Classify your image library before you move it into a new DAM

New DAMs come with AI tagging, and some can be trained on your own tags. That helps with search, but a migration needs more than search terms. It needs every image filed in the taxonomy you designed, a way to know which answers are safe to trust, and someone to check the rest before go-live. StayCharted gives you:

• A confidence on every answer, and a cutoff you choose for what a person checks.
• An accuracy report per category, measured on images the model never saw, before you run the whole library.
• Fine distinctions. In our part-identification benchmark, StayCharted named the right part out of 100 look-alike LEGO part numbers 87.8% of the time, from a photo alone, with 20 photos of each part to learn from.
• Retraining whenever you like. When your librarians correct answers in the Review Queue, those corrections go into the next version.

1. Train on the part of the library that's already filed well. Upload a ZIP of images, or connect the storage the old library exports to (Amazon S3, Cloudflare R2, Backblaze B2, Wasabi or Google Cloud Storage). Tell AMT where each image's category is: the folder it's in, its file name, or its own keywords, title or description. The model learns from the picture itself; the folder or keyword is only the label.
2. Check the report. AMT tests the model on images it didn't train on and shows how often it's right, category by category, and which categories it confuses. Add images to the weak categories and train again until the numbers are good enough for go-live.
3. Classify the rest of the library. Run the unfiled images through the model, as ZIPs or straight from the storage folder. You get the images back, or sorted copies in your storage, with the category added to each image's keywords. Everything else in the file is kept as it was: the image data, camera details, copyright and the image's other keywords. Unsure images go in a _needs-review folder with nothing written, and into the Review Queue.
4. Import into the new DAM. Most DAMs read embedded keywords on upload; Bynder, for example, fills its Tags field from the IPTC Keywords and XMP Subject of each file uploaded through its web app. Where the DAM can import metadata from a spreadsheet, use the spreadsheet of answers instead, matched by file name. Check one file in your DAM before the big batch.
5. Keep new uploads filed after go-live. Point AMT at the folder new images arrive in and rerun Fill new photos as new batches arrive. Use your storage provider’s notifications or your DAM’s supported automation to route the sorted copies into its import workflow.

Enable the option to add categories to copy metadata when you fill. We have not tested this complete import into a Bynder tenant; verify the keyword mapping with a sample before running your migration.

Most migrations change the categories. If your old folders don't match the new taxonomy, file a sample of each new category by hand first (20 or more images each is a good start) and train on that. The report shows which categories need more examples. Once the model is good enough, it files the rest of the library in the new categories.

• File formats. Keywords are written into JPEG and PNG files. AMT reads WebP, GIF, BMP and TIFF too, but returns them unchanged; their answers are in the spreadsheet for your DAM's metadata import. RAW, PSD, InDesign files and video aren't read. A JPEG rendition of a master file can be classified in its place.
• File size. Images up to 10 MB each. Export larger masters as smaller renditions for classification.
• Volume. Each plan includes a number of pictures a month. A full library migration is usually larger than that, so talk to us about Enterprise limits for the migration months.
• Consistent labels. If the old library says "Water damage" in one place and "water-damaged" in another, those are two categories. Merge them on AMT's Check your pictures screen before training.

• Product and parts libraries: file product shots and part photos by SKU or part number before they move into the new DAM.
• Packaging and label versions: separate the current artwork from old versions that look almost the same.
• Claims and inspection archives: file years of damage or inspection photos by damage type or condition.
• Agency and photographer uploads: classify a backlog of untagged shoots into product lines and campaigns.