Keep one example you can repeat
Choose a character, persona and opening message as a baseline. Change one control, repeat a similar prompt and compare the result. If you also rewrite the definition and switch models, it becomes difficult to know why the answer changed.
Use the selected model's defaults when you are unsure what a control affects. A setting copied from a different model may use another scale or interact with a different definition. Current controls are a more useful reference than an unexplained number from an old post.
Read reply length as its own setting
The plan selector lists maximum AI message-length settings of 225 for Free, 275 for Standard, 325 for Premium, 450 for Luxe, 550 for Elite and 650 for Imperial. These are the published control values, not a confirmed word count or token unit.
Luxe, Elite and Imperial list adjustable message length. Increasing it may allow more description, but it will not repair missing scene information. If an answer repeats itself, a shorter and clearer instruction can be more useful than a higher limit.
Give style instructions a clear place
For example: 'Use short paragraphs. Put spoken dialogue in quotation marks. Leave my character's actions to me.' Keep these instructions separate from the actual events, and check that examples in the card do not demonstrate a conflicting format.
For a content-diversity control, use the range shown by the current model and make a small comparison against the baseline. Avoid assuming that a larger number improves accuracy. If a response becomes incoherent, restore defaults before adding more instructions.
Common questions
What is a sensible starting point for chat settings?
Use the selected model's defaults with a clear persona and opening. Identify one unwanted behaviour, change the related control or instruction, and compare a similar exchange before making the next adjustment.
More on this topicDoes raising message length guarantee more words?
No. The control affects the available response setting, while the prompt, model and character still influence the answer. Its displayed number should not be converted into an unsupported word-count promise.
More on this topicDo I need a special command for style changes?
Plain language can state a perspective, response language or desired format. A slash command from another service is not automatically supported by our chat controls.
More on this topicIs there one correct content-diversity value?
Start with the selected model's default and use its displayed range. Compare one small change using the same scene; if the reply becomes less coherent, restore the previous setting.
More on this topicWhere can I change Dark, Light or System appearance?
Use the appearance selector. System follows your device's colour preference. The display theme is independent of content preferences and conversation-model settings.
More on this topic