I only have time to compare one 13-minute segment, so I’m trying to decide whether Prime Video AI dubbing would feel more natural if it also altered lip movements. I tried switching between the original audio and the dubbed track during the same scene, but the voice timing still pulled my attention toward the actors’ mouths.
What should I try next to judge this fairly? Is “lip-sync correction” the right term, and would adjusted mouth shapes improve immersion or become more distracting than the mismatch itself?
The missing detail is how often the segment shows clear, front-facing speech. Lip alteration is easiest to judge in close-ups and hardest to notice in wide shots, profiles, facial hair, or fast cuts. A scene with little visible articulation could make either version seem better than it really is.
Don’t keep switching audio during the same viewing. That trains you to inspect mouths instead of watching the movie. Watch the full segment once with ordinary dubbing, take a short break, then watch it with altered lips. If possible, reverse the order on a second day. Judge whether you followed the scene naturally, not whether every consonant looked exact.
“Lip-sync correction” is understandable, but “visual dubbing” or “AI lip-sync adjustment” may be more accurate. “Correction” implies there is one correct mouth movement, even though translated dialogue often changes rhythm, syllable count, and emphasis.
Adjusted mouth shapes could improve immersion when the edit is subtle. The risk is that small facial artifacts are much more distracting than loose synchronization. Watch the teeth, jawline, cheeks, and transitions between speaking and silence. If you notice those before the dialogue, the visual adjustment has probably made things worse. I’d favor a believable overall rhythm over perfect syllable matching.
Lip matching won’t fix a dub if the translated performance sounds stiff. I’d judge the voices first, because natural timing and emotion can carry imperfect mouth sync, while altered lips cannot rescue flat dialogue. For a limited comparison, pick the version you’d willingly keep listening to after you stop studying the screen.
Even a very good lip adjustment probably won’t make a dub feel completely native. At best, it removes the obvious timing mismatch. The danger is that it changes more than timing, because the mouth is part of the actor’s performance. A tightened smile, a half-finished word, or someone speaking through clenched teeth can carry emotion that a generated replacement might smooth away.
I agree with @ghostpanda71 that weak voice acting is the bigger problem, but I wouldn’t judge the voices separately. The real question is whether the sound and face still express the same mood together. A line can sound angry while the altered mouth makes the actor look oddly calm, and that disconnect may be worse than ordinary dubbing.
For your Prime Video comparison, I’d choose a segment with interruptions, jokes, or an emotional shift rather than a basic conversation. Those moments reveal whether the altered lips preserve performance or merely line up syllables. Pay attention to whether sarcasm still reads, whether pauses feel intentional, and whether reactions flow naturally into the next shot.
My preference would be altered lips as an optional setting, not the default. I can accept imperfect sync because I already know I’m watching a dub. Changing an actor’s facial performance crosses a different line, and the improvement would need to be nearly invisible before I’d leave it turned on.
Turn the subtitles off first.
If captions are running, your eyes keep dropping away from the actors, which can hide both bad synchronization and bad facial edits. The exception is if you normally watch everything with captions. In that case, leave them on because you should test the feature under your real viewing conditions. Use the same screen, viewing distance, volume, and lighting you would use for a full movie.
For the comparison, play each version straight through and mark only the moments that interrupt you without actively searching for flaws. Keep the notes simple: “noticed mismatch,” “face looked odd,” or “forgot it was dubbed.” Don’t pause or replay anything until both passes are finished. Afterward, inspect those marked moments if you want to know what caused the distraction. This separates genuine viewing problems from tiny defects you would never catch at normal speed.
I’d set a fairly high bar for the altered version. Ordinary dubbing has an obvious limitation, but your brain often adjusts to it after a few minutes. A digitally changed mouth that looks wrong for even a second can make you start monitoring every face in the scene. Prime Video’s altered lips would only improve the movie for me if I stopped thinking about synchronization sooner and still read the actor’s expression clearly.
With only one segment, I’d choose the altered version only if it wins without close inspection. If you need freeze-frames to prove that its lip matching is better, the improvement probably does not matter enough. If you spontaneously notice warped teeth, a soft jawline, or a strange return to the original face after a line ends, regular dubbing is the safer choice.
Don’t test this on an action scene or anything with heavy background music. Your ears fill in a lot when there’s noise and motion, so a busy segment will flatter both versions and tell you nothing about the lip work. Pick a quiet two-hander instead.
Honestly, I think the thread is overcomplicating the decision. @scriptengineer1545 nailed the real risk: the mouth carries performance, and a smoothed-over face during a tense line reads as emotionally off even when the sync looks clean. That’s the thing that would kill it for me, not a stray warped tooth. My rule is simple. If you catch yourself thinking ‘wait, why does he look calm when he sounds furious,’ turn it off. Loose sync you forgive in minutes. A face that contradicts the voice you keep noticing the whole movie.
Whether you know which version has the altered lips is a pretty important missing detail. Once the label says “AI lip adjustment,” most people suddenly become volunteer dental inspectors. Every odd tooth or cheek movement looks suspicious, including glitches that may have been in the original compression.
Have someone else queue the two versions as A and B without telling you which is which. Watch each straight through, then write down three things: which performance felt more convincing, where your attention drifted to the mouth, and which version you would continue watching. Don’t try to identify the technology. That turns the comparison into a guessing game instead of a movie test.
I slightly disagree with the idea that emotional or joke-heavy scenes are automatically the best sample. They reveal performance problems, sure, but they can make a short comparison depend too much on whether you preferred one line reading. A plain conversation with several uninterrupted close-ups gives the lip alteration nowhere to hide. Boring science, unfortunately, is still science.
My expectation is that altered lips could help when the original timing is wildly different from the dub, but subtle mismatches rarely bother me for long. If you cannot reliably tell which version was modified yet consistently prefer one, that is useful. If you can spot the modified version because everyone’s mouth looks freshly assembled, regular dubbing wins. And if there is no clear preference, I would keep the actor’s original face rather than “fixing” a problem I had to run an experiment to notice.
Watch a scene with an actor whose face you know cold, then watch one with a foreign lead you’ve never seen before. The alteration will read completely differently in each case. On a face you’ve studied for years, even a slightly rebuilt mouth trips an alarm you can’t switch off. On an unfamiliar actor, you have no baseline, so the same edit sails past. That gap matters here because your single segment can’t tell you which situation you’ll be in for the rest of the film.
The blind A/B idea from @silentforge2620 is the smartest thing in the thread, and I’d run with it. But the short sample is a real limit nobody’s naming. Most people don’t stop noticing ordinary dubbing because it got better, they stop because their brain gives up after ten or fifteen minutes and settles in. A short clip catches you during the noisy early stretch when you’re still adjusting, which punishes both versions and tells you almost nothing about how either would feel an hour deep. If anything, that short window favors regular dubbing, since you already have years of practice ignoring loose sync and zero practice ignoring a warped jaw.
I also think the quiet-two-hander advice and the joke-scene advice are both half right. A flat conversation exposes sync flaws, sure, but it’s exactly the kind of scene your brain forgives fastest, so a win there doesn’t carry. I’d pick something in between. A scene where one person is mostly listening while another talks. The listener’s face is where these tools tend to fall apart, because the model doesn’t always know when a mouth should be doing nothing.
My actual take is blunter than most of the thread. Loose sync is a known cost I signed up for the second I clicked the dub. A generated face is a new cost with no upside if I already accepted the first one. So my bar isn’t ‘is it better,’ it’s ‘does it earn the extra risk,’ and for a feature I’d be leaving on for a two-hour movie, that’s a much higher bar than one clean segment can clear. If you walk away unsure, that uncertainty is your answer. Keep the real face.
For anyone wanting to poke at how these systems actually reshape a mouth before trusting one, it’s worth reading up on how the underlying generated mouth reshaping is put together, because knowing where the seams usually show makes your test a lot sharper.