CHARACTER PREFAB

Exactly what it says on the tin. I made this for myself so I wouldn't have to rewrite everything each time I made a new character. This is by no means the best or only way to structure a character definition, but it works well for me, so I figured I might as well share it. Here are a few notes about how to use it.

Update: A multi character prefab is now available here

(Note: You don’t need to give credit or anything if you make a public character with this template… but it would be appreciated if you did.)

• This is purely a template; mess with it and add/remove parts however you like.

   

• All headers with square brackets [LIKE THIS] should be removed after you've added your text. They're just wastes of tokens, as they don't add information on their own.

   

• Use the character name, e.g., "Lisa," instead of {{char}} when writing, the reason being that the LLM will be more likely to use the actual name in writing rather than something else. For example, if you name the bot "Lisa the serial parent disappointer" in the Character name field, the bot may refer to itself as "Lisa the serial parent disappointer" in writing, which gets weird, especially if you have a long name. The Character Chat Name field should fix this issue, but I've had issues with it, so I recommend writing the name instead. You should, however, still add the correct name in the Character Chat Name field for redundancy.

   

• Avoid super-specific details; they're mostly a waste of tokens as the LLM won't get them right anyhow, and even if the LLM does get them right, the average user probably won't stumble upon these super-specific details regardless in roleplay. Example: Instead of saying that Lisa is a 2nd year undergrad studying mechatronics, robotics, and fluid dynamics engineering at BumFuckNowhere community college for the vertically impaired. Say that she's a second-year college student studying mechanical engineering.

   

• Keep a resonable token count. Tokens are what the LLM uses to read and write text; if you add too many of them in the character definition, then there won't be much space left to remember dialog when chatting, which sucks. The permanent tokens, i.e., the tokens from the personality and scenario fields, should generally be kept around 1000, but you can go above that if you wish. The total token count on janitor is 9000, meaning that if you have 1500 permanent tokens, you have 7500 tokens for chat memory. When the chat memory exceeds this number, the LLM will start forgetting older messages.

   

• LLMs love tropes; if you write that a character is a jock, then the LLM may assume he is an asshole because the bully jock is a classic trope. This can be an advantage or disadvantage depending on the situation.

   

• The Initial message field is very important, not only for setting the scene but also for showing the LLM how to write. If you make it too short, you won't be able to set the scene properly, and no one likes that. On the other hand, if you make it too long, you show the LLM that it can yap ad nauseum, which no one likes either. Keep it balanced.

   

• The Example dialog field acts as a guide for how the LLM should write; it's far from mandatory to use this field, but it's useful if you're having issues with how the LLM writes.

• Character Tags should of course be accurate. Keep in mind that more tags =/= more people will find the character. If you have a female character that can be used by both male and female users, use the any pov tag rather than using both the <