, such as [01]~[13], for the bot to recognize it.

Due to the lorebook character limit, too many questions may not load at once, so we recommend loading a maximum of 4 questions at a time. The full question list is provided at the bottom.

Inputting only the module number, like [11], will only display the question list and will not automatically draw or interpret cards. After selecting the desired questions, input the cards yourself or request 'draw and interpret them for me'.

After loading the question list, you can input as follows:

You can still get readings by inputting the questions accurately even if the lorebook is not activated.

[13] contains a spread designed to explore your preferences, such as the relationships, emotions, conflicts, power dynamics, and ultimate narrative goals you wish to experience through the PC.

You can freely ask about things not present in the question list.

'Just create the spread for me' will present only the questions and recommended number of cards.

'Decide the settings for me' or 'Draw cards and create it for me' will proceed with spread design, card drawing, and interpretation.

Based on the creator's testing, models with high reasoning performance like Gemini, GPT, and Claude could handle up to about 10 cards at once relatively stably. It's recommended to use fewer cards for items requiring specific explanations and more cards for complex topics.

Models with high reasoning performance (Gemini, GPT, Claude) — Approx. 10 cards

Low-cost models — 5-6 cards

The total token count is around 3,000, exceeding the 2,000 tokens recommended by CaveDuck for low-cost models. With lower-cost models, errors such as repeating actions or not following instructions properly have been observed. (It might be more cost-effective to run it once with a high-reasoning model than multiple times with a low-cost model...)

Results may vary depending on the model version, conversation length, and the amount of information in User Notes/Persona. (It has been observed that models interpret clichés differently.) A higher number of cards may reduce the depth of interpretation for each item and the ability to remember settings.

For long readings, it's recommended to proceed in batches of 3-6 cards rather than all at once, and to organize adopted/excluded items in between.

To reduce missing settings during long-term creation, intermediate summarization is recommended at the end of each module or when approximately 4-8 readings have accumulated.

Rather than a general conversation summary, create an integrated version summarizing the PC settings adopted, modified, and excluded so far, and then update the important information in User Notes/Persona.

Starting a new PC creation in the same chat room can lead to a mix-up with previous PC settings and card records. It is recommended to create new characters in new chat rooms.

[90] and [91] have different purposes.

[90] is for final PC profile compression, and [91] is for pure interpretation/recommendation integration. Neither is for summarizing intermediate conversations, so during long-term creation, please create an integrated version first before using them.

Opus is recommended for using [90] and [91]. If you have updated key settings in User Notes/Persona Information, using Gemini is also acceptable.