CraftAlign: New Framework to Detect and Guide AI Story Revision

A novel framework named CraftAlign has been launched to assess and improve AI-generated narratives. As outlined in a paper available on arXiv (2608.01377), this framework tackles the challenge of AI stories often appearing clichéd and unnatural, resulting from over-explanation, predictable causal sequences, and stereotypical conclusions. CraftAlign enhances the alignment of AI narratives with human storytelling techniques by evaluating writing patterns and offering revision suggestions. It consists of two learning modules and a guidance pipeline for inference. A feature estimator based on Qwen3.5-9B forecasts 304 distinct writing attributes related to style and narrative structure. The framework seeks to facilitate various revision approaches and influence overall changes in information flow, causal arrangement, and ending styles, surpassing traditional detection and evaluation techniques that typically rely on source labels or aggregate scores.