Researchers propose Loom, a framework to improve creative writing assistance by separating narrative intent from rendering density

A persistent failure mode in LLM-assisted creative writing is a binary oscillation between two extremes: remedial polishing—safe, surface-level edits that preserve the original text but add little value—and destructive, uncontrolled plot expansion that violates the author’s intended structure. This tension defines a trade-off between narrative fidelity and descriptive intensity.

To address this, the authors propose Loom, an assisted-writing framework grounded in the narratological distinction between story (the sequence of events) and discourse (the presentation of those events). Loom employs a three-layer pipeline that enforces precise control over narrative intent and rendering density through an intent-centered semiotic chain-of-thought.

The architecture separates the generation of perceptual material—such as descriptive details—from syntactic insertion, ensuring that enhancements are applied without altering the underlying event structure. This design aims to preserve the author’s original intent while allowing for richer, more controlled narrative rendering.

The authors report that a comprehensive evaluation—combining LLM-based metrics and human assessment—shows Loom achieves the highest overall quality score among tested approaches. They claim substantial gains in factual integrity and descriptive intensity compared to state-of-the-art baselines, indicating that the framework successfully resolves the fidelity–intensity tension.