How to train your AI Characters?
- Do not accept replies you do not like. If it is bad, or if a portion of it is something you don’t want, then do not reply. You must reroll or edit the response to avoid loops and poor-quality messages. If you reply, the AI thinks it did a good job.
- Thumbs-up quality comments. If you like the good comments, especially at the start, it lays out a foundation for how you want the AI to communicate with you
- If you get a duplicate message, reroll. If it continues, try redoing your previous message in a slightly different or longer way.
- Your replies are just as important as your Bios! Use the communication format found within this guide.
- If you enjoy asterisks or other formatting for actions and dialogue, there is no need to put it in the bio. The AI will follow your lead in the intro message most likely.
- Any NSFW discussion will trigger the NSFW AI. Be patient with this AI and remember to reply with LONG messages, otherwise, the AI will devolve into short replies that match your length. Don’t be afraid of rerolls.
- When creating a bio, if the bio contains profanity or explicit NSFW material, the character will not use the SFW DaVinci model. It will stay in the NSFW model and may impair the quality of the interaction.
- You are able to send an empty text box. This can be used to have your bot send a second reply without input from you, continuing where their last reply left off. You can also use this as an alternate means of re-generation - deleting the bot’s subpar reply and sending an empty text box will have the bot reply to your post as if it was the first time. (If you are re-generating replies several times in a row, it can be effective to try the delete/blank send)
- Please keep in mind, when a response from the character is regenerated, it counts towards our message limit, since a new message is requested when regenerate is processed.
Editing a message helps the AI learn! When the AI is communicating with you, it references back to the previous messages before it responds. When you edit the message, it will refer back to it and therefore implement the edits going forward.