Vision-language models are increasingly used to caption images in museums, archives and libraries, from alt-text to collection metadata.
But what is a caption, and what is it for? In this challenge, we will generate and evaluate captions from different AI models for images from the collections provided by GLAMhack26, starting with the SKKG Digital Collection. Participants will also describe a selection of these images themselves in short Show and Tell sessions, reflecting aloud on how they read an image, so that human descriptions become part of the dataset. Comparing the two, we ask what models see and miss, and how captions are produced and introduced into GLAM institutions.
Using the human descriptions as a reference, teams will explore how to evaluate captioning models: which criteria matter for GLAM collections, and how human and machine descriptions can be compared. Bridging the gap between looking at an object and describing it digitally, the aim is a critical framework for human–AI interaction in GLAM captioning, one that asks when, how and whether machine captions should enter our collections.
Previous
GLAMhack 2026
Next project