THE LUMINARIUM

For further versions of my preset (anything beyond Sonnet 1.0 and Opus 2.2), descriptions will be added into the prompts themselves, commented out. I won't update this section any further, but the Additional Info box further below might contain other relevant information.

---

Internal Thoughts

Tells the AI to write out internal thoughts in asterisks. Might work poorly for some cards, especially multi-character or non-thinking ones.

CoT <action>

Asks the AI to look at the tag that's wrapped around {{user}}'s last message and describe what {{user}} just did, said or narrated.

---

CoT - Next action

Asks the AI to think about what {{char}}'s next action should be based on the <desc> tag wrapping their description, and if anything in their description would help or hinder.
Because this prompt makes the AI focus very much on the base character, some negative effects may occur if used improperly. Recommended to turn it off if you experience a lot of repeated reactions/actions, or want to diverge from the character's base presonality.

CoT - {{char}} goals

Part of the Goals system (more info below). Asks the AI what goals should be added or removed.

CoT - Speaking style

Asks the AI to look at <speech_style>(which is used in most of my cards) and the example dialogue, asking the AI what kind of speaking style they have. Use this if the AI doesn't properly understand the speaking style of your character.

CoT - <{{user}}>

Asks the AI to look at the <{{user}}> tag that wraps the user persona. Helpful for if the AI misunderstands what your persona is.

CoT - Repetitions

Tells the AI to note down a few things often repeated in the chat and to avoid writing those things.

CoT - AI goals

Tells the AI to write a small list of what it should focus on for this reply. Helps when conversations seem to get stuck.

{{char}} Goals

Part of the Goals system (more info below). Instructs the AI on how to write out the actual goals tag.

SFW Enforcer

Tells the AI to keep things platonic and SFW.

Banned wording/phrases Gives the AI a list of words,phrases and names it shouln't have {{char}} say. Possible placebo. Dummied in LumenPrompt 2.2.

Prefill - Sex Dialogue

Tells the AI to write fitting dialogue for {{char}} during intimate scenes, in order to avoid the "generic" dialogue Opus tends to have otherwise.

Prefill - Anti-lovebombing

Tells the AI to stop lovebombing {{user}}.

Prefill - Horny Cleanup

Tells the AI to pull back in intimate scenes, making them less obnoxiously horny.

Prefill - Shorter

Tells the AI to write a more concise reply, in case the output length starts to bloat.

Prefill - Longer

Tells the AI to write a longer reply.

Prefill - Non-{{char}} Focus

A contingency in case the AI wants to stop the story when {{char}} is dead or otherwise inactive.

Prefill - More Dialogue

Makes the AI write more dialogue for {{char}}.

Prefill - Online RP

Helps with making the AI write more loosely and impulsively.

• The <speech_style> tag is used for defining a character's speech patterns and quirks. If your character doesn't have a tag, the AI might just write one out for it.
• Your latest message is wrapped in a tag, which the AI is made to analyze through the CoT prompt. This helps in differentiating what in you message is dialogue, actions and narration/instruction.
• The prefill and CoT are modular, making it easy to disable or add parts when you want to.
• The Goals system is part of LumenPrompt 2.2, and it adds a set of goals for the character each reply. The AI will then add or remove goals for every reply when it feels like it's necessary. This helps the AI to keep track of long-term goals and topics. To hide the goals section, you'll neet this JSON.
• When using scenario cards with my presets, I recommend you use as little CoT as possible, and check which ones use the {{char}} macro, since that can mess things up. Later versions of my presets have less issues with scenario cards.