​

​

​​​THE MAKE OF You Only Have Seven Seconds

Bridging the gap between Community Participatory Data Visualization with Cinematic AI Art

​​

​

​​You Only Have Seven Seconds is a time-based cinematic work shaped through a multi-stage pipeline that combines real-time human-machine interaction, multimodal AI, data curation, visual processing, and experimental post-production. The project originated with a public, interactive installation where participants were invited to share whispered, seven-second memory fragments. These recordings were transcribed using OpenAI’s Whisper model, then semantically expanded through a GPT-based language model. Each resulting text fragment was transformed into a still image using a fine-tuned version of Stable Diffusion.

​

From these interactions, we curated 64 complete dataset triads—consisting of the original voice recordings, the GPT-generated texts, and the corresponding AI-generated images. These served as the core material for the film’s visual structure. To bring these static frames to life, we used a text-to-video generation model, producing 10-second animated clips that maintained thematic and visual coherence with the original prompts. These sequences were further organized into five semantic categories—memories of family, food, place, achievement, and shared moments of commonality—while intentionally preserving overlaps and fluid transitions. The film opens with a deliberate narrative illusion: disparate voices recalling moments with a father, edited to momentarily feel like a singular story before unraveling into polyvocal fragmentation. Throughout, we retained intentional irregularities such as mic tests, stumbles, and re-recording attempts as meta-fragments—preserving the rawness of the participatory act and foregrounding the project’s live, mediated origins.​ The visual sequences were not used in their raw form. Instead, we processed them through a custom-built system to develop the film’s aesthetic language. Drawing inspiration from monotype printmaking and slit-scan photography, our goal was to represent the ephemeral, unstable quality of memory. Each image was converted to monochrome, then distorted algorithmically using a GLSL compute shader with additional layers included blob-tracking to recursively warped to suggest memory degradation. ​Sound and image were composed in parallel. Using FFT, frequency, and amplitude analysis, we created audio-reactive visual behavior that allowed the film to respond dynamically to the spectral qualities of each voice recording. The result is a richly layered, affect-driven film that merges generative AI with procedural aesthetics—creating a temporal visualization of memory that is fragmented, speculative, and deeply human.

​​

​

​

​

​

​

​