Collections numériques
Museum Open Data and IIIF: From Single Downloads to Cross-Collection Work
Separate downloadable images, open metadata, and IIIF endpoints, build working image URLs, and know where bulk harvesting stops.
Most people use a museum website the same way: search for the object, open the large image, save it. That reaches a small part of what open collections actually offer.
Research, teaching, and comparative work need three things instead: reusable images, machine-readable metadata, and a way to join the two. Institutions differ widely in how much of each they provide.
Three kinds of “open”
1. Downloadable images
A download button, usually at one to three resolutions. The most common layer, and enough for single-object use.
2. Open metadata
Records released as structured data — CSV, JSON, linked data — often under CC0. It tells you what exists and what it is; it does not include the image.
The Smithsonian Open Access programme is the model here: images and metadata released together under CC0, ready to drop into a dataset.
3. Endpoints (API / IIIF)
Programmatic access to images or metadata. IIIF matters most because it works across institutions.
Key distinction: CC0 metadata does not imply CC0 images. The two are declared separately and must be checked separately.
What IIIF is for
IIIF has two core specifications:
- Image API. Retrieve an image by URL, specifying region, size, rotation, and format.
- Presentation API. A manifest file describing a set of images and their structure — order, volumes, annotations.
The Image API URL pattern is roughly:
{server}/{identifier}/{region}/{size}/{rotation}/{quality}.{format}Requesting the full image at 1200 pixels wide as a JPEG looks like this:
https://example.org/iiif/obj-123/full/1200,/0/default.jpgThe parameters that matter most:
| Parameter | Meaning | Common values |
|---|---|---|
| region | Which area | full / pct:10,10,40,40 / 100,200,800,600 |
| size | Output dimensions | max / 1200, / !800,800 |
| rotation | Degrees | 0 / 90 |
| quality | Rendering | default / color / gray / bitonal |
| format | File type | jpg / png / tif |
The practical consequence: you can pull one detail at full resolution without downloading the entire file. For handscrolls, maps, and print inscriptions, that is the difference between reading the text and guessing at it.
Where to find it
- The Art Institute of Chicago publishes both IIIF and open data, with the endpoints documented for developers.
- The Rijksmuseum offers open data and an API; some endpoints require a key.
- The Getty supports IIIF across its collection and publishes research datasets alongside it.
- Gallica, the Bibliothèque nationale de France digital library, exposes IIIF for prints, photographs, and manuscripts.
- ColBase, the joint database of Japan’s national museums, provides high-resolution images with commentary, useful for East Asian material.
- The Met’s open access collection offers downloadable files and usable metadata.
Search for these terms: IIIF, API, Open data, Developer, Rights and reproductions. They are rarely in the main navigation — look in the footer or under Research / About.
Three things this is actually good for
Comparing details
Different impressions, versions, or states of conservation of the same work, cropped with identical parameters, will show differences immediately.
Annotation and teaching
Manifests load into IIIF viewers, which support region annotation, walkthroughs, and classroom demonstration. You do not have to host the images yourself.
Bulk research
Pull metadata and count: collection composition by decade, subject distribution, an artist’s spread across institutions. Fields differ between institutions, so map them first.
Where bulk harvesting stops
- Read the terms first. Many institutions limit request rates and require an identifying User-Agent.
- Do not mirror the site. It is technically possible and will cause problems for you and for them.
- Cache, do not re-request. Fetch a record once and keep it.
- Keep source and rights. Store the record URL and the rights field with every row.
- Respect robots.txt and quotas. It is courtesy, and often also a contract term.
Common mistakes
- Assuming a zoom viewer is an API. A high-resolution viewer on a page is not necessarily programmatic access.
- Ignoring rights inside a manifest. Manifests declare rights per image, and they can differ from the site-wide terms.
- Treating CC0 metadata as CC0 images. Separate declarations.
- Assuming IIIF is free for all uses. The endpoint is free; the image may still be priced for some uses.
- Comparing fields across institutions without mapping. Units, date formats, and attribution rules all differ.
Checklist
- Do I need images, metadata, or an endpoint?
- What are the separate licences for images and metadata?
- Do I need an API key, or a request-rate cap?
- Have I stored record URLs and rights fields?
- Have I mapped fields before comparing institutions?
What to do next
- Licensing and credit lines: Image licensing and citation.
- To build an ownership chain from a collection record: Provenance research.
- Video, installation, and media archives: Digital art and media archives.
- High-resolution East Asian material: Finding Chinese and East Asian art online.
- To browse entry points: Collections.