The Transaction is based on an intelligent agent that utilizes diary-like text to create videos. Diverging from generative AI models that create fictional content, the agent integrates audiovisual excerpts of archival content from Radio Télévision Suisse (RTS, the Swiss national broadcasting company) and transforms text prompts into a montage of historical clips. The Transaction then transforms the montage into a Manga strip printed by a thermal receipt printer. Each receipt strip is augmented with a QR code, facilitating access for a closer look into the associated full-length archival videos with context.
This work operates on an improved narrative-aware language-video joint embedding space based on deep learning, where text inputs are translated into vectors, and used to retrieve historical archival clips to construct a new video. The final Manga strips are transformed from the generated video through state-of-the-art image generation and style-transfer models.
Playfully sitting at the intersection of memory, history, and narrative, this work uses the dialectical and serendipitous encounter with archival footage to destabilise the fixed linearity of history and archival narratives. This transaction between private memory and public history aims to reveal the fragility, plurality, and absurdity of the accepted norm, calling for a reexamination of our perception of the past and the present.