Metadata that generated assets need
A generated asset needs everything a photographed one needs, plus provenance. Add fields for whether AI was used and how, fully generated or edited; the model or models; the brief and campaign IDs; the source and reference assets it was made from, such as a product pack shot; approvals with names and dates; and any real person, voice or licensed element involved. The IPTC photo metadata standard has a Digital Source Type property for exactly this, with values such as trainedAlgorithmicMedia for fully generated images. Meta reads these values when deciding whether to apply its AI labels, and Google Merchant Center requires them on AI-generated product images, so fill them in consistently.
Rights and expiry travel with the file
Rights metadata matters more for generated content, not less. A generated scene can still contain licensed elements: a person's likeness under a time-limited consent, a licensed font, a music bed, a product photographed under a usage agreement. Record usage terms per asset: permitted media, territories, start and end dates, restrictions. Set expiry dates so the DAM warns before an asset lapses and blocks downloads afterwards. Inherit rights from source files automatically where the system allows, because a variant made from a restricted master is restricted too, and nobody will work that out from a file name.
Findability once volume explodes
Search is where a DAM full of generated content succeeds or fails. People look for assets by product, campaign, market, format and approval status, so those fields must be controlled vocabularies, not free text typed differently by each uploader. Automatic tagging of what an image shows helps discovery but is no substitute for business metadata such as SKU and campaign. Store variants as a family under their master, so a search returns one campaign image with its ratios and languages attached rather than forty unrelated files. Make approval status and expiry filterable by default, so expired or unapproved assets never top the results.
Avoiding duplicate sprawl
The quickest way to ruin a DAM is to sync every generation into it. Hundreds of near-identical candidates bury the approved asset, and search results fill with rejects that someone will eventually publish. Keep exploration in the production system and push only approved finals, with their variants, to the DAM. Check that the DAM keeps embedded metadata when it creates renditions, because some pipelines strip it, and C2PA Content Credentials with it. The studio connects to DAM systems and logs the model behind every asset it makes, so provenance can travel with the file instead of living in a separate spreadsheet.
Updated 25 September 2026 · General information, not legal advice. Rules change, so check the current text with your legal team before relying on it.