The Likeness License · v0.1.2
One license format for every likeness.
An open, machine-readable standard for licensing people, estates, and owned characters into AI production — think Creative Commons meets a SAG rider. One JSON document that an approval engine can evaluate, a compliance tool can parse, and a court can read. Published openly, because standards win by being adopted.
Illegal or talent-hostile terms are inexpressible.
Minors, irrevocable consent, waived AI disclosure, below-scale union deals, machine-granted exclusivity — none of it is against the rules; it fails schema validation outright.
Machine evaluation fails closed.
Anything not machine-decidable — free text, union coverage, exclusivity, persistent personas — routes to a human. Free text exists for courts, never as a matching input.
No floats. No FX. No fuzzy matching.
Money is integer minor units in ISO 4217; percentages are basis points; territories are ISO 3166 sets; platforms and models are registry IDs; competitor blocks are NAICS codes.
Consent is bound to the exact terms.
The consent record carries a SHA-256 hash of the canonicalized grant. “The licensor agreed to these specific terms” is machine-verifiable; post-consent tampering is detectable.
Specific consent is forced, not encouraged.
California’s “reasonably specific description of intended uses” standard is structural: a named project, a 200-character-minimum plain-language narrative, enumerated deliverables. Boilerplate can’t validate.
The grant
Twelve dimensions, all machine-matched.
Every dimension of the grant has a 1:1 mirror in the talent’s auto-approval parameters, so an agent’s booking request is evaluated by subset and threshold checks — never judgment calls.
Use category
Closed seven-value taxonomy: advertising, film/TV, games, social/UGC, corporate training, interactive persona, music/audio.
Media and channels
Voice and visual likeness enumerated separately — statutes treat them as distinct rights. No “all channels” value exists; every channel is an affirmative grant.
Territory
Worldwide, include-list, or exclude-list of ISO country codes — with explicit geo-targeting and organic-spillover terms instead of pretending the internet has borders.
Term
Explicit dates only; auto-renewal is unrepresentable. Living humans cap at ten years — aligned with the NO FAKES draft — and renewals are new documents with fresh consent.
Exclusivity
None, category-exclusive, or full — with NAICS-coded competitor blocks. Exclusivity is priced separately and can never be granted by a machine.
Output and impression caps
Explicit capped/uncapped election with a required overage policy — success has a pre-agreed price instead of a breach claim.
Approved platforms and models
How licenses stay portable but controlled: registry platform IDs, per-platform model scope, no wildcards — self-hosted generation only as a registered, audited pipeline.
AI disclosure
Required, never waivable: C2PA manifests plus human-perceptible labeling, with EU AI Act Article 50 handling built in.
Content exclusions
A 13-entry prohibited-use baseline that cannot be negotiated away, plus talent red lines from a 25-code taxonomy, machine-matched at request time.
Approval mode
Manual, auto-approve within parameters, or standing offer — with fee floors and ceilings, rate limits, and no auto-approval on silence, ever.
Pricing
Composable fee components — flat, per-output, per-impression, revenue share, exclusivity premium — with escrow choreography, milestone schedules, and staged kill fees.
Revocation
The crisp two-effect split: new generations cease immediately on every platform; delivered outputs follow a three-way election made at execution. Real, and priced honestly.
Pressure-tested
Four real deal shapes, encoded.
The schema was drafted from three lenses — legal completeness, machine executability, deal realism — pressure-tested against full deals, then put through two adversarial review rounds (a skeptical lawyer and the engineer who has to build on it) until every finding closed. All parties fictional; every file downloadable below.
- Licensorhuman · identity verified
- Useadvertising · paid social
- Term6 months · US + CA
- Price$3,500 flat
- Approvalauto — inside standing parameters
- Notablecompetitor category block · 2-platform allowlist
- Licensorhuman · SAG-AFTRA
- Usegames · voice + face
- Term10 years · worldwide
- Price$9,000 + 1.5% net rev share
- Approvalmanual · rider attached
- Notablevoice-model destruction obligations · in-game only
- Licensorcharacter · chain-of-title verified
- Useadvertising · JP + EN social
- Term3 months
- Price¥60,000 per clip · 20 cap
- Approvalstanding offer · executed by an AI agent over MCP
- Notablecharacter-bible adherence · no fan-content sublicense
- Licensorestate · authority documented, rep age-attested
- Usefilm/TV · documentary streaming
- Term2 years · bounded by post-mortem rights to 2059
- Price$45,000 · milestone schedule
- Approvalmanual · estate representative
- NotableCA §3344.1 post-mortem regime · re-verified at execution
The floor
What can’t be negotiated away.
Certain values in the schema are constants. Change them and the document simply stops validating — the ethics floor travels with the format, even for competitors who adopt it.
No minors, structurally.
Adult verification is a constant. A license for a minor’s likeness isn’t against policy — it’s schema-invalid.
Consent is always revocable.
Irrevocable consent cannot be expressed. Revocation stops new generation immediately, everywhere.
Biometric data is never sold.
Raw reference data stays in the Rights Vault by default; any release carries a 30-day retention ceiling and certified destruction.
Union scale is the minimum.
A below-scale union license is schema-invalid. Riders reference the governing CBA and travel with the contract.
Disclosure is never waived.
Every delivered asset carries C2PA credentials and a license-keyed watermark, and declares itself synthetic.
Read it · adopt it
The standard is public. That’s the point.
If this schema becomes how likeness licenses are expressed, every adopter strengthens the network that protects talent — including our competitors.
v0.1.2 · four worked examples · adversarial mutation suite (23/23 rejected) · two review rounds, 35/35 findings closed · schema $id is live at semblance.studio