I'm talent I'm a business I'm an AI lab
v1.0 May 2026 Twinnin Standards Adopted across UK broadcasting, advertising & production

The industry standard
for AI humans.

Twinnin clears the human. You create the work. The Twinnin Framework is the operating standard for using verified digital humans safely, respectfully, and creatively in AI-assisted production. Already in use by broadcasters, agencies, and studios.

Download v1.0 PDF → Talk to us about adoption
Why the framework exists

Three legal regimes. Zero implementation guidance.

In 2026, four pieces of law and contract govern how AI can use a real human's face and voice. They all require verified identity, specific consent, scoped licence, and auditable provenance.

None of them tell production teams how to actually do it. Which tools are approved. Which uses are prohibited. What records to keep. How to handle scope changes.

The Twinnin Framework is the gap-filler. It translates regulation into operational standards a creative team can ship against on a Tuesday.

US NO FAKES Act 2026

Establishes federal property rights over digital likeness and statutory damages for unauthorised AI-generated digital replicas.

California AB 2602 Active

Requires reasonably specific descriptions of AI-likeness use; invalidates non-specific consent contracts under Labor Code §927.

EU AI Act 2 Aug 2026

Mandates transparency, labelling, and documented consent for AI-generated human likenesses in EU-distributed content.

SAG-AFTRA Digital Replica Rider 2025

Operationalises consent and compensation requirements for AI-generated likeness in US commercial production.

Core principle

Every twin is a real person. Treat them as you would talent on set.

A Twin is not a stock asset. Not a random AI face. Not a synthetic character with no human behind it. It is a licensed human likeness — and the framework exists to keep it that way.

The model

Four layers. One signature.

Twinnin runs the clearance layer for digital humans. Every twin moves through a fixed sequence — verified at each step, recorded for audit.

Layer 01

Identity

Government-ID verification confirms the twin is a real person connected to the profile. Biometric liveness check at registration.

Onboarding · biometric · KYC
Layer 02

Consent

Recorded permission for the approved use and any agreed creative parameters. Categorical exclusions locked at registration.

Per-use approval · revocable
Layer 03

Licence

Specific scope, term, territory, exclusivity, and rate. Issued per project; no buyouts, no perpetuity by default.

NO FAKES / AI Act compliant
Layer 04

Provenance

Every output traceable. Every sample cryptographically signed. Audit trail available to talent and client at all times.

SynthID-compatible · auditable
Industry adoption

Developed in consultation with the industry that has to live with it.

The framework was shaped by the people building, regulating, and performing in the AI-likeness space. We brought them together at Royal Holloway in April 2026 to debate the standard — and ship it.

Hilda Cronje at Royal Holloway summit, April 2026
Featured panellist

Hilda Cronje

West End actress · Twinnin Twin

"As performers, we've always traded on our likeness. The difference now is the scale, the speed, and the silence. Twinnin is the first place I've seen where a working actor can actually say yes — on terms they understand."

Summit speakers included representatives from
ITV · Havas · RADA · Equity · AI Cinelab · Royal Academy of Art

Hosted and moderated by Katrien Grobler · Royal Holloway, University of London · April 2026

"AI isn't the end of rights — it's the next negotiation. The laws of the day are not keeping up with the tech, so we stepped in and decided to step up. We didn't host a debate — we exposed a fault line. Now the question is: who builds the system that fixes it. We have, at Twinnin."

Katrien Grobler · Founder & CEO · April 2026 Summit
The red list

What the framework does not permit.

Eight categories are explicitly prohibited without separate, specific approval. These are non-negotiable — locked at registration, enforced through licensing.

Ethnicity alteration

A twin's ethnicity cannot be changed in any AI-generated output without express written permission from the individual.

Sexualised content

Not permitted in any commercial or non-commercial use. Locked at registration; cannot be unlocked by any licensing request.

Hate, discrimination, or slurs

Prohibited absolutely. Use in any content depicting hate speech or discrimination voids the licence and triggers takedown.

Humiliation or ridicule

Twins cannot be placed in degrading, mocking, or demeaning contexts without specific consent.

Political messaging

Specific separate approval required. Categorically excluded by default at registration for most twins.

Disability portrayal

Requires careful, specific approval. Cannot be added or removed from a twin's likeness without explicit permission.

Deceased-person recreation

Separate estate / rights process required. The framework does not authorise post-mortem use.

Model training on twin assets

Not permitted without separate written licence. No LoRAs, no embeddings, no reusable replicas outside scope.

Tool standards

Which AI tools are cleared. Which aren't.

The framework specifies which generative AI platforms can be used with verified twins, in which workflows, and under what account settings.

Approved · Preferred

  • Twinnin Platform — mandatory source for all twin assets
  • Runway Gen-4 References — character consistency, image & video
  • Google Gemini — image generation with approved account settings
  • Adobe Firefly Enterprise — backgrounds & non-likeness assets
  • Seedance 2.0 — AI video with logged references
  • ElevenLabs — voice (if licensed separately)

Conditional · Project-specific

  • OpenAI Image / DALL-E — twin use approved per-project
  • Synthesia / Hour One — only where avatar use is specifically approved
  • Midjourney — moodboards only, no final twin outputs
  • Stable Diffusion / ComfyUI — only in closed approved workflows. No public checkpoints, no public LoRAs

Prohibited

  • Consumer face-swap apps — including all deepfake apps
  • Unapproved public web tools — anything with unclear data handling
  • Social scraping — no Instagram, TikTok, press images, or unofficial sources
  • Unauthorised supplementation — no augmenting an approved twin with non-Twinnin reference images
Get the framework

The standard. In your hands.

Read the full Twinnin Framework v1.0 — operating standards, tool tier list, asset handling, reference pack specs, and QA checklist. Free to use. Free to adopt. Built to be the category baseline.

Request the v1.0 PDF → Adopt the framework