Japan's Portrait Rights (肖像権) in the Age of Generative AI

Two Separate Rights

Japan does not have a single statute covering portrait rights. The protection comes from two overlapping civil law concepts developed through court decisions.

肖像権 (portrait rights) is the right to control how one's appearance is used and published. It applies to ordinary individuals and public figures alike. Courts have held that using someone's image without consent can constitute an infringement, but the remedy is civil, not criminal. Victims can sue for damages or seek an injunction.

パブリシティ権 (publicity rights) covers the commercial value of a name and likeness. It applies specifically to public figures whose image carries economic value. The Supreme Court's 2012 Pink Lady ruling established that unauthorized commercial use of a celebrity's likeness constitutes an unlawful act under civil law.²

Neither right has ever been tested against a photorealistic image of a person who was never photographed in the first place.

Where Criminal Law Applies — and Where It Stops

Several criminal provisions can reach deepfake distribution. The Act on Prevention of Damage Related to Private Sexual Images (the revenge porn law), amended in 2023 to extend to AI-generated intimate imagery, makes distribution a criminal offense punishable by up to three years imprisonment or a fine of ¥3 million.³Defamation (名誉毀損) and obstruction of business (業務妨害) can apply when deepfake content is used to damage someone's reputation or livelihood.

The critical gap is creation. Making a deepfake, including a sexually explicit one of a real, identifiable person, is not by itself a criminal offense under current Japanese law. South Korea amended its Sexual Violence Punishment Act in 2024 to criminalize production regardless of any intent to distribute. Japan has not made that change.⁴

Why AI Makes the Old Framework Harder to Apply

Portrait rights developed around a simple fact pattern: someone took a photograph, the subject did not consent, the photograph was published. AI-generated imagery removes the photograph entirely.

If a model generates a photorealistic face that resembles a specific person, no original image was used, no camera was pointed at anyone, and nothing was "copied" in the traditional sense. Whether a purely AI-generated likeness falls within 肖像権 at all is legally unsettled, and courts have not yet ruled on the question.

The Unfair Competition Prevention Act (不正競争防止法) offers a partial answer. A 2025 METI analysis concluded that AI-generated portraits used commercially for a well-known person could be challenged under provisions on known display confusion and misrecognition acts, but these require the subject to be sufficiently famous and address commercial use, not intimate imagery.⁵

The 2025 AI Promotion Law and the 2026 Push for Penalties

Japan's Act on Promotion of Research and Development, and Utilization of AI-related Technology, enacted in May 2025, established a framework for government oversight. It allows investigation and guidance when AI causes rights violations, but it contains no criminal penalties for non-compliance. An AI operator can ignore a government inquiry without legal consequence.⁶

In April 2026, the LDP's AI and Web3 panel proposed adding penalties to this law: criminal sanctions for operators who repeatedly ignore compliance requests, and new disclosure obligations requiring AI providers, including foreign companies offering services in Japan, to explain their training data practices and content safeguards.⁷ As of May 2026, those proposals are under review and have not been enacted.

What This Means in Practice

For private individuals targeted by AI-generated intimate imagery, the path to relief is slow. The fastest route is platform removal: most major platforms have policies against non-consensual intimate deepfakes and respond to formal takedown requests within 24 to 72 hours. Japan's revenge porn law also supports rapid injunctions when the distributor can be identified.

Civil suits under 肖像権 are possible but expensive and time-consuming. Criminal complaints work for distribution cases when the distributor is identifiable, which is harder when content originates on anonymous platforms or overseas services.

The most reliable protection today is upstream. Limiting publicly accessible photos, keeping social media accounts private, and understanding that any photo reachable online can serve as a reference for AI generation are practical steps available to anyone, regardless of how the law develops.