Labelling video

Standing time, the recommended setting, and what the video editor offers.

A video asset opens an editor of its own: the stage with the label over it on top, the timeline below, settings and the save card on the right.

There are no two editing steps here. The „Zuschnitt“ / „AI-Label“ tabs belong to the image editor; with video everything sits on one screen at once.

The signed-in product speaks German, so its labels and messages are quoted in German throughout, with the English rendering in italics.

Standing time

The timeline knows three states:

  • permanent — the label stands for the whole runtime.
  • timed — it stands within a window you set on the timeline.
  • faded — the same, with a fade in and out at the edges.

The editor always starts on permanent. Leave that state and a note appears:

Permanent empfohlen für die AI-Act-Kennzeichnung — ein zeitweises Label kennzeichnet das Video nur währenddessen.

Permanent is recommended for AI Act marking — a temporary label marks the video only while it shows.

Whether, and to what extent, the EU AI Act requires a marking from you depends on your role and on the individual case. This is not legal advice.

The note does not stop you — all three versions can be saved.

Switching to a window starts at the full runtime; the fade starts at one second, but at most at half the window length. The window cannot be dragged arbitrarily small — the floor is a tenth of a second, and the fade may take at most half of it.

While the video duration is unknown, the mode cannot be left. The toggle then does nothing and no message appears — wait a moment for the metadata.

What the video editor does not offer

No format picker and no crop. A video derivative keeps the source's resolution; there is no target format and so nothing to crop to.

No on/off switch for the label. A video derivative always carries a label; the switch row is not shown at all. It follows that without a label family no video derivative can be saved — there is no valid specification, and the button stays inactive.

No automatic variant selection. The video editor always shows the variant tiles for a manual choice — even for a family with only one variant.

What it does offer

Automatic size calculation. The „Größe & Rand“ toggle with „Automatisch“ and „Manuell“ is here too, and its canvas is the native video frame.

While the metadata is missing — or never arrives after a playback error — the stored numbers apply; saving still works. Free placement does disappear in that case: with no frame dimensions there is no draggable overlay, so only the anchor grid remains.

For a portrait video the editor turns the label to the vertical orientation by itself — once the metadata is in, and provided your workspace has not explicitly set the rotation otherwise.

Below the stage the timeline shows ten real frames from the clip. Where that fails, a gradient stands in their place — the labelling is unaffected.

When the frame is too small

Where the visible type does not fit the video frame, nothing is burnt in: no specification is produced, the preview shows no label, and the save button stays inactive. "Too small" is a statement about the visible type width, not about a pixel count.

The way out is „Größe & Rand“ on „Manuell“; the button beside the warning switches exactly that.

Prerequisites

Verification has to be through. Until it is, the lock message with a retry button stands in place of the editor.

The processing limits apply at upload already: 3 minutes of runtime and Full HD. Both are checked when the file is picked, not at render time.