A rights clearance checklist before delivery

Most rights problems in generated work are found by the client, after delivery, by someone who was not in the creative conversation. That is a workflow failure: nobody owned a pass over the finished asset, so the pass never ran.

This is that pass. Five categories, run once, on the cut you are about to send. Every item names the failure it prevents, because a checklist whose items you cannot justify gets skipped.

General information, not legal advice. Anything material goes to counsel.

Run it on the preview, not the export

Sequencing first, because it changes the cost of the exercise.

The Versely editor is EDL-based: one timeline, re-renderable, so a fix is an edit to the list rather than a regeneration of everything downstream. A preview pass at 480p is free and carries a short per-user cooldown, and the final export is charged once regardless of how many clips are on the timeline. Clearance review therefore belongs on a preview render, where finding a problem costs an edit rather than a re-export.

The practical shape is a client review loop built on previews and share links, with clearance as the gate before the client sees anything. The timeline model is described on the video editing page.

1. Likeness

Prevents: a replica claim, a platform takedown, and the version of both where the subject is a real person your client knows.

Does any human in this asset resemble a specific real person? Recognisability is the test, not fidelity. A near-miss carries the exposure of a replica.
If yes, do you hold written consent naming this use — medium, platforms, territory, run dates, and whether any model artefact is retained afterwards?
Is there a voice as well as a face? A likeness grant is not a voice grant. Tennessee's ELVIS Act made voice a protected personal right from 1 July 2024. Anything involving voice cloning is a second clearance, always.
Is the person deceased? California's AB 1836 has required estate consent for digital replicas of deceased personalities in expressive audiovisual works and sound recordings since 1 January 2025, subject to listed exemptions.
Was consent obtained through a performer contract? California's AB 2602 (Labor Code § 927), effective the same day, makes a replica clause unenforceable as to new performances if it substitutes for in-person work, lacks a reasonably specific use description, and the person was unrepresented. A blanket clause with no described use, signed without counsel or a covering union CBA, may be worth nothing.

Two things people get wrong. There is no federal right of publicity in the US, so this is a state-by-state analysis turning on where the person's rights sit and where the audience is. And the NO FAKES Act, which would create a federal digital-replica right, advanced out of Senate Judiciary on 18 June 2026 as S.4591 and remains a bill. Do not plan against it as law.

2. Music

Prevents: a claim on the recording or composition, a monetisation block, and the quieter penalty where the track is lawful but the platform buries it.

Is the track generated, licensed, or library? Write it down. "It came with the project" is not an answer.
If generated, what does the provider's terms actually grant for commercial use in a paid ad? Read the tier you are on, not the marketing page.
If a voice sings or speaks, is it recognisably a real artist's? Content ID matches recordings and compositions, not style, so a soundalike reproducing no protected recording generally passes through. Passing Content ID therefore tells you nothing about right-of-publicity exposure. YouTube announced synthetic-singing identification inside Content ID in 2024; treat that as incomplete coverage, not as clearance.
Does the client know the track may be tagged or deprioritised on DSPs? Spotify adopted DDEX AI disclosure fields in credits in September 2025 and states disclosure does not affect royalties or recommendations. Deezer tags AI-generated tracks and excludes them from editorial and algorithmic playlists.

Label settlements are not clearance. UMG settled with Udio in October 2025 and Warner with both Udio and Suno in November 2025, each paired with forward licences for opt-in artist catalogues. Sony has settled with neither, and UMG v. Suno continues. As we set out in why a signed label deal is not a cleared model, a deal announcement and a licensed model shipping in the product are different events.

3. Trademarks

Prevents: an infringement claim, an ad rejection, and the awkward call where a competitor's logo is in frame at 0:04.

Scan every frame for marks, logos, packaging and trade dress that arrived uninvited. Models produce plausible branding without being asked.
Are any characters recognisable as protected properties? Studio litigation against Midjourney remains in discovery and targets outputs depicting protected characters. It is unresolved, and user-side exposure is real regardless of how it resolves.
If the asset creates a mark, a fictional brand or a product name, has it been checked rather than invented?
Does the asset imply a partnership or endorsement that does not exist? An integration badge or an "as seen on" treatment is a claim.

Provider terms do not help here. A terms-of-use assignment can only transfer what the provider holds, and none of them immunise you against a third party's infringement claim.

4. Provenance

Prevents: an unverifiable asset when the client asks how it was made, and a marking obligation quietly missed.

Does the delivered file carry Content Credentials? The current public spec family is C2PA 2.x, with a Conformance Program and Trust List gating who can sign credibly.
Did they survive your pipeline? The manifest is a hard binding to the file. Any re-encode, screenshot, format conversion or non-C2PA-aware editing step destroys or invalidates it, and major social platforms re-encode on upload. We traced this end to end in sign, strip, survive.
If provenance matters to this client, are you relying on Durable Content Credentials — manifest, plus an invisible watermark, plus a perceptual fingerprint? That combination survives recompression and is still defeated by heavy crop or adversarial removal.
Are you re-signing at export? Assume the original manifest did not make it unless every tool in the chain is C2PA-aware.

Treat C2PA as an EU AI Act compliance aid, not proof of provenance. Separately, the California AI Transparency Act as amended by AB 853 moves latent provenance marking by covered generative AI providers to 2 August 2026, with large online platform duties from 1 January 2027 and capture devices from 1 January 2028.

5. Disclosure

Prevents: a pulled ad, a fine, and a client discovering the obligation after the media is booked.

Where does this run? Territory first. Nothing else in this section is answerable without it.
EU: AI Act Article 50 applies from 2 August 2026. Deployers must label deepfakes, disclosed on first exposure rather than buried in metadata. Evidently artistic, creative, satirical or fictional works need only a disclosure that does not hamper enjoyment. Penalties run up to €15 million or 3% of worldwide turnover, whichever is higher. The grace period to 2 December 2026 for systems on the market before 2 August covers the marking obligation only, not deployer labelling.
New York: conspicuous disclosure when an ad reaching New York consumers features a fully synthetic human performer, effective 9 June 2026, at $1,000 for a first violation and $5,000 thereafter.
China: labelling measures effective 1 September 2025 require both explicit visible labels and implicit metadata or watermark labels.
UK: no blanket duty. The test is whether omission misleads, which catches synthetic endorsements and imagery implying real product performance.
Platforms: separate, contractual, stricter than law in places, and enforced by classifiers. Run the cross-platform labelling checklist rather than assuming one label satisfies five destinations.

The item everyone skips

Ask the client whether they need a protectable asset, because delivering the file and delivering copyright are different things.

In the US, human authorship is required and purely AI-generated work is unregistrable. The Supreme Court denied certiorari in Thaler v. Perlmutter on 2 March 2026, leaving that line in place. Prompts alone do not make the user an author. Human-authored selection, arrangement and modification are what is protectable, and AI-assisted works are registrable with disclosure of the AI material. If the client needs an asset they can enforce, log that human control while the work happens. Reconstructing it after delivery is close to impossible.

FAQ

Can this pass be delegated to whoever is exporting?

Partly. Frame-level checks for marks and recognisable people are fine to delegate. Territory, consent documents and disclosure duties are not, because they require knowing where the asset runs and what was signed. Split the checklist into a mechanical half and a contractual half, and give the second to whoever holds the client relationship.

How long does it actually take?

On a single short-form asset with no real people and no licensed music, ten minutes. It rises sharply the moment a named person, a real track or a multi-territory buy is involved, which is the correlation you want.

What do I hand the client at the end?

One page: what was generated versus captured, which models, what consents are held and their scope, what disclosure is applied and where, and whether Content Credentials survived to the delivered file. It is also the document that makes the next project faster.

Do I need this if we only generate backgrounds and B-roll?

A reduced version. Likeness and music drop out. Trademarks stay, because uninvited branding is a background problem more than a foreground one. Provenance and disclosure stay, because they attach to the finished asset rather than to the interesting parts of it.