Sudowrite

Story Engine generates plot outlines and story beats by analyzing narrative structure patterns (three-act structure, hero's journey, story arcs). The system likely uses a combination of template-based generation and LLM fine-tuning on published fiction to understand pacing, turning points, and dramatic tension. It maintains awareness of story progression state to suggest contextually appropriate plot developments rather than random ideas.

Expand capability extends existing prose while maintaining established character voice, dialogue patterns, and narrative perspective. The system analyzes voice markers in the input text (vocabulary choices, sentence structure, emotional tone, dialect patterns) and applies those stylistic constraints during generation. This likely uses prompt engineering with voice examples or fine-tuned models trained on character-consistent writing samples to ensure expanded text doesn't break character consistency.

Describe feature generates vivid sensory descriptions (visual, auditory, tactile, olfactory, gustatory) for scenes, objects, or characters. The system likely uses a sensory-focused vocabulary model and prompt engineering that explicitly requests multi-sensory details rather than visual-only descriptions. It may analyze the genre, tone, and existing description style to match sensory language intensity and type to the narrative context.

Brainstorm feature generates creative ideas (plot twists, character motivations, dialogue options, scene concepts) while maintaining awareness of existing story context and narrative constraints. The system analyzes the current manuscript state and generates ideas that fit the established world, character arcs, and story direction rather than producing disconnected suggestions. This likely uses context-aware prompting or retrieval-augmented generation to ground suggestions in the specific story.

The system understands narrative pacing and generates content (scenes, dialogue, descriptions, plot beats) that matches the intended pacing of the story. It likely analyzes existing text to infer pacing patterns (fast-paced action vs. slow character development) and generates new content that maintains consistent pacing rhythm. This may involve understanding scene length, sentence structure, action density, and emotional intensity as pacing signals.

The system adapts writing style, vocabulary, tone, and conventions based on the detected or specified genre (romance, thriller, literary fiction, science fiction, fantasy, horror, etc.). This likely involves genre-specific training data, vocabulary models, and convention libraries that inform generation. The system may analyze existing text to infer genre and apply appropriate stylistic constraints, or accept explicit genre specification to guide generation.

All generation features maintain awareness of the broader manuscript context, including character names, established plot points, world-building details, and narrative history. The system likely uses a context window or retrieval mechanism to access relevant manuscript sections and ensure generated content doesn't contradict or ignore established story elements. This enables coherent multi-chapter generation and consistent world-building across the manuscript.