No filming:An AI avatar or clone yourself
    Versely
    Industry

    NY law skips audio and translation

    NY §396-b skips audio-only ads and translation-only AI. A talking-head video of a fake person is not exempt. TTS and video split.

    Versely Team••6 min read

    New York's synthetic-performer law skips audio-only ads and AI used solely for language translation of a human performer. A talking-head video of a fake person is not exempt. That is the practical split for TTS versus a generated face. N.Y. GBL §396-b has been in force since 9 June 2026. Penalties $1,000 first / $5,000 subsequent. This page is the exemption. The comparison to the EU rule is NY synthetic performer vs Article 50.

    This is not legal advice. Recaps from AP (9 June 2026) and firm notes (McDermott / Crowell / Manatt, June 2026) are the public layer we are using. Counsel reads the actual buy.

    What the statute carved out

    §396-b is an advertising rule about a synthetic performer: a digitally created asset that appears as a human performance by someone not recognisable as an identifiable natural person. Advertisers with actual knowledge must conspicuously disclose in ads that reach New York consumers.

    The same recaps name the exemptions that matter to production:

    • Audio-only advertisements. No picture of a fake human. Voice-only.
    • AI used solely for language translation of a human performer.
    • Promotional material for expressive works (film, TV, games) when the synthetic use matches the work.
    • Publishers and platforms hosting the ad, as hosts.

    This page is the first two. They are how a team tries to keep a cheap voice pipeline without a conspicuous "this person is not real" on every unit — and how that team overreaches the moment a face appears.

    Audio-only means audio-only. A still of a generated person under a voice-over is not audio-only. A 6-second video with a talking mouth is not audio-only. A waveform graphic with no human impression might be a different object; a waveform next to a fake head is not. If the unit would play as a video in a feed, treat it as video.

    Translation-only means that is the sole AI use, and the performer is a human. Dubbing a real creator's ad into Spanish, when the face on screen is that creator, sits in the carve-out the recaps describe. Generating a new face that speaks Spanish is not translation of a human performer. It is a synthetic performer plus a language. Lipsync on a generated stranger is not translation-only.

    TTS versus talking-head ads

    TTS on a real plate. You filmed a product, a store, a real employee who is not the voice. A TTS bed reads the script. NY's synthetic-performer definition is about a human performance that appears to be a person who is not identifiable. A voice with no fake human on screen is the audio-only shape if the whole ad is audio, or a voice-over on real non-person footage if there is no synthetic performer in the picture. Do not invent a face "to make it UGC." The face is what pulls you into the disclose duty.

    TTS as a talking head. You generate a person, then drive the mouth with TTS or a lipsync row. That is a synthetic performer in a video ad. Not exempt. Conspicuous disclose. The voice being synthetic does not cancel the face being synthetic.

    Translation of a real talking head. Real creator, real performance, AI used only to change language. Recaps treat that as exempt. Keep the use sole. If you also restyle the face, swap the room, or generate extra takes of a person who was not there, you have left "solely translation."

    Translation of a fake talking head. There was never a human performer. Translation is not the exemption you are in. Disclose the synthetic performer (and still check Article 50 if EU viewers would take the person as real).

    Native audio on a video model is not "audio-only." Native audio means the clip shipped with sound. If that clip is a generated person, it is video of a synthetic performer.

    A UGC video generator is almost always the talking-head side of this split. Veo 3.1 with a photoreal person is the same. TTS rows are the audio side only when the picture does not invent a human.

    What still stacks

    NY's exemption is not a platform holiday. YouTube, TikTok, and Meta can still want an AI content label on realistic synthetic audio. Fastlane's 24 August 2026 summary of the shared photoreal test includes synthetic voice that could be mistaken for real. Cite that only as the shared hinge. A NY audio-only exemption does not tick YouTube's checkbox off.

    NY's exemption is not Article 50. Audio can be a deepfake under the Act if it resembles a real person and would appear authentic. Deployers still label at first exposure for EU audiences. Translation into a realistic cloned mouth can still be a deepfake-shaped file in the EU even when NY calls the use translation.

    FTC endorsement rules still apply to a voice that praises a product in a paid unit. #ad is not the NY synthetic-performer line, and it is not an AI label. FTC endorsement plus an AI label is the stack.

    Digital twins of identifiable people are outside §396-b's synthetic-performer definition. They are not "saved" by the audio exemption either; they are a different legal pile. NY law is not a digital-twin statute.

    A production default

    If the ad can reach NY consumers:

    1. No fake human in picture, audio-only unit: NY synthetic-performer disclose may not attach. Still check platform audio rules and Article 50.
    2. Real human, translation-only: NY may not attach. Do not add a generated extra and keep calling it translation.
    3. Generated face, any language: conspicuous disclose in the ad. $1k / $5k is the statute. Burn the line. Do not hide it in a caption pile.
    4. Borderline still + voice: if a viewer would take the still as a human performer in the ad, treat it as in.

    If you need the talking-head and the disclose line on one timeline, sign up and burn the conspicuous mark on the video units. Keep TTS-only versions in an audio package that is actually audio. Do not "just add a face" to a radio spot and expect the exemption to follow.

    The carve-out is narrow on purpose. Video of a fake person is the thing the law wanted on the label.

    FAQ

    Is a podcast ad with no video exempt?

    If the advertisement is audio-only, the recaps treat it as exempt from §396-b. The moment you attach a generated host image or a talking-head cutdown for YouTube, that cutdown is a different ad. Disclose on the video unit.

    Does auto-dub of my real face count as translation-only?

    If the only AI is language translation of your human performance, that is the exemption the recaps describe. If the dub is applied to a generated lookalike, you are not in that exemption.

    Can I skip YouTube's AI checkbox because NY exempted the audio?

    No. NY §396-b is not YouTube Help. Realistic synthetic audio can still be in-scope for platform disclosure. Run each test on the file you are uploading.