Netflix Recreated Gene Wilder's Voice — With Permission. Permission Is the Hard Part.

On September 23, a Netflix competition series called Wonka's The Golden Ticket premieres with a voice nobody recorded. Gene Wilder died in 2016. Netflix and Eureka Productions worked with the AI voice company ElevenLabs to re-create the voice he gave Willy Wonka in 1971, and it narrates a show in which 12 golden ticket winners and their chosen partners compete through a version of the chocolate factory — with Rusty Goffe, who played an Oompa Loompa in the original film, appearing alongside them. Almost every AI voice story that reaches the news is a story about someone who never agreed. This one is the opposite: Wilder's estate granted permission. "We are delighted that Wonka's The Golden Ticket celebrates the warmth and imagination that he brought to the role," Karen B. Wilder said on the estate's behalf. Which makes it the more useful story to study — because the interesting part isn't the cloning. It's the paperwork.

Most coverage stops at four words: the estate approved it. But "approved" is carrying an enormous amount of weight in that sentence. Approved for this one series, or for any Wonka project that follows? For how long? Approved to generate entirely new lines, or only to reassemble ones he actually spoke? Approved for the show itself, or for the trailers, the social clips, the merchandising, the theme park tie-in? And — the question people forget to ask about anyone who has died — who exactly held the right to say yes at all? These are not pedantic. They are, increasingly, the precise questions that the law and the union contracts now require to be answered in writing, before anything is generated.

California answered the last one first. AB 1836, which took effect on January 1, 2025, amended Civil Code section 3344.1 to prohibit producing or distributing a digital replica of a deceased personality's voice or likeness in an expressive audiovisual work or sound recording without prior consent from the estate. It defines a digital replica as a highly realistic representation "readily identifiable as the voice or visual likeness" of the person, in which they either did not actually perform or the fundamental character of their performance has been materially altered. Get it wrong and the floor for damages is $10,000, or actual damages, whichever is greater. The statute even spells out who inherits the right to consent, in order: a surviving spouse, then children and grandchildren, then parents. There are carve-outs — news, comment, criticism, satire, parody, scholarship, documentary and historical works, fleeting uses — but a reality show narrated by the dead man's voice is not one of them.

The union contracts answer the rest — and they are far more specific than most buyers realize. Under SAG-AFTRA's 2025 Interactive Media Agreement, consent for a digital replica must be "clear and conspicuous," set out in a separate document or a specifically signed rider — not buried as a blanket clause in an initial employment contract. The employer has to supply a "reasonably specific description" of the intended use, including whether the replica will be used for real-time generation, and whether the performer will be asked to handle sensitive material such as profanity or slurs. Crucially, studios cannot grab consent up front for games that don't exist yet; they have to come back and negotiate again. Vocal replica compensation is metered on generated lines, with one "line" defined as roughly ten words, and the employer owes a usage report within 90 days of release. The Commercials Contract runs on the same logic: consent through a Digital Replica Rider or a separate signed writing, based on a reasonably specific description of the intended use, with distinct forms for creating a new replica, using an existing one, and authorizing commercial use.

Read those two frameworks side by side and a pattern emerges. Consent is being redefined from an event into a document with fields. The question is no longer "did they say yes" — it is yes to what, for how long, in which medium, generated by what method, and on whose authority. That is why a tickbox can never satisfy it: a tickbox records that a click happened, not what was agreed. And it is why the Wilder project is worth studying rather than just arguing about. Whatever anyone thinks of resurrecting a performance, a major studio did the one thing the emerging rulebook actually asks for — it went and got a documented yes from the people entitled to give one, before generating a single line.

Here is where it lands on you. Netflix has a legal department; your campaign probably doesn't. But the exposure travels the same direction — downstream. The synthetic voice in your ad sits on your brand, in your channels, under your name, and "the vendor said it was fine" is not a document. Notice too that the Wilder case had to clear two independent gates: state law required the estate's consent, and the applicable union agreement required its own. A supplier who can satisfy neither is not offering you a cheaper voice; they are offering you an unpriced liability with a very specific damages floor attached. The practical test is unglamorous and completely decisive: ask your voice supplier to produce the consent. Not describe it — produce it. Who is the person, what did they sign, what does it cover, and does it cover what you are about to do?

That question is one we built Onyx Studios to be able to answer. Every AI voice in our library traces back to a named professional who walked into a session, recorded under an explicit Voice ID authorization, and signed a contract that stays on file — scope, medium and term written down, per project, the way the new rulebook assumes. We don't scrape, we don't quietly promote a "test" recording into a product, and we don't ship a voice whose selling point is that it resembles someone famous. The actor is paid and credited; the document exists; we can show you both. And because we came up as a Taiwan voice studio — since 2008, with more than 1,500 professional actors — a native speaker still verifies every delivery before it reaches you. That is what "AI-Generated. Human-Perfected." means in practice: provenance you can produce, and a read you can ship.

The Wonka show will divide people, and reasonably so — an estate can consent on a performer's behalf, but it cannot ask him what he would have wanted. That argument is worth having. What isn't in dispute is the smaller, more portable lesson sitting underneath it: the projects that survive contact with the new rules are the ones where somebody can produce the document. If you are buying voice, make that your first question rather than your last. Tell us what you need voiced — and ask us to show you the paperwork behind it. We'd consider that a fair test.