AI Flavor Remover for Chinese Fiction

Problem

Chinese web fiction often contains formulaic phrases such as “缓缓”, “嘴角勾起”, and “not A, but B” structures, along with overused psychological declarations and tidy endings. These issues may pass simple keyword checks while still sounding synthetic. AI Flavor Remover treats this as post-generation editing for Chinese fiction: classify severity, apply rule-based rewrites, and verify naturalness.

How it works

• Diagnosis: score banned-word density, parallel passages, psychological-word share, dialogue-tag density, and sentence count per paragraph.
• Three-pass editing: Pass 1 removes generic phrasing and stock tags; Pass 2 trims formal connectives and boilerplate structures; Pass 3 restores rhythm with varied sentence length, concrete action, and restrained endings.
• Self-check: run from L1 hard rules to L4 human-readability review to avoid over-correction.
• Master-style layer: express emotion through actions, setting, absence, or objects, for example using “fingers turning white” instead of “he was terrified”.

Scope and caveats

It is better used as a post-processing step than as a creative writer. Context-sensitive items such as metaphors, foreshadowing, character voice, and chapter hooks are reported for human review. Punctuation rules keep exceptions for formatted labels like 【channel】 and 【system】.