AI-generated scripts often struggle with character consistency, where characters’ personalities, motivations, and speech styles shift unpredictably. This project aims to develop a framework to improve character consistency in AI-generated scripts and explore methods to improve it using structured prompting techniques.

Project Objectives: 1. To develop the Character Consistency Modelling Framework (CCMF) Create a multi-dimensional framework for modelling consistent fictional characters. •Synthesize insights from psychology (Big Five), holistic development (BMSEST, 3H Model), and narrative theory. •Define structured categories that the character will be modelled against •Justify each framework element in terms of its relevance to character coherence and narrative believability. 2. To operationalize the CCMF into usable prompt templates Translate theoretical constructs into practical, reproducible input structures for LLMs. •Design prompt scaffolds that ensure character realism and prevent consistency drifts across scenes. •Develop example profiles using the CCMF and dialogue instructions to guide consistent LLM output. •Test prompt adaptability across diverse character types and narrative scenarios. 3. To design a survey to assess character consistency using human evaluators Measure human perceptions of character consistency across generated scenes using the CCMF and basic prompts. •Develop rating criteria grounded in the CCMF’s dimensions (e.g., emotional coherence, behavioural alignment). •Construct a structured survey using both Likert-scale and qualitative prompts. •Recruit participants and guide them through evaluation tasks using sample character scenes. 4. To evaluate the effectiveness of CCMF in guiding consistent character generation Analyse how well the CCMF performs as a prompting tool for narrative consistency. •Compare participant ratings against intended character traits and behaviours. •Identify patterns in character drift or breakdown across different types of prompts. •Reflect on limitations of the framework and propose refinements or future directions.