The Democratisation of Digital Storytelling Using Large Language Models: A Case Study on Describing Coercive Control

Background
Coercive control is a complex and under-recognised form of psychological abuse involving patterns of intimidation, isolation, surveillance, and behavioural regulation within intimate relationships. Despite its serious personal and public health implications, awareness among young people remains low. Previous research in Northern Ireland found that only 16% of 16-year-olds reported understanding the term, highlighting a clear need for accessible and developmentally appropriate educational interventions. Previous work adopted a participatory, co-design methodology for the production of digital stories.
While effective, producing high-quality digital stories requires specialist skills, limiting scalability. This project investigates how large language models (LLMs) and generative media tools can democratise digital storytelling by lowering barriers to content creation while retaining co-creation principles and trauma-informed design.
Methods
Findings from Lagdon, Jordan et al. 's (2023) study, alongside insights from earlier workshops and stakeholder consultations, informed the development of three coercive control scenarios. These narrative scripts were revised in collaboration with young people to improve clarity, ensure age-appropriate language, and strengthen narrative coherence. The digital stories are rooted in themes identified by young people, including emotional dependence, psychological distress, stigma, peer support, and hope.
Visual assets were generated using Adobe Firefly and Midjourney to create youth-oriented imagery without the need for specialist illustration. The image creation process required iterative prompt engineering to control tone, character age, setting, and emotional expression. Experimentation revealed challenges in maintaining visual continuity across scenes—a known limitation of generative image models. To address this, Midjourney was employed due to its support for style parameters and reference-based prompting, which enabled greater consistency in character appearance, environments and aesthetic style across storyboards. This resulted in coherent visual narratives appropriate for the intended age group.
Stories were assembled and animated combining text, motion graphics, and automated text-to-speech to support multimodal accessibility. This AI-assisted workflow enabled rapid, low-cost production while maintaining ethical and trauma-informed storytelling practices. The stories are embedded in an evaluative questionnaire for distribution to young people at Ulster University to assess comprehension, engagement, and relevance.
Outcomes
The project produced three digitally animated narratives that depict overt and subtle forms of coercive control in relatable, youth-centred contexts to help young people understand coercive control in its various forms. Using generative tools substantially reduced production time and technical barriers, enabling researchers and non-specialist stakeholders to independently create professional-quality educational media. The multimodal format supports diverse literacy levels and learning preferences, improving accessibility and engagement. Feedback indicates that the stories are visually coherent, emotionally resonant, and developmentally appropriate for the target audience.
Conclusions
Integrating LLMs and generative design tools into participatory research workflows offers a pathway for the democratisation of digital storytelling in digital mental health education. Rather than replacing co-design, these technologies enhance it by enabling rapid, scalable, and inclusive content creation. Careful prompt design and controlled image generation techniques are critical to ensuring visual coherence and ethical representation. This approach provides a replicable model for developing accessible, evidence-informed resources that address sensitive issues such as coercive control, with potential applications across public health, education, and prevention initiatives.