Vizard Agent

How to remove a logo or watermark from a video

Last updated 2026-08-16 · 5 min read

Upload the video to Vizard Agent and say which mark to remove. Vizard Agent extracts full-resolution frames, crops into the corners to read exactly what is there, works out the pixel coordinates of the mark, and rebuilds that region alone. Everything outside those coordinates is left exactly as it was.

What is the short version?

Removing a mark from a video is a small, precise job that is tedious by hand and easy to do badly. The usual bad version is a blur box, which announces that something was removed. The good version rebuilds what should have been behind it, and that starts with knowing exactly where "there" is.

  1. Go to Vizard Agent and upload the clip.
  2. Say which mark to remove and roughly where it sits.
  3. Say if there is other branding that must stay.

What do you need before you start?

The clip and a description of the mark. Vizard Agent will find marks on its own by looking at the frame, but corners often hold more than one thing — an export watermark from an editing tool, a channel logo, a timestamp — and it needs to know which of them you want gone and which are yours.

What do you type into Vizard Agent?

Name the position and the wording if you know it. Vizard Agent reads the corners itself, and giving it the text is the fastest way to be certain you are both talking about the same mark, particularly when a frame carries two.

Prompt

Remove the [brand label / watermark] from this video — the one in the [top left] and the [clideo.com] text at the bottom right. Leave everything else unchanged.

Variants worth knowing:

What does Vizard Agent actually do?

Here is the order Vizard Agent worked in on a real watermark removal. Almost the whole run happens before any pixel is changed: it is finding out precisely what is there and precisely where, because a rebuild that is off by ten pixels looks worse than the mark did.

  1. Extracts frames from the video to see the marks that need removing.
  2. Checks the video's technical information.
  3. Extracts several frames at once and reads them as a set.
  4. Extracts a full-resolution frame to analyse the coordinates precisely.
  5. Crops into the top-left and bottom-right regions separately for a closer look.
  6. Reads each cropped region — the logo in one corner, the watermark text in the other.
  7. Works out the exact pixel coordinates of the white logo, and separately of the text.
  8. Crops the text alone to confirm its position before rebuilding.

Step 8 is the tell. Vizard Agent checks its own answer by isolating the mark one more time, which is what stops the rebuild from being applied to slightly the wrong rectangle.

What does the result look like?

From the run this page is written from, probed on the delivered file: 720x1280, H.264, 25fps, 20.08 seconds, audio preserved. The source resolution and frame rate were kept, because this is a repair rather than a re-encode for a new destination, and Vizard Agent had no reason to change either.

Ask for a different output shape in the same conversation and Vizard Agent re-frames the cleaned video rather than re-doing the removal.

Expect tens of minutes rather than minutes. This category was not separately measured; comparable work runs a median of 28 to 38 minutes end to end, and across all projects the median cost by tier is Flash 47, Pro 55, Max 242, Ultra 263 credits.

When does this not work well?

The rebuild is generated rather than recovered. Vizard Agent puts back something plausible where the mark was, and how convincing that turns out to be depends entirely on what the mark was sitting on top of — which is why the same job is trivial in one shot and impossible in the next.

How do you fix a result that came back wrong?

Point at the corner. Vizard Agent keeps the coordinate analysis and the rebuilt region, so a correction adjusts that rectangle rather than re-scanning the video, and the parts of the frame you were happy with are never touched.

How does Vizard Agent compare to doing it yourself?

By hand this is masking and content-aware fill, frame by frame if the shot moves. The technique is well known and the work is fiddly: find the exact box, feather it, fill it, then check every frame for the moment the fill slips.

By hand Vizard Agent
Finding the mark Zoom in and eyeball the box Crops in and reads the coordinates
Confirming the box Trust your eye Isolates the mark again to check
Filling Content-aware fill, then patch Rebuilds the region
Two marks Two passes Treated as two regions in one pass

Common questions

Can it remove a watermark from a video I downloaded? Technically yes. Whether you may use that footage is a separate question, and one that removing the mark does not change.

Can it remove my old logo and add a new one? Yes. Say both, and Vizard Agent puts the new one in the same place.

What if the mark moves around the frame? Say so. It is harder and slower than a fixed corner mark, and Vizard Agent will track it rather than assume a fixed rectangle.

Can it handle two marks in different corners? Yes, and Vizard Agent treats them as separate regions with their own coordinates, so one can be removed while the other is kept.

Will the rest of the video be re-encoded? The file is written fresh, but Vizard Agent keeps the source resolution and frame rate unless you ask for something else.