Dubbing has always carried a visible flaw: the mouth on screen forms the sounds of a language the audience is not hearing. Tools that reshape the mouth to match the new audio have changed the economics of releasing video across languages.
The mismatch was previously handled by script
Traditional dubbing employs adapters who rewrite dialogue so that translated lines approximate the original mouth shapes and line lengths.
This constrains meaning considerably, since the closest visual match is often not the closest translation.
Removing the constraint lets the translation optimise for sense, which is a larger change to dubbing quality than the visual fix itself.
Only the mouth region is regenerated
These systems detect the face, isolate the mouth and surrounding area, and regenerate it frame by frame conditioned on the new audio.
The rest of the frame is untouched, which keeps computation manageable and preserves the original performance in the eyes and body.
Failures appear at the boundary of the regenerated region, particularly with facial hair, strong side lighting or a hand crossing the face.
Voice cloning is the more contested half
Full localisation pairs the visual work with a synthesised voice carrying the original speaker's timbre into the new language.
This is where performer consent becomes the live issue, since a voice is a recognisable attribute that can be reproduced without the person being present.
Union agreements in several markets now treat voice reuse as a separately licensed right rather than as part of a performance fee.
The economics changed at the low end
Conventional dubbing is expensive enough that it is reserved for titles with confident international demand.
Cheaper localisation extends coverage to back catalogue, training material, corporate video and independent work that was previously subtitled or not released at all.
The effect is less about improving flagship releases and more about the volume of material that becomes viable to localise at all.
Detection of the technique is becoming a requirement
The same mechanism that dubs a film also makes a person appear to say something they did not.
Legitimate localisation is therefore converging on disclosure: provenance metadata attached at export, and a visible statement that a track was synthesised.
That distinction between labelled and unlabelled use is likely to matter more than the technical quality, because the quality question is close to settled and the labelling question is not.