Guide
English
Search the docs⌘K
4.4

Remove subtitles

A clip you downloaded comes with a layer of subtitles burned into it, and using it as material for an AI video means cleaning them off first. That is exactly what Remove subtitles on the DOTV video node's toolbar does, and Erase, Change and Replace subject sit in the same spot. This section covers where it is, the two ways to erase, and how they relate to Edit frame.

Where it is

Select a video node that has generated and click Edit on the toolbar above the picture — the Remove subtitles group is in there:

Remove subtitlesTwo options: Auto erase (the model finds the subtitles itself) and Box erase (you draw the boxes). Chinese and English subtitles only.
Remove subtitles sits in the toolbar's Edit dropdown; that group has two ways to erase.

Two ways to erase

Click the arrow on that button and there are two:

  • Auto erase — the model finds the subtitle area in the picture by itself, and you box nothing.
  • Box erase — you drag out the areas you want gone, and several boxes are fine. The bar carries Box, Undo, Redo and Clear all boxes.

Box erase won't submit until you have actually boxed something; Auto erase can be submitted whenever you like.

The erase bar under the frameAfter you pick Box erase: hold Box and drag over what you want gone (several boxes are fine). Next to it are Undo, Redo and Clear all boxes; the right side shows the credits this run is estimated to cost, and nothing is erased until you hit ↑.
The control bar below the picture once Box erase is chosen.

Once you have picked, a control bar hangs off the picture, and erasing only starts when you confirm. The new video node shows its own progress, then lands to the right when it finishes.

Erase, change or replace a subject

Erase / Change / Replace subject, in the same Edit dropdown, is in the same family as Remove subtitles — both "take one thing in the picture away or turn it into something else"; the difference is that subtitles are found for you, while these three you mark yourself. Opening any of them spreads a full-screen stage:

  • Erase subject — mark what should go, and every occurrence of it through the whole clip — moving along with the shot — gets wiped and the background filled back in.
  • Change subject — mark one thing and write what it should become; the model edits the source in place using the subject crop from the marked frame; camera, motion, duration and audio stay intact.
  • Replace subject — mark one thing and add an image; that thing is substituted in place while other moving parts of the clip are preserved.

Three ways to mark, pick whichever fits: Click (tap it and a vision model names what's there), Box or Brush (with brush width and an eraser). A mark only sticks once the thing has a name; everything you marked is listed in the bar at the top, where you can jump back to that frame or delete one. Up to 4 marks at a time (replace only allows 2); Erase, Change and Replace take 3–15 seconds of source — over 15 seconds only the first 15 seconds is sent, and a short side below 720 px is upscaled locally first. The film strip at the bottom shows what this run will cost; confirm and a new video node lands to the right.

Edit frame

Edit frame on the toolbar (Inpaint / Expand frame) works on the current frame: it captures the frame into an image node and opens the editor right there — see 4.1.

Try it in DOTVReading only goes so far: open DOTV and make an AI video with the feature this section covers.