Reference Image Tool Crashes With 1000+ Images: Fixes and Alternatives (2026)

A large collection can expose performance problems, but image count alone does not identify the cause. Start with a backup and a repeatable example. Image dimensions, cache settings, storage, and database state can matter as much as the number of files.

Correction, September 7, 2026: An earlier version treated individual Allusion reports as fixed database limits and gave unsupported card-count failure thresholds for Milanote and Miro. Those claims have been removed. This guide does not present a cross-product performance benchmark.

First, find the action that fails

Before changing tools, record the app version, operating system, available memory, and storage location. Then identify when the failure happens:

Change one thing at a time. A repeatable failure with a small set of files is more useful to support than an image count alone. Do not erase a database or overwrite your only board to test a theory.

PureRef: check the performance mode and scene contents

PureRef 2.x defaults to disk caching and offers other modes with different memory tradeoffs. Its current behavior should not be judged from older descriptions of all-in-memory loading. Preserve a scene copy, then compare the same scene under another documented mode. PureRef performance settings

The Manage images tools can change image resolution, format, and storage location. Some changes affect stored image data, so experiment on a copy and keep the originals. PureRef image management

A board holding references for one painting and a searchable archive of years of work are different tasks. You may find it easier to keep a smaller working board and a separate library. That is workflow advice, not proof of a fixed PureRef image limit.

Allusion: use issue reports as test cases

Two reports describe specific failures in Allusion 1.0.0-rc.10:

• On November 5, 2024, a Windows 11 user reported 14.4 GB of memory use while generating thumbnails for 358 images. They also reported normal memory use when browsing after the thumbnails existed. This makes first-time thumbnail generation a useful test to separate from ordinary browsing. Allusion issue #640
• On September 6, 2023, a Windows 10 user reported missing image results in a library of more than 120,000 files. They suspected a relationship with a database around 81 to 82 MB, but listed reproduction steps as unknown. Restoring an older database helped temporarily in their case. This does not establish a database-size ceiling for all libraries. Allusion issue #604

We checked these reports on September 7, 2026, but did not reproduce them. They do not establish that every current build or fork has the same problem, or that another app will handle the same files without failure.

If your symptoms match, include your version, OS, the action that fails, and a shareable sample in a bug report. Keep the full library and database backed up before trying a restore.

Milanote and Miro: separate board load from library size

For Milanote, this review does not establish a reliable card-count threshold for lag or crashes. If one board is slow, compare it with a small test board and record the media types and actions involved. A slow session alone does not show that the product has reached a fixed capacity limit.

Miro's official guidance says performance depends on board content, collaborators, device, browser, and connection. It identifies heavy images and documents as possible contributors and recommends trying a smaller board, another browser, or a duplicate of the affected board. Follow its current troubleshooting steps before concluding that you need to migrate. Miro performance guidance

Keep an intact copy before reducing image resolution or splitting content. A board can remain useful for sharing and planning even when you prefer to keep the full reference archive elsewhere.

How to evaluate another tool

Use the same representative files on the same machine. Record the build and settings for each app. Test the jobs you actually need:

1. Add your sample and wait for thumbnail generation to finish.
2. Close and reopen the app, then browse and search.
3. Build a working board with your usual image sizes.
4. Try the bulk selection and edits you use in normal work.
5. Check the media and metadata after export or migration.
6. Restore a small backup sample in a separate location.

Keep initial indexing separate from later browsing when recording timings. A fast empty grid does not show how the app handles your full workflow. Do not compare a small board in one app with an entire library in another and call the result a capacity benchmark.

Where refern fits

refern combines a searchable local library with a canvas. A normal workspace can index media already in a folder; explicit imports and copy actions can create files. Its database and generated thumbnails use an associated or configured data location.

Its streaming indexing and virtualized grid are implementation choices, not promises of fixed memory use, unlimited capacity, or crash-free bulk operations. This guide publishes no verified 100,000-item or million-item benchmark. Test the current build with your own formats and workflow before moving a large collection.

Before migrating

Use the source app's documented export options and inspect the result. Do not assume a board export contains every original image, note, link, or editable layout. Keep the source until you have checked what transfers.

For a PureRef board, consult the PureRef migration guide. For Eagle, use the Eagle import guide. Opening a normal folder is a separate workflow from using a dedicated importer.

Keep independent backups of source media, boards, and organizer data. An app's recovery feature does not replace those copies.

Try refern with a sample of your library, or compare reference managers for artists before choosing.