FIAT/IFTA World Conference 2026

Early 20th-century archaeological expeditions produced a rich body of visual material: film footage, field photographs, lantern slides documenting landscapes, excavation methods and cultural contexts. Dispersed across museums, university libraries and national film archives, these film collections remain difficult to discover, cross-reference or integrate into contemporary research. This presentation reports on an ongoing project that applies AI techniques to support descriptive cataloguing and access in digitized early film archives.
The project centers on a corpus of records from institutions in Europe and North America. Using vision language models, it explores how AI-assisted workflows help integrate contextual metadata, including site, estimated date ranges, artefact types. The presentation goes through opportunities and challenges of existing models trained on modern visual conventions that differ from historical image quality, aesthetic and stylistic composition. It observes how black-and-white historical footage with inconsistent framing and labels requires domain-specific approaches, and which solutions can be envisaged to preserve the richness of early audiovisual information.
Beyond cataloguing, the presentation examines how automated archival description affects and transforms cultural memory. Text matches and visual similarity now allow researchers to reconstruct early film sequences across archives and curate thematic collections that reframe and reposition our understanding of historical events. The paper reflects critically on the digitization of early footage and its entanglements with automation: which screen memories can machines unlock?
By sharing epistemological reflections on early documentary film culture, this presentation aims to contribute to broader conversations in the FIAT/IFTA community about the responsible integration of AI into archival workflows and the renewed relevance of historical film collections as forms of cultural mediation.