Introduction

Accordent is a research project from the NUS FinTech Lab tackling one of the fastest-growing harms in generative AI: the use of a real person's face, voice, or likeness without their consent. It provides a shared, verifiable record of likeness consent that AI generation and distribution platforms can check automatically — turning consent from an inconsistent, platform-by-platform afterthought into a portable, auditable standard, and shifting the burden of misuse off the individual and onto the platforms responsible.

Overview

Accordent is a framework that makes likeness consent explicit, verifiable, and machine-checkable at the moment AI content is generated, and independently auditable at the moment it is distributed. It closes the structural gap between AI generation platforms — which manage consent internally and inconsistently — and distribution platforms, which can today only react after harmful content is already public. By binding machine-readable consent terms (ODRL) to a C2PA-carried, blockchain-anchored, non-repudiable record, Accordent shifts accountability for unauthorised likeness use away from the individual depicted and back onto the platforms that create and distribute content.

Be it a rights holder (a public figure, performer, or estate), a regulator, an insurer, or even an AI generation engine or distribution platform, Accordent makes a single question answerable across the whole content pipeline: was this use of a person's likeness authorised?

Current Version

Key Features

Feature 1 — Consent registration & rights definition

A rights holder's representative registers a likeness (with identity verification) and defines permitted uses as a machine-readable ODRL policy — permissions, prohibitions, and time constraints. The default state is no consent until explicitly activated.

Feature 2 — Generation-time consent lookup

Before producing content, a generation engine queries Accordent for the relevant likeness and its permitted scope. Matching is biometric (embedding-based on the reference image and the generated frames), so it holds even when a text prompt is substituted or a reference image is uploaded instead of a name.

Feature 3 — Tamper-evident anchoring & C2PA embedding

The ODRL agreement is stored (encrypted) off-chain on IPFS; its content identifier, signature, and timestamp are anchored on-chain from the engine's own wallet (making the transaction non-repudiable) and embedded into the content's C2PA manifest.

Feature 4 — Independent verification

Any downstream party — a platform, auditor, court, or regulator — can extract the manifest, retrieve the agreement, verify the signature (authenticity), recompute the content identifier (integrity), and read the consent terms without contacting the rights holder or Accordent. Verification lookups are free.

Use Cases

This project can be used for:

• Verifying likeness consent at AI generation time, consistently across platforms
• Independently checking authorisation at distribution time, without a complaint
• Managing talent and estate likeness rights (scope, time-limit, revoke)
• Providing evidence for litigation, takedown appeals, and regulatory audit
• Underwriting media-liability insurance conditioned on verified-consent generation
• Countering celebrity-endorsement scams, non-consensual imagery, political disinformation, and voice cloning

Benefits

For Users (rights holders)

• Free registration and protection — the person depicted never pays
• Explicit, machine-readable control over how their likeness may be used
• A default-deny model that makes "never authorised" provable, lifting the burden of detection and complaint off the individual
• Support for revocation and time-limited grants

For Organizations (engines, platforms, regulators)

• Converts unbounded, unprovable liability into a bounded, provable record
• Non-repudiable proof supporting safe-harbour defences (US NO FAKES Act) and machine-readable marking duties (EU AI Act Article 50, China's labelling rules)
• Cross-platform interoperability instead of siloed internal consent systems
• Free, independent verification — a compliance line item, not a cost centre

Roadmap

Project Team

Developed and maintained by the NUS FinTech Lab Accordent project team and contributors. We welcome feedback and suggestions to help improve the project.

License

This project is distributed under the applicable project license.