Where in a pipeline the language decision is made
A language can be chosen before anything is generated or applied to a finished file afterwards. The two produce the same deliverable through different pipelines, with different costs, review cycles and failure modes. As of 2026-09-12.
| Stage | Decided up front | Applied afterwards |
|---|---|---|
| Number of passes | One | Two or three |
| Mouth movement | Produced with the line | Repaired after the fact |
| Line length | Unconstrained | Has to fit the original timing |
| Cost of another language | Another generation | Another conversion pass |
Inclusion rule. The two arrangements compared as production stages. Neither column asserts anything about the quality of either result. Order. The single-pass arrangement first.
1Deciding late is constrained by a picture that exists
A replacement line has to fit timings, mouth movements and pauses made for a different sentence. That is why translating for length matters as much as translating for meaning once the picture is locked.
Deciding early removes the constraint entirely, because nothing was timed to another sentence. It moves the language decision to the front of the schedule, where it constrains everything downstream instead.
2Changing your mind costs differently
With the language applied afterwards, adding one is a new pass over an existing picture. With it decided up front, adding one is a new picture as well, at whatever a generation costs.
For a simultaneous multi-language release the second arrangement can be cheaper anyway, since every version is produced the same way rather than one being converted from another. For a catalogue already shot, it is not available at all.
3The failure modes are different too
A late decision fails visibly at the mouth and at the reading limit. An early decision fails at consistency: a character whose voice or manner drifts between episodes, because each was generated separately.
Neither failure appears in any language list. What a list can tell a reader is which arrangement a vendor is describing, and that is enough to know which of the two sets of problems to plan for.
4Ask which arrangement before comparing numbers
Two vendors with the same language figure can be selling a conversion service and a production tool. The figure is the same and almost nothing else is.
So the first question about any language claim is not how many but at what point. The answer is on a vendor page surprisingly often, and it reorganises the comparison more than the counts do.
An explainer about method and structure, not a count. No vendor number appears here that is not also in the register. The sourced material is on the register. Related: Scripts in pictures, Voices for a series.