Skip to content

Rights and contracts

Media asset clearance dataset: is this asset cleared for this use?

By SourceX Editorial · Reviewed by Noah Loul ·

9 sections

Short answer

To verify that a creative asset is cleared for a particular use, write the request as dimensions (purpose, media, channel, territory, dates, exact version) and test each one against every rights layer: the license, talent and property releases, music and other embedded elements, plus approvals for that version. A media asset clearance dataset keeps that request, the evidence, the rule applied, the decision and a later outcome check together.

License terms

Media asset clearance

This page covers media asset clearance dataset: is this asset cleared for this use.Illustrative

Key takeaways

  • An asset that is findable in the DAM is not necessarily cleared; clearance is a judgment about one use of one version.
  • Break every request into the same dimensions so requests and rights records can be compared consistently.
  • Approvals and rights attach to versions, so a new crop or cut needs its own approval record.
  • Prove the decision with the documents it relied on and the outcome with a later record, such as a placement log or unpublish event.
  • Your clearance decision records and the third-party creative files they describe raise separate licensing questions.

How do you verify that a creative asset is cleared for a particular use?

Treat clearance as a test of one requested use against every rights record attached to one exact version of the asset. Write the request down as dimensions (purpose, media, channel, territory, start and end dates, the version being used), then check each dimension against the license, each talent and property release, any music or third-party element, and the required internal approvals. The asset is cleared only if every layer allows the request and the evidence for each answer can be shown.

That definition matters when you turn clearance work into a dataset. A model or agent that learns from your history should learn the reasoning that connects a request to its evidence, not just copy the final label. So each training or evaluation task in a media asset clearance dataset needs the request, the evidence references, the rule that applied, the decision and a later check that the decision held up.

Most creative operations teams already make these calls every week: an account lead asks whether last spring's hero image can run in a new market, or whether a video cut can move from organic social to paid placement. The answer usually sits across a digital asset management system, a contract folder, a talent tracker and an approval tool. The task pack below pulls those pieces into one record per decision.

Why does an asset being available in the DAM not mean it is cleared?

Availability only shows that a file can be found and downloaded; clearance is a judgment about a specific use. A DAM can hold an asset that is fully licensed for print in one country and not licensed at all for paid digital advertising anywhere else, and the file looks the same in both cases.

Rights features inside DAM platforms support the judgment but do not replace it. Adobe's documentation for Experience Manager Assets 6.5 describes asset expiration as a way to enforce license requirements: when an asset expires it is unpublished, and users without administrator permissions cannot edit, copy, move, publish or download it. The same page describes a license acceptance step, where the xmpRights:WebStatement metadata property points to the page holding the license agreement for a protected asset. Both features handle dates and access. Neither tells you whether a given territory, medium or placement is inside the license.

Embedded metadata has the same limit. The IPTC Photo Metadata Standard defines fields such as Model Release Status and Property Release Status, which summarize whether releases exist and how far they reach, plus Web Statement of Rights, a link to a page stating copyright and usage terms, and Licensor, the party to contact for a license. These are summaries and pointers. A field that says releases are unlimited is a claim that needs the release itself behind it before it can support a decision in a dataset.

Which dimensions of a requested use must each rights record match?

Each requested use should be broken into the same small set of dimensions every time, so that a reviewer, a model or an auditor compares like with like. The W3C ODRL vocabulary is a useful reference for naming them: it defines constraint operands for purpose, media context (with examples such as print, advertising and marketing), delivery channel, a geospatial named area for territory and a date or date-time for when the action happens.

Using a fixed vocabulary does not mean you must publish ODRL policies. It means the request and the rights record speak the same language, which is what lets a task be checked mechanically before a person reviews the hard parts. The table maps each dimension to the record that usually answers it.

Which dimensions of a requested use must each rights record match?
Use dimensionWhat the request statesRecord that usually answers itTypical mismatch
PurposeAdvertising, editorial, internal training, pitch workStock or commissioned license terms, statement of workEditorial-only license used in a paid ad
Media contextPrint, broadcast, digital, out-of-home, packagingLicense grant clause, rights-managed usage scheduleLicense names digital only; request adds out-of-home
Delivery channelOrganic social, paid social, website, retail partner siteMedia plan linked to the license or talent agreementOrganic cut moved into a paid placement
TerritoryCountries or named regionsLicense territory field, talent agreement territoryAsset cleared for one country, requested for a region
TermFirst use date and last use dateLicense start and end, talent use period, DAM expiration dateCampaign runs past the talent or music term
VersionThe exact rendition and edit being usedDAM version history, approval record for that versionApproval covers version 3; version 4 adds a new element
People and propertyWho and what is identifiable in the frameModel releases, property releases, location agreementsRelease covers one person; a second person is identifiable
Embedded elementsMusic, fonts, artwork, logos, screen contentSync and master licenses, font licenses, third-party permissionsMusic cleared for 30 seconds; request is a 60-second cut

What does one clearance decision record need to hold?

The asset this page offers is a task pack: one record per clearance decision, specific enough that a separate reviewer can rebuild the decision from the linked evidence without asking the original team. Use it as a schema for an evaluation set, as a structure for supervised training examples, or simply as an audit format for your own clearance desk.

At a glance

What does one clearance decision record need to hold

Every field below should be filled from a system of record, not retyped from memory. Where a value is missing, the record says so in the field rather than guessing; the review status then decides whether the task stays in the dataset.

What does one clearance decision record need to hold?
FieldWhat it holdsValidation ruleWhy it matters
task_id and pack_versionStable identifier and the pack schema versionUnique; version recorded on every recordLets you cite and re-run a task after the schema changes
asset_refDAM asset identifier, rendition and version numberMust resolve to one version, with a file hashApprovals and rights attach to versions, not to file names
use_requestPurpose, media, channel, territory, start date, end dateAll six dimensions present; dates in ISO formatA decision without a full request cannot be checked
rights_layersList of license, releases and embedded-element permissions, each with document reference and scopeEach layer has a document ID and an effective date rangeClearance is the intersection of all layers
approvalsApprover role, version approved, date, scope of approvalVersion approved equals asset_ref versionA sign-off on an earlier cut does not cover a later one
applicable_ruleThe clause or internal policy line that decided the outcomeQuoted or cited by document sectionTeaches the reasoning, not just the label
decisionCleared, cleared with conditions, not cleared, or needs new rightsOne value; conditions listed when presentA yes or no hides conditional clearances
outcome_checkWhat later happened: placement log, takedown, renewal or claimLinked record dated after the decisionSeparates a sound decision from a status field
counterexampleA nearby request that changes the answer, with the dimension that changedDiffers from use_request on exactly one dimensionShows the boundary of the rule
evidence_labelReal record or illustrative constructionRequired; illustrative tasks never mixed into evaluation scores without a flagKeeps invented cases out of measured results

Which revisions, talent terms or approver changes alter what may run?

Three kinds of change move the answer, and a dataset that ignores them teaches a model to trust stale approvals. The first is a revision to the asset: a crop that removes a logo can widen what is permitted, while a re-edit that adds a new person, a new track or a product shot can narrow it. The version field and the approval field must agree, or the task fails validation.

The second is a rights condition with its own clock. Talent terms in advertising often run on cycles separate from the photo or footage license. SAG-AFTRA describes its 2025 commercials contracts as setting a maximum period of use of 24 months, starting 10 working days after the first day of on-camera production, with holding fees due 13 weeks after that first production day. A clip can therefore be inside its footage license and outside the period its performers agreed to, on the same date.

The third is a change of approver scope. Brand approval, legal approval and client approval answer different questions, and a record that lists only the last sign-off loses the reason a version was held. Keep every approval with its role and the version it covered. When a later version arrives with only a brand sign-off, the correct label is often 'needs legal review', and that label belongs in the data as a legitimate outcome.

Worked task case (illustrative): one photo, two requests

This case is illustrative and built to show the structure; it does not describe any real client, campaign or asset. A creative operations team at a mid-sized agency receives two requests for the same lifestyle photograph, version 3 in the DAM, showing one adult model in a kitchen with a branded appliance in the background.

At a glance

Worked task case (illustrative): one photo, two requests

  • Starting state: asset_ref points to version 3; the license is a rights-managed grant for digital advertising in one named country for 12 months; the model release covers advertising use without a territory limit; no property release exists for the appliance brand; the DAM expiration date matches the license end date.
  • Request A: paid social in the licensed country, starting in month 4 and ending in month 9 of the license term. Evidence check: purpose, media, channel, territory and term all sit inside the license; the model release covers advertising; the appliance logo is a trademark question that the team's policy resolves by requiring a crop for paid use.
  • Decision for request A: cleared with conditions, using version 4, a crop that removes the logo, which then needs its own approval record. Outcome check: the placement log shows version 4 ran on the stated dates, and the DAM expiration unpublished the asset at the end of the term.
  • Request B (counterexample): the same version 4 for out-of-home posters in the same country. Only the media dimension changed. The license names digital advertising, so the decision is 'needs new rights', and the applicable_rule field quotes the grant clause.
  • What a model should learn: that one changed dimension flips the answer, that the crop created a new version with its own approval, and that a later placement log, not the DAM status, confirms the outcome.

What proves a clearance decision held up after the asset ran?

The decision is proved by the documents it relied on; the outcome is proved by a record created later. Keep those two apart. A clearance status of 'approved' in a work management tool is a statement that someone decided, not evidence that the use stayed inside its limits.

Good outcome records include media plan or ad platform placement logs with dates and markets, the DAM's own expiration and unpublish events, renewal paperwork when a term was extended, and any rights complaint or takedown that followed. Each outcome_check should link one of them and carry a date after the decision. Where no outcome record exists, label the task 'outcome unverified' rather than assuming the decision held.

This split also gives you a cleaner evaluation design. A model can be scored on whether its decision matches the reviewed decision, and separately on whether it flags the tasks whose outcomes later went wrong. Teams that only keep final statuses can measure the first and never the second.

Which clearance histories are too thin to keep?

Drop or quarantine a clearance decision when its evidence cannot be rebuilt by someone who was not in the room. Common reasons are approvals given by email or chat with no reference to a version, licenses stored as PDFs with no link to the asset they cover, and renewals negotiated by phone and never recorded.

Quarantine is better than deletion when the gap is fixable. A record missing only a version number can often be repaired from DAM history; a record missing the license itself cannot. Set a simple rule before the build starts: tasks with any rights layer lacking a document reference are excluded from training, and tasks lacking an outcome record are allowed only in a separately labeled split.

Stop and bring in a rights specialist when the history involves disputed ownership, union or guild agreements whose terms your team cannot interpret, or client contracts that restrict how campaign materials and internal records may be reused. Those questions decide what you may build, not just how to label it.

Who owns the clearance records, and who owns the creative files they describe?

Clearing a photo for a campaign and clearing your clearance records for AI licensing are different questions with different owners. The photos, footage and music in your DAM usually belong to photographers, stock libraries, labels and clients; your right to use them in ads does not give you a right to hand them to an AI developer. The decision records your team created (requests, rule citations, approval trails, outcome checks) are more often yours, subject to client confidentiality and any personal data inside them.

Machine-readable signals help you respect third-party terms when deciding what to include. Version 1.4 of the C2PA specification defined a training and data mining assertion with separate entries for data mining, AI inference, AI training and generative AI training, each marked allowed, notAllowed or constrained, and it says constrained is treated as notAllowed when no further information is given. The IPTC standard added a Data Mining property in 2023, and its specification notes that regional law may overrule the value a rights holder selects. Read both as signals to honor, not as grants.

An agency or brand team that wants to license its decision trails can bring them to SourceX, which carries them through rights review, contracting, preparation and delivery. No record moves until the team's company has agreed the scope and signed, and checking whether it qualifies takes a brief assessment rather than a file transfer. Third-party creative files belong in a licensed dataset only when their own licenses clearly allow it.

Frequently asked questions

Can a model release alone clear an asset for advertising?

No. A model release covers the person's likeness, not the copyright in the photo, the music in a video or the trademarks in the frame. Clearance needs every layer to allow the use, so the release is one row in the rights list, not the whole answer.

Does an expiration date in the DAM prove an asset was used within its license?

It shows the system was set to stop access after a date, which is useful evidence. It does not show where the asset ran before then, so pair it with placement logs or media plans before treating the outcome as verified.

What is the difference between 'cleared with conditions' and 'needs new rights'?

Cleared with conditions means the existing rights cover the use if something specific happens, such as a crop or a credit line. Needs new rights means no change to the asset brings the request inside the current grants, so someone must negotiate a new license or release.

Should illustrative tasks be mixed with real clearance decisions?

Keep them in the dataset only with an evidence label, and never count them in evaluation scores without saying so. Illustrative tasks are useful for teaching rare boundaries, but a model measured on invented cases will look better than it is on real requests.

Can an agency license its clients' campaign assets as AI training data?

Usually not on its own authority. The creative files typically belong to clients, photographers or libraries, and client agreements often restrict reuse. The agency's own process records may be licensable, subject to confidentiality terms, personal data review and counsel's reading of the contracts.

Sources

  • Adobe Experience Manager Assets 6.5 documentation describes asset expiration that unpublishes expired assets and blocks non-administrators from downloading them, and a license acceptance step where xmpRights:WebStatement points to the license page. Source
  • The IPTC Photo Metadata Standard defines Model Release Status, Property Release Status, Web Statement of Rights and Licensor fields that summarize release coverage and point to usage terms. Source
  • IPTC added a Data Mining property in version 2023.1, and the 2023.2 specification notes that regional laws may overrule the value selected. Source
  • The W3C ODRL vocabulary defines constraint operands including purpose, media context, delivery channel, geospatial named area and date-time. Source
  • SAG-AFTRA describes a 24-month maximum period of use under its 2025 commercials contracts, starting 10 working days after the first day of on-camera production, with holding fees due 13 weeks after that day. Source
  • C2PA specification 1.4 defined a c2pa.training-mining assertion with data mining, AI inference, AI training and generative AI training entries marked allowed, notAllowed or constrained, with constrained treated as notAllowed absent more information. Source

See if your company qualifies

A short company assessment. No data uploads are needed.

See if your company qualifies