GateTrue

What each generator documents about marking its output

The two distinctions the rest of this register depends on

Two ideas do most of the work on this site. One separates a mark inside the picture from a note attached outside it. The other says what a blank cell on a pricing grid can honestly be read to mean. As of 2026-09-12.

What survives an export, split by where the mark livesA mark drawn into the picture survives a re-encode, a format change and a crop of the untouched area, because it is part of the picture. A note attached to the container survives none of those. Both are described with the same handful of words on product pages, which is the reason the distinction is worth a page.Where the mark livesOnly inside the pictureSurvives a re-encodeA visible watermark, An invisiblewatermarkBoth listsSurvives bothNothingOnly outside itRemoved by a re-exportContainer metadata, Generation notesin the file headerRead from the two explainers below
Fig. 1 The middle column is empty for a reason: almost nothing survives both routes, and a product page rarely says which one it means.

Both are the kind of distinction that feels pedantic until a file has been through an export. Metadata and watermarks fail in opposite ways: one is removed by an ordinary re-encode and leaves the picture untouched, the other stays with the picture and cannot be lifted out without touching it. Products describe both with the same three or four words, which is how a buyer ends up with the wrong one.

The second explainer is about reading tables rather than about marking at all. A feature grid is a sales document: blanks are usually there to make an upgrade obvious. Sometimes a blank is a line cut for space, and sometimes it means the question has never been addressed anywhere. Three different readings, one visual — and a register that treated all blanks alike would be inventing findings.

Neither page adds a row to the register. They exist so the rows can be read the way they were written, and so a reader who disagrees with a reading can see exactly which step of it to argue with.

1The explainers

  • Metadata vs watermark — Two marks that survive different things: one rides inside the picture, the other is attached to the container around it.
  • Reading a plan table — What a blank cell on a feature grid can mean, and why only one of the three readings is about the table.

Where a page in the register turns on one of these distinctions, it links here rather than restating the argument in a shorter and worse form.

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

2A generator's own account

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

SceneMixer writes dialogue into the script analysis, assigns each line to a shot with a speaker and a short delivery note, and the video model performs it — picture and audio generated together, so the mouth forms that sentence in the language you chose.

SceneMixer, AI Dialogue Generator (read 2026-10-13)

Each line is generated with the picture, in the voice chosen for that character, and the speaker's lips move with the words.

SceneMixer, Dialogue Lines: Delivery Notes, Voice-Over, Captions and the Voice Tag — “Is the dialogue really spoken in the video?” (read 2026-10-13)