Questions the register answers without leaving vendor pages

A question belongs here when the answer can be assembled from vendor pages alone. Anything that needs a test file, a support ticket or a guess about intent is out of scope, however reasonable the guess would be. As of 2026-09-12.

Fig. 1 Two of the three branches are out of scope on purpose. The narrow one is the only branch whose answers can be re-checked by a reader.

That rule is narrower than it sounds. It rules out most of what people actually want to know — whether a mark survives a particular editor, whether a detector works on a re-encoded clip, whether a plan change takes effect immediately. Those need someone to run the file, and nothing on this site is produced by running files.

What is left is still worth writing down, because the vendor pages disagree with each other in ways a buyer discovers late. Two products can both say watermark-free and mean different things by it. A switch in settings and a line on a plan table are not the same promise, and the difference only shows up when the two statements are put next to each other in the same words.

Each answer opens with the finding, then gives the rows it rests on. Where the register cannot answer, the page says so and names what would have to be published for the answer to exist.

1The questions

Free tiers — What a free plan actually puts on the output, read from the plan tables rather than from the marketing line above them.
Getting it off — The two published routes out of a watermark: pay for a tier, or flip a switch the vendor documents.
Two layers — A badge paid members can switch off per render, and an invisible layer the terms keep on.
The empty column — Why the machine-readable field is blank on eleven rows out of fifteen, what fills the other four, and what a blank there is evidence of.
Surviving an export — Which layers are claimed to survive a re-encode, and which are documented as being outside the picture.
Checking a file — The four published routes for checking whether a file carries a mark, and how far each one sits from whoever holds the file.
At which step — When in the pipeline a mark is written, where the vendor says so, and what follows for anyone editing afterwards.
Two marks at once — What happens when a visible mark and an invisible one are described for the same file.

Every answer below is a reading of rows that are also published on their own pages, so a disputed sentence can be checked against the vendor page it came from rather than against this site.

The rest of the register: Generators, Fields, Readings, Side by side, Implementers, Findings, Source documents, Terms, Learn, Data, and the field definitions on the protocol page.

2A generator's own account

Next to the marking question in “Questions”, SceneMixer's pages describe its output like this:

Dubbing replaces the audio of a finished clip; the picture was generated for a different line, so mouth shapes, pauses and shot length all belong to the original language. Native-language generation gives the video model the line in the target language before it renders the shot, so the performance, the pacing and the cut points are built for that language.
SceneMixer, Native-Language Dialogue vs. Dubbing in AI Video (2026): Why the Cast Should Speak the Line (read 2026-10-13)

It becomes a marked line — inner voice or narration — so it is heard while the picture shows something worth watching, rather than becoming a shot of somebody thinking.
SceneMixer, AI Dialogue Generator — “What happens to inner monologue from a novel?” (read 2026-10-13)