Prime Video lip-sync technology went live on 9 September 2026, and the announcement that introduced it runs to 357 words. Those 357 words contain eight numbers. Five of them are season numbers, two are the same December release date written twice, and one is the year a novel was published. Not one of the eight measures the technology the page exists to announce. Amazon says the feature “synchronizes actors’ lip movements with human-dubbed audio” using “AI and VFX technologies” that are “responsibly applied,” and then the page ends.
The feature itself is real and shipped. It is live on Seasons 1 and 2 of Maxton Hall — The World Between Us, a German-language series that Amazon calls its most-watched International Original ever, and it will be there when the third and final season premieres on 9 December 2026. What is unusual is not the technology, which is a well-studied computer vision problem with a published literature. What is unusual is how little the announcement is willing to say about it.
This piece reads the Prime Video lip-sync announcement against three other documents Amazon published itself: the March 2025 AI dubbing announcement it replaces, the entertainment-desk page about the same television show, and the four papers Amazon’s own research organisation has published on exactly this problem. The gaps between them are large, consistent, and countable.
Table of contents
- What the Prime Video Lip-Sync Announcement Actually Says
- The 357 Words: What the Prime Video Lip-Sync Page Counted
- Prime Video Lip-Sync Versus the 2025 AI Dubbing Pilot
- The Show Page Counts What the Prime Video Lip-Sync Page Does Not
- Amazon’s Own Researchers Already Published a Score for This
- What Prime Video Lip-Sync Does to a Filmed Performance
- The Consent Question the Prime Video Lip-Sync Page Never Raises
- What Prime Video Lip-Sync Means for the Dubbing Industry
- How to Read an AI Announcement Like the Prime Video Lip-Sync Page
- What to Watch Next on Prime Video Lip-Sync
- Frequently Asked Questions About Prime Video Lip-Sync
- References and Further Reading
What the Prime Video Lip-Sync Announcement Actually Says
The document is short by design. It carries a “1 min read” label, a byline reading “Written by Amazon Staff,” three key takeaways, six paragraphs of body copy and one executive quote. Stripped of navigation and unrelated promo cards, it comes to 357 words.
The mechanism, in Amazon’s own words
Amazon’s description of what happens is a single sentence: “The technology synchronizes actors’ lip movements with human-dubbed audio, creating a seamless viewing experience that allows audiences to remain fully immersed in the story.” A second sentence adds that “AI and VFX technologies power the process behind the scenes, responsibly applied to deliver a better customer experience.”
That is the whole technical account. Between them, those two sentences establish that something is done to an actor’s mouth, that two categories of tool are involved, and that Amazon considers the result responsible.
What “human-dubbed” is doing in that sentence
The most precise word in the Prime Video lip-sync announcement is “human-dubbed.” It appears once, and it is attached to the audio. The voice performance stays human; the announcement is explicit about that. The visual half of the pipeline gets “AI and VFX technologies” and no adjective at all.
The asymmetry is deliberate and worth naming. Amazon flags the part of the process a person still performs and leaves unlabelled the part where a machine alters a filmed performance. A reader who skims will come away thinking the dubbing is human. Half of it is.
The scope, which the page never states plainly
The Prime Video lip-sync dubs are available “globally in English” on two seasons of one show. The word “language” does not appear in the document. Neither does “German,” the language the series was shot in. Neither does “episode.” The reader is never told that this is a one-language, one-title deployment; that fact has to be assembled from a sentence about availability.
The 357 Words: What the Prime Video Lip-Sync Page Counted
Counting the announcement is quick, because there is very little to count. Eight numeric tokens appear in the body, and they resolve to three subjects: season numbering, which accounts for five of them; the 9 December release date, written out twice; and 2018, given as the publication year of the Mona Kasten novel the season adapts.
Zero of the eight numbers describe the technology
There is no accuracy figure. No percentage of frames altered, no viewer preference result, no A/B test, no error rate, no processing time, no count of episodes treated, no count of shots left untouched. The announcement contains no % symbol and never uses the word “percent.”
This is not a case of a company withholding a bad number. It is a case of a company publishing a feature page with no measurement of any kind on it, then describing that feature as an improvement seven times over.
The words that do appear, and how often
“Customer” outnumbers “actor” seven to two in a document about redrawing actors’ mouths. “Experience” is the most frequent noun in the piece, at eight uses in 357 words — one every 45 words.
The terms it never uses at all
| Term | Uses | Why a reader would want it |
|---|---|---|
| face / facial | 0 | Names what is being modified on screen |
| model | 0 | Identifies the system doing the work |
| generative / synthetic | 0 | Distinguishes generation from retouching |
| train / training corpus | 0 | Says what the system learned from |
| consent | 0 | Covers the performers whose faces change |
| accuracy / percent / % | 0 | Would let the claim be checked |
| language | 0 | States the actual scope of the rollout |
| German | 0 | Names the source language being replaced |
| episode | 0 | Quantifies how much content was processed |
| opt out / toggle / choice | 0 | Tells a viewer whether it can be switched off |
| label / watermark / disclose | 0 | Tells a viewer when they are watching it |
| vendor / partner | 0 | Says whose technology this is |
The entire ethical vocabulary of the document is four words — “responsibly,” “oversight,” “artistic” and “integrity” — one use each, 1.12% of the text.
Prime Video Lip-Sync Versus the 2025 AI Dubbing Pilot
Amazon has announced an AI localisation feature before. On 5 March 2025 it published a page introducing “AI-aided dubbing” on licensed titles. That announcement is 308 words, also labelled “1 min read,” and also quotes Raf Soltanovich, VP of technology at Prime Video and Amazon MGM Studios. The two documents are the same shape, 553 days apart.
The earlier, less invasive feature disclosed more
| Measure | AI dubbing pilot, 5 Mar 2025 | Lip-sync, 9 Sep 2026 | Change |
|---|---|---|---|
| Words in the announcement | 308 | 357 | +15.9% |
| Stated read time | 1 min | 1 min | unchanged |
| Titles counted | 12 | 0 (one named) | −12 |
| Languages named | 2 | 1 | −50% |
| Audience figure given | 200 million | none | −1 |
| Process words used | 7 | 0 | −7 |
| Experience words used | 2 | 21 | +19 |
| Human role named | localization professionals | none | −1 |
| Uses of “pilot” | 2 | 0 | −2 |
| What AI touches | the audio track | the filmed performance | — |
The 2025 page called its own release a pilot, twice. It named the human job title in the loop. It said the AI option applied only to titles that had no dubbing at all, which is a genuine limit that made the claim falsifiable. The Prime Video lip-sync page does none of those things while shipping a materially more invasive change.
The vocabulary swap, measured
Six process words appear across the 2025 announcement — “quality,” “control,” “professional,” “hybrid,” “expert” and “pilot” — for seven total uses. In the 2026 Prime Video lip-sync announcement, every one of them is used zero times. Five experience words — “experience,” “enhance,” “seamless,” “immersive” and “global” — go the other way, from two uses to twenty-one.
Experience language is 9.1 times denser in the newer document. Process language is gone entirely. The company did not get quieter overall — the lip-sync page is 15.9% longer. It got quieter about method and louder about feeling.
The Show Page Counts What the Prime Video Lip-Sync Page Does Not
Amazon published a second document about Maxton Hall the same week, on the same site, in the same news section. It is the Season 3 preview, updated 3 September 2026 and carrying a named human byline rather than “Amazon Staff.” It is labelled “3 min read.”
Twenty-three named people, and a full price list
The show page names 23 people: thirteen cast members, a director, two head writers, three additional writers, two more executive producers, a producer and the novelist. It states that the final season has six episodes, that it premieres in more than 240 countries and regions, that Prime membership costs $14.99 monthly or $139 annually in the United States, that a 50% discount exists for some customers, that a 30-day free trial is available, and that Prime Video carries over 900 free ad-supported channels.
Two named people on the technology page
The Prime Video lip-sync announcement names two: Raf Soltanovich, who is quoted, and Mona Kasten, who wrote the novel. Neither of the two lead actors whose filmed performances the system modifies is named anywhere on that page.
The two pages disagree about the show itself
There is a smaller tell in the same pair of documents. The Prime Video lip-sync announcement says “The upcoming Season 3 is based on the 2018 novel Save Me by Mona Kasten.” The Season 3 preview, published six days earlier, says “Season 3 is based on the third book in the series, Save Us.”
Both pages are Amazon’s, both are about the same six episodes, and they name different books. It is a trivial error in isolation. It matters here because it is the only factual claim on the technology page precise enough to be checked against another Amazon page — and it does not survive the check.
The comparison that matters
| Document | Date | Read time | Named people | Countable claims |
|---|---|---|---|---|
| Maxton Hall S3 preview | 3 Sep 2026 | 3 min | 23 | Episodes, countries, two prices, discount, trial, channel count |
| Prime Video lip-sync page | 9 Sep 2026 | 1 min | 2 | None about the technology |
| The Verge report | 9 Sep 2026 | — | 1 byline | Adds the English-only limit and the source language |
A romance drama gets a cast list, a price and an episode count. A system that alters filmed human performances gets an adjective.
The press filled one gap in 169 words
The Verge’s report on the launch runs to 169 words of body copy — 47% the length of Amazon’s own announcement — and still adds two facts the announcement omits. It states that the feature “is only available with the English dub,” and it identifies Maxton Hall as a German series. Neither sentence required access; both required only a willingness to state the scope plainly.
Amazon's Own Researchers Already Published a Score for This
The most striking gap is internal. Search amazon.science for “lip sync dubbing” and it returns four papers. Amazon’s research organisation has been publishing on this exact problem for years, with the numbers a product page would need.
The four papers
| Paper | Venue | Year | What it quantifies |
|---|---|---|---|
| Perceptual synchronization scoring using phoneme-viseme agreement (PhoVis) | WACV workshop | 2023 | A grounded score for how well lips match dubbed audio |
| SIDGAN: high-resolution dubbed video generation | ICCV | 2023 | Mouth movement synchronised to driving audio while preserving identity |
| Dubbed audio sync detection using compressive sensing | WACV workshop | 2023 | Audio-video desynchronisation as a measurable quality defect |
| Dubbing in practice: a large scale study of human localization | TACL | 2022 | 319.57 hours of video across 54 professionally produced titles |
A metric exists and went unused
The PhoVis paper’s stated motivation is that comparisons between lip-synchronisation methods are “weakly substantiated due to the lack of a generalized and visually-grounded evaluation method.” Amazon researchers built that method and published it. Three years later, Amazon shipped a Prime Video lip-sync feature to a global audience and published no score from it, or from anything else.
The words “PhoVis,” “viseme,” “synchronization score” and “evaluation” appear zero times in the launch announcement. The newest of the four papers is dated 2023, roughly three years before the launch — so the research is not fresh work still under wraps.
The corpus figure the product page could have borrowed
The 2022 paper describes a corpus of 319.57 hours of video from 54 professionally produced titles — an average of 5.92 hours per title. That is what disclosure looks like when the audience is other researchers. The Prime Video lip-sync announcement had the same publisher and a much larger audience, and gave a number for none of it.
What Prime Video Lip-Sync Does to a Filmed Performance
Strip the marketing and the mechanism is not mysterious. A performer was filmed speaking German. A voice actor recorded English dialogue. The system then changes the mouth region of the original footage so it appears to form the English words.
This is a change to the image, not the sound
Traditional dubbing leaves the picture alone and accepts the mismatch. The Prime Video lip-sync approach accepts the audio and alters the picture instead. That is a different category of edit, and it is why the absence of the word “face” from the announcement is more than a stylistic quibble.
Where the human oversight sits is unstated
Amazon says the work is “all done under creative oversight to ensure the integrity of their artistic vision is preserved.” The sentence has no subject. It does not say who exercises the oversight, at what stage, on how many shots, or with what authority to reject a result. The 2025 page, by contrast, named localization professionals and described a hybrid workflow with quality control.
Nobody says whether a viewer can turn it off
Across the announcement, the show page and the press coverage, no document states whether Prime Video lip-sync is a selectable audio-and-video track or an unavoidable property of the English dub. “Opt,” “toggle,” “setting” and “choice” appear zero times in Amazon’s page. For a feature that changes what an actor’s face does, that is the single most useful sentence nobody wrote.
The Consent Question the Prime Video Lip-Sync Page Never Raises
Two lead actors, Harriet Herbig-Matten and Damian Hardung, are named on Amazon’s show page. Their filmed performances are what the system modifies. Neither is named on the technology page, and the word “consent” appears nowhere in it.
What the announcement does say about creators
The Soltanovich quote addresses creators, not performers: “for creators, we’re making it easier for their stories to reach an even broader and global audience.” Stories and audience reach are the frame. The person whose mouth is redrawn is inside the phrase “actors’ lip movements” and nowhere else.
Why the distinction is not pedantic
Performer likeness has become the central bargaining issue in screen-industry labour negotiations worldwide, and the reason is precisely this class of tool. A company can behave impeccably here and still leave the question open by not addressing it. Amazon may well have contractual permission from every performer involved; the announcement gives a reader no way to know.
What a disclosing version would have said
One sentence would close it: which performers were involved, whether they reviewed the output, and whether the arrangement was negotiated. The Prime Video lip-sync page has room — it is 357 words in a format Amazon routinely runs to three times that length, as the show page proves.
What Prime Video Lip-Sync Means for the Dubbing Industry
Localisation is a large, unglamorous business built on voice talent, adaptation writers, dubbing directors and quality control. The 2025 announcement described that ecosystem. The 2026 one does not mention it.
The audio jobs are explicitly preserved
“Human-dubbed” is a real commitment, and it should be read as one. Amazon is not claiming to have replaced voice actors on Maxton Hall. The English dialogue was performed by people.
The visual work is the new line item
What the Prime Video lip-sync pipeline adds is a post-production step that did not exist in traditional dubbing at all. Amazon attributes it to “AI and VFX technologies,” which is two industries, and names no vendor, no internal team and no tool. Anyone trying to understand where this work will be done, and by whom, gets nothing.
The expansion is stated but unbounded
Amazon says it “plans to expand the technology to additional titles.” No count, no timeline, no criteria for which titles qualify. The 2025 pilot at least published a rule — AI dubbing applied only to titles with no existing dubbing support. The Prime Video lip-sync announcement publishes no equivalent boundary, which means the deployment could be narrow or total and the page reads identically either way.
How to Read an AI Announcement Like the Prime Video Lip-Sync Page
The pattern here is not unique to Amazon, and it is easy to test on any launch page in a few minutes. A useful AI strategy starts with being able to tell a specification from an adjective.
Four questions that separate the two
| Question | What a specification looks like | What this announcement gives |
|---|---|---|
| What changed? | Named artefact, named modification | “lip movements,” never “face” |
| How well does it work? | A score, against a stated method | No number of any kind |
| Who checked it? | A named role at a named stage | “creative oversight,” no subject |
| Can it be refused? | A setting, a label, or a stated default | Not addressed |
Count the adjectives against the nouns
The quickest tell is the ratio this article opened with. Eight numbers, none about the product. Twenty-one uses of experience vocabulary, zero uses of process vocabulary. A page that has genuinely measured something almost always tells you what it measured, because that is the cheapest way to be believed.
The comparison document is usually on the same site
Every gap in the Prime Video lip-sync announcement was visible by reading another page the same company published. The 2025 pilot page supplied the missing process language; the show page supplied the missing willingness to count; the research site supplied the missing metric. This is the most reliable method available, and it costs three browser tabs. Organisations that treat artificial intelligence and machine learning claims as evidence to be checked, rather than positioning to be absorbed, tend to run this comparison as routine. Our AI models and tools hub tracks launches on the same basis.
What to Watch Next on Prime Video Lip-Sync
Three dated things are now on the record, and each is checkable.
9 December 2026
Season 3 premieres with the lip-sync dub included. That is Amazon’s stated commitment, and the season has six episodes according to the show page. Whether the announcement accompanying it carries a number is the thing to look for.
The “additional titles” expansion
Amazon has committed to expanding the technology without saying to what. The first sign will be a second title receiving it, and the useful question then is whether the page announcing it names a second language.
A published score, or none
Amazon’s researchers built PhoVis to make claims like this comparable. If a future Prime Video lip-sync announcement quotes any evaluation figure at all, the pattern in this article breaks. If the next one is another 357 words of experience vocabulary, it holds. Teams building on top of any vendor’s data management and analytics claims face the same test in a less visible form.
Frequently Asked Questions About Prime Video Lip-Sync
What is Prime Video lip-sync?
It is a feature Amazon launched on 9 September 2026 that alters actors’ on-screen mouth movements so they match human-recorded dubbed dialogue. Amazon describes it as visual dubbing powered by “AI and VFX technologies.”
Which shows have it?
Maxton Hall — The World Between Us, Seasons 1 and 2, in English, globally. Amazon says it will also be on Season 3 when that premieres on 9 December 2026, and that it plans to expand to unnamed additional titles.
Are the voices AI-generated?
No. Amazon specifies “human-dubbed audio.” The AI and VFX work is applied to the picture, not the voice track. That is the opposite of Amazon’s March 2025 AI dubbing pilot, where the AI worked on the audio.
Can you turn Prime Video lip-sync off?
No published document says. Amazon’s announcement does not use the words “opt,” “toggle,” “setting” or “choice,” and the press coverage does not address it either.
How accurate is it?
Amazon has not said. The announcement contains no accuracy figure, no percentage and no evaluation result, despite Amazon Research having published a purpose-built synchronisation metric, PhoVis, in 2023.
Is this the same as a deepfake?
The underlying technique family overlaps, but the framing differs: this is applied to licensed content, with the rights holder’s involvement, to match a translation. Amazon does not use the word “deepfake,” “synthetic” or “generative” anywhere in the announcement.
References and Further Reading
Prime Video introduces new lip-sync technology on ‘Maxton Hall’ — About Amazon
Prime Video begins an AI dubbing pilot program on licensed movies and series — About Amazon
Watch the teaser trailer for ‘Maxton Hall’ Season 3 — About Amazon
Amazon Prime Video’s new AI tech matches lips to dubbed audio — The Verge
Maxton Hall Debuts Prime Video Feature Syncing Mouths to Dubbed Dialogue — Unite.AI
Perceptual synchronization scoring of dubbed content using phoneme-viseme agreement — Amazon Science
SIDGAN: High-resolution dubbed video generation via shift-invariant learning — Amazon Science
Dubbing in practice: A large scale study of human localization — Amazon Science
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.