Prime Video now changes the picture to fit the dub, starting with Maxton Hall

What Amazon shipped on 9 September, and what it replaced

5 March 2025. Prime Video launched an AI-aided dubbing pilot on 12 licensed movies and series, in English and Latin American Spanish, including El Cid: La Leyenda, Mi Mamá Lora and Long Lost. The AI produced the audio. Localization professionals reviewed it. Raf Soltanovich, VP of technology at Prime Video and Amazon MGM Studios, said it was "only available on titles that do not have dubbing support". The picture was untouched.

9 September 2026. Amazon introduced lip-sync technology on Maxton Hall, a German series it describes as Prime Video's most-watched International Original. The English lip-sync versions of seasons 1 and 2 went live globally that day, and season 3 is scheduled to carry it when it releases on 9 December 2026. Amazon describes the work as a combination of AI and VFX, credits Prime Video and Amazon MGM Studios teams, and says it is done "under creative oversight to ensure the integrity of their artistic vision is preserved". Soltanovich again gave the quote. Amazon says it plans to extend the technology to more titles and gives no list and no date.

The inversion is the news. The 2025 pilot replaced the human voice and kept the picture. The 2026 feature keeps the human voice and replaces the picture. Amazon has now run AI on both halves of a dub, on different titles, eighteen months apart.

What the announcement does not contain matters as much. It names no technology vendor. It says nothing about the cast's consent to having their faces altered. It does not describe how a viewer can tell the lip-synced version from the original or from a conventional dub, or whether the version is labelled in the audio menu. It gives no language roadmap. We checked the Amazon post and two trade reports on it (Unite.AI and PPC Land) and found none of these details.

Who alters the picture, who only replaces the voice

Four consumer platforms and one API vendor now ship AI in the dubbing chain. Only two of them touch the image, and only one of those does it on long-form professional drama.

Plot the language counts and the shape of the market is plain. Voice replacement is broad and cheap. Picture replacement is narrow and expensive, and so far it has been reserved for a platform's biggest non-English original.

YouTube is the clearest signal of how hard the picture side is. Its 4 February 2026 update opened auto dubbing to all creators in 27 languages, reported more than 6 million daily viewers watching at least 10 minutes of auto-dubbed content in December, and described lip-sync as still being tested. Seven months later we found no YouTube announcement that it had shipped.

Visual dubbing moves the lip-sync constraint from the script to the pixels

In conventional lip-sync dubbing the picture is fixed, so the script bends. The adapter rewrites the translation to fit the length of each line and the visible mouth shapes on screen, and the voice actor records to picture. Meaning is traded for fit, line by line. Visual dubbing reverses the dependency: the translation and performance are chosen for meaning and delivery, and the mouth region is re-rendered to match the audio.

The disclosure question is concrete, not abstract. Article 3(60) of the AI Act defines a deep fake as "AI-generated or manipulated image, audio or video content that resembles existing persons, objects, places, entities or events and would falsely appear to a person to be authentic or truthful". Article 50(4) requires deployers to disclose such content, and then limits the duty for fiction:

"Where the content forms part of an evidently artistic, creative, satirical, fictional or analogous work or programme, the transparency obligations set out in this paragraph are limited to disclosure of the existence of such generated or manipulated content in an appropriate manner that does not hamper the display or enjoyment of the work."

Our reading, not legal advice: a drama whose actors' mouths have been AI-altered is manipulated video resembling existing persons, and the entire purpose of the alteration is that it looks authentic. If that meets the definition, the fiction carve-out still leaves a duty to disclose the existence of the manipulation, in an unobtrusive way. The European Commission published its Code of Practice on marking and labelling AI-generated content on 10 June 2026, including EU icons for labelling. Maxton Hall is available globally, which includes the EU. Amazon's announcement says nothing about any label, and we could not verify how the version appears in the Prime Video app.

Meta already shows what the unobtrusive version looks like: every translated reel carries "Translated with Meta AI", and viewers can switch translation off. That is a disclosure in the menu layer, not on the picture.

What this changes for a dubbing buyer or a localization lead

Script adaptation gets looser, but only where the platform pays for the VFX pass. If visual dubbing is applied, adapters can translate for meaning instead of mouth shape. Today that condition holds for one title and one language. A distributor preparing English dubs for its own catalogue cannot assume the platform will fix the lips, and should keep conventional lip-sync adaptation until a platform commits to a title list in writing.

Talent contracts need a clause for the on-screen cast. Conventional dubbing contracts cover the voice actor. Visual dubbing alters the performance of the original actor, who may be under a different contract in a different country. The Amazon post does not describe the consent arrangement. A producer licensing content for international release should now ask whether the grant covers AI alteration of facial performance for localized versions, and get the answer in the licence rather than in a press release.

QC gains a picture step. Dubbing QC traditionally listens. A lip-synced deliverable also has to be watched, frame by frame around dialogue, for artifacts in the mouth region. That review needs people who can judge both the language and the image. Budget it as a separate pass.

Plan the disclosure before the platform asks. If you deliver AI-altered video into the EU, decide where the disclosure lives: an audio-track label, a title-card line, or metadata the platform surfaces. The fiction carve-out makes it light, not optional, on our reading. Waiting for a platform policy means inheriting whatever they choose.

The input that makes this work is consented face and voice data. A lip-sync model is trained on footage of people speaking. Our judgement: as visual dubbing spreads from one title to a catalogue, the question buyers will face is not only whether the output is disclosed, but whether the training footage was licensed. Amazon has not said what its model was trained on.

Check it yourself

Every factual claim above traces to five pages. Read them yourself and date-stamp what you see, because none of these announcements is versioned.

# 1. the two Amazon announcements, 18 months apart
curl -sL -A "Mozilla/5.0" https://www.aboutamazon.com/news/entertainment/prime-video-lip-sync-technology \
  | python3 -c "import sys,re,html;t=html.unescape(re.sub(r'<[^>]+>',' ',sys.stdin.read()));print('\n'.join(s.strip() for s in re.split(r'(?<=[.])\s',t) if re.search(r'lip|human|oversight|December|season',s,re.I))[:3000])"
curl -sL -A "Mozilla/5.0" https://www.aboutamazon.com/news/entertainment/prime-video-ai-dubbing-english-spanish \
  | grep -o -i "[^.]*12 licensed[^.]*\."

# 2. count disclosure and consent language in the visible text (tags stripped,
#    because raw markup contains aria-label attributes)
curl -sL -A "Mozilla/5.0" https://www.aboutamazon.com/news/entertainment/prime-video-lip-sync-technology \
  | python3 -c "import sys,re,html;h=re.sub(r'(?s)<(script|style).*?</\1>','',sys.stdin.read());t=html.unescape(re.sub(r'<[^>]+>',' ',h));[print(k,len(re.findall(k,t,re.I))) for k in ['label','disclos','consent','watermark','vendor','lip-sync']]"
# 2026-09-15: label 0, disclos 0, consent 0, watermark 0, vendor 0, lip-sync 5

# 3. the legal text
#    https://artificialintelligenceact.eu/article/50/   (paragraph 4)
#    https://artificialintelligenceact.eu/article/3/    (definition 60)

# 4. the comparison rows
#    https://blog.youtube/news-and-events/youtube-auto-dubbing-expressive-speech/
#    https://about.fb.com/news/2025/10/discover-reels-around-world-meta-ai-translation/
#    https://elevenlabs.io/docs/changelog/2026/8/10

In the Prime Video app, open Maxton Hall season 1, open the audio and subtitles menu, and record the exact name of each English track. If one is labelled as lip-synced or AI-altered, that is the disclosure, and it is a detail the announcement left out. We could not check this from a non-subscriber session.

What would prove this wrong

The claim under test is that picture-altering visual dubbing is, in September 2026, a flagship showcase rather than a localization default. It is wrong if, by 30 June 2027, Prime Video offers lip-synced dubs on at least 10 titles or in at least 3 languages, stated in an Amazon announcement or visible in the app. Today it is 1 title and 1 language.

A second prediction, marked as judgement: before Maxton Hall season 3 releases on 9 December 2026, the lip-synced English version will carry a visible or menu-level label identifying it as altered, whether Amazon calls it lip-sync, visual dubbing or AI. The 9 September announcement described none. If season 3 launches with no such label anywhere in the app, this reading was wrong.

FAQ

What is Prime Video's AI lip-sync feature?

A visual dubbing process Amazon announced on 9 September 2026 that alters actors' on-screen mouth movements to match human-performed dubbed dialogue, using AI and VFX. It launched globally on the English dub of Maxton Hall seasons 1 and 2 and will be on season 3 from 9 December 2026. The voices are human; the picture changes. Amazon credits its own Prime Video and Amazon MGM Studios teams and names no outside vendor.

Is Prime Video's lip-sync dubbing AI-generated voice?

No. The lip-sync is matched to human-dubbed audio. That separates it from the AI-aided dubbing pilot of 5 March 2025, on 12 licensed titles in English and Latin American Spanish, where AI produced the audio and localization professionals reviewed it.

Does the EU AI Act require disclosure of AI lip-sync in dubbed films?

Possibly, in a light form. Article 3(60) defines a deep fake as AI-generated or manipulated video resembling existing persons that would falsely appear authentic, and Article 50(4), applicable from 2 August 2026, requires deployers to disclose it. For evidently artistic or fictional works the duty shrinks to disclosing the manipulation in a way that does not hamper enjoyment. Whether lip-synced drama qualifies is a legal judgement. This is our reading, not legal advice.

Which platforms offer AI lip-sync dubbing in 2026?

Prime Video, on one title in one language. Meta, as an optional creator setting on Instagram and Facebook Reels with a "Translated with Meta AI" label. YouTube's February 2026 update said lip-sync was still in testing. ElevenLabs Dubbing v2 translates audio into 90+ languages and leaves video untouched.

Sources

1. Amazon, Prime Video introduces new lip-sync technology on Maxton Hall, 9 September 2026. AI and VFX, human-dubbed audio, English, seasons 1 and 2 globally, season 3 on 9 December, Prime Video and Amazon MGM Studios teams, Raf Soltanovich quote.
2. Amazon, Prime Video begins an AI dubbing pilot program on licensed movies and series, 5 March 2025. 12 titles, English and Latin American Spanish, hybrid process, titles without dubbing support. Date corroborated by Broadband TV News, 6 March 2025.
3. Trade coverage of the 9 September announcement, used to confirm scope and the absence of vendor, consent and disclosure details: Unite.AI and PPC Land.
4. YouTube, auto dubbing update by Chandralekha Motati, 4 February 2026. 27 languages, Expressive Speech in 8, more than 6 million daily viewers of 10+ minutes in December, lip-sync in testing.
5. Meta, Discover Reels from around the world with Meta AI translation, 9 October 2025, updated 14 July 2026. Voice in the creator's tone, optional lip-sync, "Translated with Meta AI" label, viewer opt-out.
6. ElevenLabs, changelog, 10 August 2026. Dubbing v2 via API, more than 90 languages, voice, tone and pacing preserved. Watermark toggle status from BLOMEGA's dubbed-minute cost note.
7. European Union, AI Act Article 50 and Article 3(60), transparency obligations applicable from 2 August 2026. European Commission, Code of Practice on marking and labelling of AI-generated content, 10 June 2026.

Related BLOMEGA guides: Article 50 applies to your dub, and the watermark it asks for dies in your mix · A dubbed minute costs $0.33 to $9.00 · Localization is the new default.