User Avatar Image

This image is shown on your profile and across the platform. We recommend a square 1:1 aspect ratio image.

Public Description

This text is displayed on your public profile

Background Image

Background images are shown when chatting with a character. Having “Display Custom Chat Background Images” enabled will show either the creator added background image or the image you selected. We recommend a 16:9 image if on desktop or 9:16 image on mobile.

Response Styles

Relaxed (Default) - Characters will generally respond with shorter responses and will let the user drive the conversation forward.

Active - Characters will attempt to proactively move the conversation and will frequently take action and make narrative decisions.

Users note

Special instructions to the AI that get included with every message. This can be useful if you want characters to respond in a special format.

Advanced Settings

These settings allow you to finely control the way the AI responds to you. We have selected default settings that work well for most characters. Change these at your own risk!

Max new tokens

Determines the max number of tokens the AI will respond with.

Temperature

Temperature is a value between x….y. A lower temperature results in milder, more predictable conversations. A higher temperature results in more creative but sometimes unpredictable and incoherent conversations when taken too high.

Repetition Penalty

A higher repetition penalty makes the AI less likely to say the same things over and over again. While a lower repetition penalty will encourage the AI to repeat things. A repetition penalty that is too high will result in the AI not responding because words repeat themselves in real conversations.

Top K

For every token the AI wants to say - it chooses between a set of most likely tokens; Top K determines how many options (K) to consider. A Top K value too low might result in erratic behavior.

Top P

For every token the AI wants to say - it takes the smallest set of tokens whose probability adds up greater than P and chooses between that set. A lower top P value will leave out less likely tokens, and a higher top P will include a lot more choices. A higher top P value might allow for more variety and originality but at the risk of returning more incoherent text.