The only online magazine devoted to the history, art, craft and conservation of stained glass

Conference Report: University of Münster Autumn School

Modern Stained Glass – Metadata – AI

Maximilian Berkel, University of Münster

Fig.1 Speakers and participants at the conference. © Universität Münster | ACHRIBI | Johanna Dierschke

From 16 to 18 September 2025, the University of Münster’s Arbeitsstelle für Christliche Bildtheorie, theologische Ästhetik und Bilddidaktik (ACHRIBI, the Centre for Christian Image Theory, the Aesthetics of Theology, and Image Didactics) hosted an interdisciplinary conference on modern stained glass and the possibilities opened up for it by automated metadata recognition and evaluation and the application of AI (Fig. 1).

The impetus for the event was a project funded by the diocese of Münster: Digitalisierung des christlichen Kulturerbes im Bistum Münster (DCK, Digitization of Christian Cultural Heritage in the Diocese of Münster). This multi-year research project aims to digitize systematically the c.700 churches and chapels found in the Nordrhein-Westfalen part of the diocese and to make the records accessible to all interested members of the public by publishing research data and mounting digital exhibitions. Important milestones have now been reached in this project, such as the development of a data model for use in an fylr database.1 The DCK team’s main focus has been on making new records of churches, as the majority of these buildings either had no inventory data, or incomplete data, some of which was in analogue form.

The venue chosen for the conference was Hotel SeeZeit on Lake Aasee, near Münster. For the first session, Norbert Köster and Maximilian Berkel kicked things off with an introduction to the concept to be explored through the conference. Reinhard Köpf then presented the ‘Situation of Modern Stained Glass in Germany’.2 He reported that as a result of damage occasioned during the Second World War, there is an enormous number of modern church windows in Germany, but that these have only been catalogued in a few areas. At the same time, these windows are now themselves under serious threat, with churches being deconsecrated and/or closed. While other furnishings from such churches can be moved to sacred spaces or museums elsewhere, stained glass – especially dalle de verre windows – is nearly always such an intrinsic part of the building that in most cases the decision is made to dispose of it as part of the demolition process. Furthermore, the contexts for windows, which were made for specific locations, are destroyed during demolition.

The second session dealt with ‘Databases, inventory and recording in the various countries’. Jasmine Allen began with a presentation entitled ‘Stained Glass in Danger’.3 She reported on databases established by various private initiatives that record endangered or demolished churches. Indeed, the founding idea for the Stained Glass Museum at Ely and its first acquisitions were windows removed from deconsecrated or demolished churches. In turn, the research results made available by the citizen scientists behind these initiatives are now in danger, as most of the database operators are of advanced age and the future of their work is unclear. In collaboration with the British Corpus Vitrearum, Allen had previously instigated a meeting at the Stained Glass Museum and proposed establishing a joint platform where these results could be collected and made available to the public in the long term.4

Martin Crampin then gave a presentation entitled ‘Stained Glass in Wales’, reporting on the various platforms with good metadata structures that he had established over the course of numerous projects.5  The treasure trove he had uncovered in the process is enormous, especially as it includes the holdings of the Art College in Swansea, where modern German glass-painters such as Johannes Schreiter (c.1930) taught British glass-painters. The difficulties in financing such research subjects makes reliable planning and the establishment of comprehensive and systematic catalogues very fraught in the long term.6

A project very similar to the DCK project, but on a much larger scale, has been underway in the Netherlands since the late 1990s. Anique de Kruijf reported on the ‘Project of digital inventories of all church and convent buildings in the Netherlands’, a long-term undertaking in collaboration with all religious communities throughout the Netherlands, in which sacred spaces were inventoried and catalogued digitally.7 The project is complete, and the inventories are now being adapted to modern data standards, transferred to a new database system, and revised. The database contains over 300,000 records and has a special feature, in that each community can view its own data online at any time and submit changes to the inventory team via the platform. This means that records can always kept up to date thanks to the work of local volunteers.

Karin von Wartburg concluded the second session with a presentation on ‘Vitrosearch’, the multilingual platform (German, English, French) on which the Vitrocentre, Switzerland’s national glass centre, publishes both the holdings of the Vitromusée, the national glass museum, and the research conducted by the Swiss Corpus Vitrearum, which has its seat at the Vitrocentre at Romont.8 As Vitrosearch is considered to set a good standard for the processing of academic data in the field of stained glass, Karin von Wartburg also presented the FAIR principles for research data (Findable, Accessible, Interoperable and Reusable). The DCK project and other projects also prepare data according to the FAIR standard.

At the end of the first day, Anja Gerber presented her lecture ‘Object Biography for Stained Glass Research’, setting out the possibilities offered by LIDO (Lightweight Information Describing Objects) and/or RDF (Resource Description Framework) models for in-depth indexing and compiling a complete object biography.9 One focus was on CIDOC CRM (the Conceptual Reference Model of the Comité International pour la Documentation) and the knowledge graph offered by the NFDI (Nationale Forschungsdateninfrastruktur). 10

While the first day of the conference had focused primarily on modern stained glass and the current state of affairs, the focus on the second day was primarily on what applications of modern technology are desirable or already possible.

Maximilian Berkel launched the day and the third session with his presentation ‘Concept studies: new ways of digital inventory’. As project manager and data curator of the DCK project, the question of what lessons could be learned from such a project and what adjustments could be tested in future projects were particularly relevant for him. Given the tension between the large number of windows and the considerable time pressure, the idea was raised of how modern technology should or could ensure these windows are at least preserved digitally. Digital preservation does not only mean photography, but also, for example, the recording of the space in which a window is situated, and time-lapse video recordings (as the perception of church windows changes with the time of day and the seasons); there is also the possibility of recordings made by drones to produce details and good results in a short time from windows installed high up. When it comes to mass work, AI can be used to make suggestions to a human data processor. A human processor must still be present at the beginning and end of the capture process, but the time-consuming mass processing can be effected gradually with the help of AI.

Folke Obermark-Stiller, a member of the DCK project team, reported on problems arising from this and other issues. With her presentation ‘Reports from everyday inventory work’, she gave a workshop report on site conditions and the problems that arise when the workplace is a sacred space that is in use. In archive and museum collections, it would be possible to clarify many issues quite differently, but working in a space in active use can be compared to performing open-heart surgery. At the end of the third session, there was a ‘Presentation of the data model and the project idea at ARCHIBI’. During this, the DCK project’s data model was discussed in detail, highlighting similarities and differences with the data models presented on the first day of the conference.

The fourth and final session was entitled ‘Possibilities of machine learning, automated metadata recognition and AI’. Christopher Pollin, opened with his presentation ‘Promptotyping a Stained-Glass-Metadata-Annotation-Tool with Frontier-LLM’.11  In addition to the opportunities afforded by LLMs (Large Language Models), for example in the mapping or enrichment of metadata, Pollin presented how processes and procedures could and must be broken down into many different individual tasks in order to limit sources of error (such as AI hallucinations), but also to comply with character limits in LLM chats.12

Dominic Eickhoff from the LWL Archives Office followed with a presentation on ‘Integrating Digitisation and AI-Supported Collection Workflows: A Critical Reflection’. Eickhoff reported on use cases in which AI-supported work integrations have been applied, as well as on the opportunities afforded by setting up an AI workflow meaningfully and, during the digitisation of objects, creating metadata that can be evaluated at a later stage. ‘AI use in practice’ was presented by Sebastian Klarmann from Programmfabrik, the developer of the fylr software being used in the DCK project. In a live demonstration, Klarmann presented various plug-ins that enable direct querying of language models in fylr, both in past projects and future ones.13

When talking about metadata recognition and the use of AI, the question naturally arises as to which metadata should or can be recognized. Etienne Posthumus from ICONCLASS explored this issue. The multilingual platform serves as a reference point for many projects for the consistent and precise cataloguing of iconographic content. Posthumus explored the question: ‘Now that we have large language models, are metadata standards still necessary?’. He averred that since AI statements and values do not produce the same results even in response to the same query, the question of comparability of results can only be answered by referring to models that offer comparability, i.e., standard data such as ICONCLASS. In the early stages, there were problems with querying such data models. Posthumus reported on recent attempts to use image recognition to build queries that were able to deliver valid results.14

In a sense, this presentation not only concluded the last session, but also served as a transition to the round-table discussion, which focused on how we could use machine learning and AI for digital inventories. As well as highlighting the needs of research, the workshop also made clear which scenarios have already been tested and are already available for use today. An image description may not fully meet art-historical criteria, but can deliver results that are sufficiently reliable, for example for citizen-science projects, that most users can work with. Machine learning is also very well developed and usable for the recognition and enrichment of standardized data, as it leaves little room for error. Such ideas can be particularly helpful for mass-processing tasks that are very strenuous for human operators to perform consistently. In the future, important human labour could be used for quality assurance and compliance with standards rather than for repetitive, individual work steps.

On the third and last day, the group visited churches and chapels in the immediate vicinity to view modern stained-glass windows in situ. Once again, it became clear what a cultural treasure modern stained glass represents. Some of the works viewed were by members of the ‘magnificent seven’, an expression coined by the glass artist Amber Hiscott to denote the seven great German stained-glass artists working after the Second World War, whose work was radically different from anything that had gone before: Georg Meistermann, Johannes Schreiter, Ludwig Schaffrath, Joachim Klos, Wilhelm Buschulte, Jochem Poensgen, and Hubert Spierling. The excursion began with a viewing of windows by Joachim Klos in the Liudgerhaus (Fig. 2), Münster’s Institute for Diaconate and Pastoral Services. At the nearby Liebfrauen-Überwasserkirche, attendees then saw works by Valentin Feuerstein (Fig. 3), which feature striking contemporary references, such as the depiction of a mushroom cloud. The group went on to the St.-Paulus-Dom (Münster Cathedral) to view windows by Georg Meistermann, his last works, in the chapels of the choir ambulatory, as well as windows by Wilhelm Buschulte in the tower. The tour concluded with works recently created by Johannes Schreiter for the chapel at Münster’s seminary, the Collegium Borromaeum (Fig. 4).

Fig.2 Window by Joachim Klos at the Liudgerhaus. © Universität Münster | ACHRIBI | Maximilian Berkel
Fig. 3 Admiring the stained glass by Valentin Feuerstein in the Liebfrauen-Überwasserkirche. © Universität Münster | ACHRIBI | Maximilian Berkel
Fig. 4 Johannes Schreiter’s recent work in the chapel of the Collegium Borromaeum. © Universität Münster | ACHRIBI | Maximilian Berkel

Acknowledgements

Organization: ACHRIBI

Conference concept: Maximilian Berkel

With thanks to Joseph Spooner for his editorial work. Without his help, publication of this report in English would not have been possible.

Footnotes

  1. A development of EasyDB. See https://fylr.io.[↩]
  2. Reinhard Köpf is currently head of the Untere Denkmalbehörde (Lower Monument Authority) in Mönchengladbach (Nordrhein-Westfalen). He was previously a research assistant at the Institute of Art History at Heinrich Heine University in Düsseldorf, as well as preparing inventories and expert reports for the Büro Rauch (https://www.buerorauch.de). [↩]
  3. Curator and director of the Stained Glass Museum, Ely (UK). [↩]
  4. There is a detailed report on this in Vidimus (https://www.vidimus.org/features/digital-future-of-stained-glass/, 17.12.2025). [↩]
  5. Centre for Advanced Welsh and Celtic Studies, University of Wales, Aberystwyth (UK). [↩]
  6. Future plans: https://www.vidimus.org/news/online-resources-for-wales-and-beyond/ (Last seen: 16.12.2025, 11:16). [↩]
  7. Head of the Erfgoed in Kerken en Kloosters (Department Cultural Heritage in Churches and Convents), at Museum Catharijneconvent, Utrecht, Netherlands.[↩]
  8. Vitrocentre, Romont, Switzerland: https://vitrocentre.ch (Last seen 22.02.2026). [↩]
  9. Klassikstiftung Weimar, representing the NFDI4Objects consortium of the NFDI.[↩]
  10. For CIDOC CRM, see https://cidoc-crm.org. For Gerber’s presentation, see https://zenodo.org/records/17140245 (accessed 17.12.2025). [↩]
  11. Digital Humanities Craft, Kainbach (near Graz), Austria.[↩]
  12. https://chpollin.github.io/stained-glass-metadata-annotation-tool/docs/version-2/index.html and https://github.com/chpollin/stained-glass-metadata-annotation-tool (accessed 16.12.2025). [↩]
  13. press release: https://www.programmfabrik.de/fylr-und-die-integration-von-ki/ (Last seen: 16.12.2025, 11:16). [↩]
  14. For the presentation, see https://zenodo.org/records/17151141 (accessed 17.12.25). [↩]