CreAgentive: Agent-Driven Narrative Engine

• CreAgentive System is a multi-agent creative generation engine that decouples narrative structure from stylistic realization using a Story Prototype.
• The system employs a three-stage workflow—Initialization, Generation, and Writing—to construct complex, long-form narratives while ensuring coherence and genre diversity.
• Evaluation via the HNES framework demonstrates that CreAgentive outperforms traditional LLMs in narrative quality, output length, and structural complexity.

The CreAgentive System is an innovative approach that addresses several core challenges in the field of creative generation using artificial intelligence. It leverages an agent workflow-driven methodology to tackle limitations faced by traditional LLMs in producing extended narratives and diverse creative outputs.

1. Introduction to CreAgentive

CreAgentive is a multi-agent creative generation engine developed to enhance the capabilities of AI in producing complex narratives across various genres. It introduces a framework that decouples narrative structure from stylistic realization, allowing for greater control over creative processes. This system emphasizes a structured representation called the Story Prototype, which serves as a foundational model for narrative development.

2. System Architecture

CreAgentive employs a three-stage workflow, utilizing multiple specialized agents to manage different aspects of the creative process:

1. Initialization Stage: This stage involves constructing a narrative skeleton based on user specifications. It serves as the basis for subsequent content generation.
2. Generation Stage: During this stage, agents collaboratively develop detailed plot and character arcs. This involves setting both short-term and long-term objectives guided by agent dialogues to instantiate the Story Prototype.
3. Writing Stage: The final stage uses the Story Prototype to generate multi-genre text that incorporates complex structures such as retrospection and foreshadowing.

3. The Story Prototype

A central feature of CreAgentive is the Story Prototype, a genre-agnostic, knowledge graph-based representation that encodes characters, events, and environments as semantic triples. This abstraction facilitates the management of narrative coherence over long texts and supports advanced storytelling constructs by maintaining detailed relationships and event sequences.

Key Features:

• Role Graph: Captures character attributes and relationships, supporting dynamic evolution over the narrative.
• Plot Graph: Represents interconnected events and scenes, enabling complex narrative interactions.

4. Addressing Limitations of LLMs

CreAgentive effectively mitigates several challenges associated with LLMs in long-form content generation:

• Genre Diversity: The system's architecture is designed to be genre-agnostic, allowing for cross-genre adaptation of narratives.
• Output Length: By utilizing a structured narrative representation, CreAgentive can manage extensive storylines extending over thousands of chapters.
• Narrative Coherence: Through incremental versioning and agent collaboration, the system maintains coherent and consistent story development.
• Structural Complexity: Supports non-linear narrative elements by leveraging the detailed structure and relationships within the Story Prototype.

5. Evaluation Framework: HNES

CreAgentive is evaluated using the HNES framework, which considers both quality and length of the generated narratives:

• Quality ($S_q$): Assesses various narrative attributes including coherence, creativity, and empathy.
• Length ($S_l$): Evaluates the extent and structure of the narrative.

Evaluation combines automated scoring with human assessments, ensuring comprehensive and robust measurements of narrative capability and quality.

6. Experimental Results

Experiments with CreAgentive show significant advances in narrative quality and length across diverse genres:

• The system consistently outperforms baseline models in terms of quality and scale, producing narratives that closely approach the quality of human-authored works.
• Across thousands of chapters, CreAgentive exhibits robustness and stability, demonstrating its versatility in narrative generation.

Conclusion

CreAgentive represents a significant leap forward in AI-driven creative writing. By employing a structured, agent-based approach, it addresses critical challenges such as genre diversity, narrative coherence, and structural complexity. This system not only enhances the capabilities of AI in creative tasks but also provides a scalable and efficient framework for generating human-like narratives across diverse applications.