In February 2025, the AI4Culture hackathon issued a challenge: use AI tools to innovatively transform, research, or showcase digital cultural heritage data. The AI4Culture platform contains a wealth of resources to feed these projects: datasets from cultural heritage institutions and the data space for cultural heritage, an arsenal of AI tools and training resources and upskilling materials.

Five teams took up this challenge and came up with engaging use cases combining innovative AI tools and cultural heritage datasets. Explore their projects below.

'We had a lot of fun exploring what data was available via Europeana, even if in the end we used other sources!' said participant Laurens Dhaenens.

ABC: Automating Blender Code

The team of Zita Baronnet, Francesco Gavioli and Lara Peeters aimed to extract 2D objects from images and convert them into 3D models using AI, automating the 3D modeling process. Their goal was to streamline the time-consuming and technically complex workflow of creating 3D models for cultural heritage applications. The team used the Europeana API, SegmentAnything, and the image to 3D asset HuggingFace model to create a pipeline that took a 2D image from Europeana and output a 3D model for reuse in Blender (software for creating 3D computer graphics). Explore their project.

Patina: de:color of time

This project from Stefanie De Winter, Laurens Dhaenens, Angelica Fieschi and Stefano Fanelli focused on addressing the issue of art degradation and patina formation, aiming to create a tool that identifies and communicates the aging of digitised artworks.

They sought to enhance public understanding of how time shapes art. The team created a training dataset of artworks that were affected by patina and aging, together with scans of those artworks before they were affected by age. They used a Convolutional Neural Network (CNN) for image classification and developed a web interface for science communication. The team identified the issue of having limited training data and the need for more accurate models.

'We look at these images all the time but we forget that they're shaped by time. We want to add this layer that is there but is invisible. What we want to show is the shape of time.' - Stefanie de Winter

Deepculture

The DeepCulture team, consisting of Ioannis Kapsalis, Katerina Zourou, Hannieh Habibi and Marianna Ziku, aimed to perform sentiment analysis on cultural heritage data to uncover hidden narratives and connections.

ArcAIVision

The ArcAIVision team of Sercan Kıyak, Knar Ohanjanyan and Elçin Istif Inci developed an AI tool to detect migration-related themes in historical videos, aiming to uncover hidden connections and forgotten stories in archival footage.

They extracted video content from the Netherlands Institute for Sound and Vision from the Europeana API, and used BERTopic for topic modelling and K-Nearest Neighbors (K-NN) clustering to analyse frames extracted from videos. The team addressed the inherent challenges in addressing biases in metadata. Even though the topic modelling AI correctly clustered similar themes together, the team manually clustered migration-related terms and frames.

'What if we could see history through a new lens - one that reveals hidden connections and forgotten stories? There are vast amounts of archival footage that are hard to navigate. It's not only hard to navigate but the traditional metadata is limited and often biased. This makes it hard to analyse, themes, emotions and more in these datasets' - Knar Ohanjanyan

Discover ArcAIVision's project.

Un2Structured

This project from Arnoud Wils aimed to extract structured JSON data from unstructured PDF files in the Corpus Rubenianum, focusing on provenance and iconography information.

They used Llamaparse, Llamaindex, Cohere LLM API, and Pydantic for data extraction and structuring. Un2Structured tweaked their template prompts extensively to get the results they were looking for. They also raised valid questions around how to valorise this extracted data.

Would you like to work on your own innovative projects using cultural heritage data? Explore the Europeana APIs.