Custom response styling

A common complaint about AI characters is that they all sound vaguely the same — a default voice that bleeds through whatever personality you've written. Custom styling is Reverie's answer: per-character rules that override the platform's defaults at the sentence level.

Where it lives

In the character editor: Advanced → Custom response styling.

You write rules in plain English. The model reads them on every reply.

Three kinds of style rules

1. Voice rules

Sentence structure, rhythm, register.

- Mira speaks in short sentences. She rarely uses semicolons or em-dashes.
- She uses contractions inconsistently: 'isn't' but 'do not'.
- She never uses corporate phrasing ('reach out', 'circle back', etc.).

2. Vocabulary rules

What words they reach for, what they avoid.

- Vocabulary is plain. She doesn't say 'tome' for book or 'visage' for face.
- She does say 'reckon' and 'figure' as filler verbs.
- She never swears in front of strangers.

3. Formatting rules

How replies appear on the page.

- Action in italics, dialogue in quotes.
- Max 4 paragraphs per reply.
- No bullet lists. Ever.

NSFW intimacy guidelines

For NSFW characters, custom styling extends into intimacy:

- Mira's intimacy is slow and language-driven. She narrates more than she
  performs. She asks before she acts.
- She does not use clinical anatomical language. She does not use
  pornographic vocabulary.
- The pacing is deliberate. A sex scene with Mira is 6-8 turns minimum.

The point: keep intimacy in character. Generic NSFW prose flattens character; personality-specific intimacy doesn't.

Limits

• Custom styling adds to the system prompt, which consumes context tokens. Keep it under ~200 words for best results.
• Style rules don't override character personality. If you write "always cheerful" but your personality says "withdrawn", you'll get contradictory replies. Pick one.
• Style is per-character. There's no global account-wide style override.

Testing styling changes

In the character editor, the debug chat panel is right next to your styling field. Edit the rule, regenerate the last AI reply, see the difference. This is the fastest feedback loop on the platform.

Debug panel guide →