The short version: migration is a metadata problem, not a file problem. Run a 100-file round-trip pilot before you sign. Find out which fields live only in the old vendor's database. Copy critical rights data into fields that survive. Budget a week per 50,000 unorganized files. Don't cancel the old subscription until you've run a real search, a real share link and a real download against the new library.
Why migrations lose metadata, not files
Every DAM can export your files. That part is never the problem. The problem is that a photo carries two kinds of information, and only one of them is guaranteed to travel with it.
Standard IPTC and XMP fields are written into the file itself. Move the file anywhere and the caption, the copyright notice and the keywords come along. But a custom field — an internal project code, a model-release expiry date, a product SKU — frequently exists only as a row in the vendor's own database. Export the file and that row stays behind.
This is why our Bynder review's advice is phrased the way it is: it's locked-in metadata, not your tool choice, that makes the eventual upgrade painful. The tool you pick today matters less than whether the data you put into it can leave.
What actually survives: our measured round-trip
We run the same test on every tool: write known IPTC/XMP fields into a file, import it, export it back out, and count what changed. These are the results across the six tools we've reviewed in depth — not vendor claims, our own measurements.
| Tool | IPTC fields preserved | What that means in a migration |
|---|---|---|
| Daminion | 100% | The only tool we tested with zero measured loss. |
| Pics.io | 94% | Second best; files never leave your Drive or S3 to begin with. |
| Canto | 82% | Captions and rights survived; creator contact and scene codes did not. |
| Brandfolder | 76% | Fine for finished JPGs; lossy for a working archive. |
| Bynder | 71% | Partial XMP export; built for finished brand assets. |
| Filecamp | 68% | Folder-thinking, not metadata-thinking — but files export cleanly. |
Percentages are field survival through a full export/re-import cycle against the IPTC standard, measured in our June 2026 test cycle. A lower number is not automatically disqualifying — it tells you how much reconstruction work a future move will cost. Full methodology in our metadata fidelity ranking.
In every migration I've watched go wrong, the loss was the same shape: captions and copyright notices survived, so the export looked fine. What vanished was structured data — creator contact details, scene codes, internal project references. Nobody noticed for months, because nobody searches on those fields day to day. They matter exactly once: when somebody asks who authorized a specific use of a specific image, and the answer used to be in a field that no longer exists.
The five steps
1. Run a 100-file round-trip pilot before you commit
Export a hundred representative assets from the old system, import them into the new one, export them again, and compare field by field. Do this before signing, not after. Our own measured round-trip survival across six tools ranges from 100% to 68% — the gap is entirely in structured fields nobody checks until they're gone.
2. Inventory what only exists in the old vendor's database
Standard IPTC/XMP fields are embedded in the files themselves and travel anywhere. Custom fields frequently live only in the DAM's own database. Anything in that second category will not appear in an export unless the tool writes it into extended XMP — list those fields explicitly before you move.
3. Duplicate critical rights data into fields that survive
If your round-trip pilot showed that creator contact or scene codes don't survive, copy that information into fields that do — caption, copyright notice, keywords — before the migration, not after. Rights and creator data is the category that costs real money to reconstruct.
4. Budget a week per 50,000 unorganized files
This is the figure we plan with, and it holds across tools: roughly one week per 50,000 unorganized files for content migration. Indexing itself is fast — on our tests, about 45 minutes per 25,000 RAW files over gigabit LAN. The week is human time spent deciding structure, not machine time.
5. Keep the old system readable until the new one is proven
Don't cancel the old subscription the day the import finishes. Run a real search, a real share link, and a real download against the new library first. A migration that appears complete can still be missing the one metadata field a rights audit will eventually ask for.
The one test that matters: before signing anything, export a hundred representative assets from your current system, import them into the candidate tool, then export them again and diff the metadata field by field. Include your weirdest assets, not your cleanest ones — the RAW file with a model release, the image with a scene code, the asset with a custom field somebody added three years ago. A migration plan built on your best-case files will fail on your real ones.
Which direction you're moving changes the risk
Migrating into a metadata-faithful tool is straightforward: whatever survives your old system's export gets embedded properly going forward, and the next move will be cleaner. Migrating out of a tool that stored custom fields in its own database is where the cost lands, because that data has to be reconstructed by hand or re-derived from a CSV export you hopefully took while you still had access.
Two practical consequences. First, if you know you might outgrow a tool, take a full metadata CSV export on a schedule while you still have a subscription — not on the day you cancel. Second, some tools make leaving genuinely easy by design: Filecamp keeps your folder structure intact and files export cleanly, which our review describes as making the move to a heavier DAM "a straightforward weekend job." Pics.io goes further — the files never leave your own Google Drive or S3, so there is no migration cliff at all.
What it costs, in time
Plan roughly one week per 50,000 unorganized files. That figure holds across the tools we've tested and it is not machine time — indexing runs at about 45 minutes per 25,000 RAW files over gigabit LAN. The week is people deciding what the folder structure and keyword hierarchy should be, which is work you're doing whether or not you migrate. A library that is already well-organized moves in a fraction of that.
Migration is also the moment when the accumulated cost of never organizing becomes visible: metadata that was never captured cannot be migrated, only re-entered. Every month spent postponing structure makes the eventual move more expensive.
FAQ
How long does a DAM migration take?
Plan roughly one week per 50,000 unorganized files. Indexing is fast — about 45 minutes per 25,000 RAW files over gigabit LAN in our tests — but the week is human time spent deciding structure and cleaning metadata, not machine time. A library that is already well-organized migrates far faster.
What actually gets lost when you migrate a DAM?
Almost never the files. What disappears is structured metadata: creator contact, scene codes, and custom fields that lived only in the old vendor's database rather than embedded in the file as IPTC/XMP. In our round-trip testing, field survival across six tools ranged from 100% down to 68%.
Should I run a test migration first?
Yes, and before you sign rather than after. Export a hundred representative assets, import them into the candidate tool, export them again, and compare field by field. This is the single check that separates a clean migration from one that quietly drops rights data.
Sources & references
- Metadata fidelity ranking — PhotoLib, IPTC/XMP round-trip measured across six tools, July 2026.
- Daminion review — 100% round-trip; indexing at ~45 min per 25,000 RAW files, June 2026.
- Canto review — 82% round-trip; the "week per 50,000 unorganized files" migration figure, June 2026.
- Filecamp review — clean folder-preserving export; migration described as "a straightforward weekend job," June 2026.
- Pics.io review — 94% round-trip; files-in-place architecture with no migration cliff, June 2026.
- IPTC Photo Metadata Standard — the reference our round-trip test is measured against.
- PhotoLib test lab — June/July 2026. See how we test.