Dxniz/NaNovel-27B
Dxniz/NaNovel-27B is a 27 billion parameter autoregressive language model from the Novelist series, fine-tuned on Dxniz/Novelist-CoT for creative writing, literary transformation, and instruction following. Based on Qwen3.5-27B, it excels at high-control scene writing, consistent voice work, and prompts mixing literary output with explanation, offering stronger prose control and narrative consistency. With a 32768 token context length, it is designed as a balanced flagship for users prioritizing output quality in creative writing tasks.
Loading preview...
NaNovel-27B: Flagship for Creative Writing
NaNovel-27B is a 27 billion parameter autoregressive model, part of the Novelist series by Dxniz, built upon the Qwen3.5-27B base. It is specifically fine-tuned on Dxniz/Novelist-CoT to enhance creative writing, literary transformation, stylistic analysis, and reasoning-heavy language tasks. This model utilizes a long-context supervised fine-tuning pipeline with explicit planning behavior, enabling it to reason about structure, emotion, pacing, and voice.
Key Capabilities
- Enhanced Prose Control: Offers superior control over writing style and narrative elements.
- Narrative Consistency: Maintains consistent voice and plot elements across longer texts.
- Instruction Following: Provides reliable adherence to complex creative instructions.
- Literary Analysis: Capable of prompted literary analysis and editorial explanations.
- Balanced Performance: Evaluated as the strongest autoregressive model in its series, performing well across prose, rewriting, translation, worldbuilding, and emotional continuity tasks.
Recommended Use Cases
- Long-form fiction drafting and high-control scene writing.
- Rewriting prose while preserving specific voices or styles.
- Literary analysis and editorial assistance.
- As a general-purpose Novelist model for balanced creative writing needs.
While offering high quality, users should be mindful of its larger hardware footprint compared to smaller models and the need for prompt discipline in very long multi-chapter planning. Generated text should always be reviewed for factual claims and continuity.