Solo creators using actor or voice-actor voices in AI are in scope. Industry guidelines ask for written consent, credits, and a one-year cap (Japan Actors Union; not a statute)

What happened

On 15 October 2026, the Japan Actors Union (JAU; a cooperative of about 2,600 actors and voice actors) held a press conference at Geinoh Kaden-sha in Shinjuku, Tokyo, and published “Guidelines for the Use of Generative AI on Vocal Audio” Ver. 1.0 (22 articles). A “business operator” covers corporations and individuals, including sole proprietors, freelancers, doujin creators, and individual streamers (Art. 2(3)). “Voice” includes singing, ad-libs, breath, and laughter (Art. 2(2)). Written prior consent, with a stated purpose, is required at the training, generation, and publication/use stages (Arts. 7–9). The numbers and periods are the guideline’s own standards, not statutory duties. This is not legal advice.

Needed background

JAU’s announcement page says Japan has no comprehensive written law that protects a person’s voice as such. Portrait rights and publicity rights (the right not to have the commercial drawing power of a name or likeness used without consent) are discussed, but there is no umbrella statute that directly polices unconsented voice training or synthetic-voice use, in the union’s account. This site previously reported the 30 September 2026 Tokyo District Court decision, which said a human voice can fall within publicity rights in principle, while dismissing the takedown claim and leaving infringement in that case undecided (https://getkominami.com/en/ai-music/2026-10-01-tokyo-voice-publicity-rights). The guideline does not replace that judgment; it is an industry rulebook for the path from training to release. JAU says it will also lobby the government to legislate “voice rights.”

What it means for producers

This matters for commercial producers who put AI vocals or narration on YouTube or in games, and for individuals who build or sell a voice model with a performer. A voice-converted song that sounds like a known voice actor, an indie-game narration made with AI, and a model sold with the performer herself are three different starting points. The guideline is not legally binding. Meeting a tool’s terms of service does not mean you have met this guideline’s written-consent process.

• Training (Arts. 7–9, 8): written prior consent, with purpose, at training, generation, and publication. New digital replicas (models that reproduce a specific performer’s timbre) or new model development: notice at least 7 days before start. In-project generation on an already consented model: at least 48 hours before. Broad buyouts and oral consent are treated as invalid under the guideline.
• Credits and distribution (Arts. 10, 13): credit that the audio is generative AI, the AI vendor/product/model name, and the performer who supplied the source data. The performer may demand a withheld or assumed name. Political or religious propaganda, adult use, and defamatory use are prohibited. Indie-game AI voices still need those three credit items under the guideline.
• License term (Arts. 12, 16, 15, 6): consent lasts at most one year in principle. Long game or animation work may exceed one year within a reasonable scope; perpetual consent is not allowed. Training, use, advertising, and multilingual fees are negotiated separately from ordinary performance fees. Sellers of a performer-backed model must give buyers a credit template (Art. 10(4)). New-work use of a deceased performer’s voice is in principle not allowed (Art. 6).
• Licensed replicas versus others (Art. 20): takedown, model destruction, and “special damages” equal to sales from the start of unconsented use. Art. 20(5)(1): a general user of a replica sold with the performer’s written consent, within personal video or doujin use, does not trigger prior consultation or special damages if Arts. 10 and 13 are kept. Generic TTS that does not imitate a specific performer has a similar exception in Art. 20(5)(2). Not a legal safe harbor.

Do this today

In each project folder, write one line for (1) whether the voice is styled after a real actor or voice actor, (2) whether it is a licensed replica or not, and (3) the three credit items plus the consent expiry (one year in principle).

Sources

Primary sources: JAU announcement page (15 Oct 2026) and the full PDF Ver. 1.0. Supporting: PR TIMES the same day. Secondary: Sankei, Nikkei, Anime News Network (16 Oct 2026). Clickable URLs are at the end.