Design considerations for photosensitivity warnings in visual media
Laura South, Caglar Yildirim, Amy Pavel, Michelle A. Borkin · 2024 · Proceedings of the 26th International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS '24) · doi:10.1145/3663548.3675643
Summary
This paper investigates how photosensitivity warnings should be designed for digital visual media to effectively help people with photosensitive epilepsy (PSE) navigate content safely. Photosensitive seizures can be triggered by flashing lights, strobing effects, and rapid colour transitions in video content, and while warnings are commonly displayed before potentially dangerous content, there has been almost no research into what makes these warnings effective. The authors conducted two complementary studies: first, a thematic analysis of 595 crowdsourced warnings from the DoesTheDogDie online forum (where users voluntarily tag seizure-inducing content in films and TV), and second, semi-structured interviews with five people diagnosed with photosensitive epilepsy about their experiences and preferences regarding warning design. For the interviews, participants were shown examples of both crowdsourced warnings (text descriptions from DoesTheDogDie) and a high-fidelity prototype of an automated warning system that might be deployed by a streaming platform. The automated prototype provided machine-detected information including the number of flagged sequences, their timestamps, and duration — information that crowdsourced warnings rarely include. The paper maps out the design space for photosensitivity warnings across three dimensions: warning structure (where and when warnings appear), warning content (what information is included), and data sourcing (whether warnings are crowdsourced, automated, or hybrid).
Key findings
Participants unanimously preferred warnings that provided specific, actionable information over vague general disclaimers. The most valued details were precise timestamps of when flashing sequences occur within a video, the duration and severity of each sequence, and descriptions of the visual characteristics (such as colour, frequency, and whether flashing fills the entire screen or is localised). Participants wanted the ability to make their own risk decisions rather than having content blocked entirely — they preferred to skip past dangerous sequences rather than avoid content altogether. The crowdsourced warnings from DoesTheDogDie, while valued for their human perspective and contextual detail, were inconsistent in format and coverage and sometimes contained spoilers. Automated detection was preferred for its consistency and ability to provide precise timing data, but participants also valued the descriptive richness of human-written warnings. Several participants advocated for a hybrid approach combining automated detection with human context. A key finding was that current industry practice — a single static warning before content begins — is insufficient; participants wanted in-stream warnings or controls that would let them skip ahead when dangerous sequences approach. Participants also noted that their photosensitivity varied by day depending on factors like sleep, stress, and medication, meaning their risk tolerance was not fixed. The thematic analysis revealed that crowdsourced warnings most commonly described flashing or strobing lights, rapid scene transitions, and flickering effects, with horror, action, and animated content being the most frequently flagged genres.
Relevance
This paper addresses an under-researched area of accessibility with direct implications for web and media platform design. WCAG 2.1 Success Criterion 2.3.1 (Three Flashes or Below Threshold) prohibits content that flashes more than three times per second, but much existing video content violates this guideline, and the standard does not address how to warn users when violations exist. For streaming platforms, social media sites, and any service hosting video content, the design recommendations are immediately actionable: provide timestamp-level flash data, allow users to skip flagged sequences, offer per-content rather than blanket warnings, and combine automated detection with human descriptions. The paper also highlights the inadequacy of a one-size-fits-all approach to risk — users want control over their own exposure rather than paternalistic content blocking. The main limitation is the small interview sample (five participants), though the authors note the difficulty of recruiting people with PSE for research. The combination with the large-scale crowdsourced data analysis strengthens the overall findings.
Tags: photosensitive epilepsy · seizure safety · video accessibility · content warnings · web accessibility
Standards referenced: WCAG 2.1