AuthorMix: Modular Authorship Style Transfer via Layer-wise Adapter Mixing

In three linesAuthorMix introduces a modular authorship style transfer framework using style-specific LoRA adapters mixed layer-wise. Trained on few examples, it outperforms SoTA baselines and GPT-5.1 while better preserving original meaning.

Summary generated by Claude — human-verified