AreaAudit

Which languages each generator says it supports, and where

Text inside the frame: the one layer nothing can fix later

Text inside the frame is any writing a model draws into the picture: a sign, a note, a phone screen. Subtitles sit on top of it and dubbing does not touch it, so it is the only language layer that cannot be repaired after the shot exists. As of 2026-09-22.

Four attempted fixes, and the two that workA subtitle lands on top of the picture. A dub replaces the soundtrack. Neither touches writing drawn into the frame. Changing the script before generation costs nothing, and shooting the writing as its own insert costs seconds per market.How do you change writing inside a generated picture?After the shot existsRegenerate itNeither a subtitle nor adub can reach it, so thepicture has to beproduced again.Before the shot existsKeep it out of frameFree, and availablewherever the story doesnot need legiblewriting.Before, if it must showShoot it as its owninsertEach market regeneratesa few seconds instead ofa scene.One claim about this capability exists in the whole register
Fig. 1 Both fixes that work happen before the picture exists, which makes this a script-stage decision rather than a post-production one.
Which language layers can reach text drawn into a picture. Recorded 2026-09-22.
LayerCan change itWhat would be needed instead
SubtitleNoA new render of the shot
DubNoA new render of the shot
Script change before generationYesNothing further
Shooting the text as its own insertYesOne short regeneration per market

Inclusion rule. The ways a production can change writing that appears inside a generated picture. Two of the four require the picture to be made again. Order. Cheapest attempted fix first, which is also the one that fails.

1It is a rendering problem wearing a language costume

Drawing legible letters is about glyph shapes, spacing and joining rather than about grammar or pronunciation. A model can speak a language convincingly and be unable to write it, and the reverse is equally possible.

That is why this register keeps such claims outside all three columns. Counting a frame-text claim as evidence about dialogue would be the same category error as counting a translated menu.

2One claim about it exists in this register

A vendor launch post read for this register claims multilingual text generation inside frames, with accurate placement, naming two languages and then saying others. It is the only statement of its kind here.

Which scripts it covers is not stated, and scripts are where the capability breaks. A model trained largely on Latin and Chinese characters is not evidence about a diacritic-heavy or right-to-left script.

3The cheapest arrangement is to avoid it

Where the story does not require legible writing, keeping it out of the frame removes the whole problem. Where a sign has to be readable, shooting it as its own short insert means each market regenerates a few seconds rather than a scene.

That is production advice this register can support, because it follows from what the layers can and cannot reach rather than from any vendor claim about quality.

A definition, not a measurement: no vendor figure appears here that is not also in the register. The one frame-text claim here is recorded beside the empty voice cell it sits next to.

Nearby terms: Burned-in text, Reading speed. The whole vocabulary is at terms; nothing on this page is a claim about a named product.