Vizard Agent

How to remove a detail from a product that hands keep covering and uncovering

Last updated 2026-09-17 · 8 min read

Ask Vizard Agent to find every moment the feature is visible and treat them as separate windows rather than as one continuous removal. A detail that is bare in one shot, half under a lid in the next and being covered by a hand in a third needs a different approach in each, and a single pass across the clip will fail at exactly the moments somebody is touching it.

What is the short version?

A continuous removal works on something continuously visible. As soon as a hand or a lid moves across the feature, what needs removing changes shape, and a pass built for the clear shot breaks where the object is handled.

  1. Ask Vizard Agent to list every moment the detail is visible.
  2. Ask for each state handled as its own window.
  3. Ask for a before-and-after comparison of each.

What do you need before you start?

The clip and a reference for what it should look like instead. Vizard Agent rebuilds the surface, so a picture of the version without the feature — a render, a prototype, a similar product — gives it something to match rather than invent.

What do you type into Vizard Agent?

Mention the handling to Vizard Agent explicitly. It is the detail people invariably leave out of the brief, and it is the single thing that decides whether this job is one straightforward pass or four separate ones with different methods.

Prompt

Remove the screw thread from this jar for the whole twenty-six seconds so it reads as a jar with no thread. Watch the moments where the lid goes on and comes off — the thread must not be visible there either. Find every state it appears in, handle each separately, and show me before and after on each one.

Variants worth knowing:

What does Vizard Agent actually do?

Here is the Vizard Agent sequence on a real product clip where a surface feature had to disappear entirely. The survey takes considerably longer than the retouch itself, because working out how many separate problems you actually have is most of the job.

  1. Looks at the reference of how the product should appear.
  2. Samples frames across the whole clip.
  3. Builds a closer overview of the area at its different stages.
  4. Checks up close every moment the feature could appear under the lid.
  5. Reviews the opening frame by frame, then each lid state in turn.
  6. Prepares separate fragments with exact boundaries for retouching.
  7. Removes the feature in the opening and under the first lid.
  8. Removes it during the lid being taken off and put back on.
  9. Builds frame-by-frame comparisons of each fragment.
  10. Reviews each window before and after at full size.

Step four is the one that defines the job. Listing where the feature is obviously visible is easy; finding the frames where it is half-hidden, partly reflected or just catching the light is what stops the finished clip flickering.

Step six is why this works. Exact boundaries mean each window can use whatever approach suits it, and a window where a hand crosses the feature does not have to share a method with a clean static shot.

Step nine is not optional on product work. At playback speed a retouch looks perfect; frame by frame you can see it breathe, and a client looking at a product close-up will pause it.

What does the result look like?

The product as you want it shown, throughout. No flicker at the moments it is handled, the surface underneath matches the rest of the object, and Vizard Agent gives you the before-and-after pairs for each window rather than one overall view.

When does this not work well?

Some of this should not be done at all. If the video is an advertisement for the product as sold, then removing a feature it actually has is misrepresenting it, and the honest version is to show what people will receive.

How do you fix a result that came back wrong?

Name the window it happens in. Vizard Agent handled each state separately, so a problem at the moment the lid comes off gets fixed in that window alone, without disturbing any of the clean static shots elsewhere in the clip.

How does Vizard Agent compare to doing it yourself?

By hand you mask the feature, track it, and it works beautifully until a hand crosses the frame and the tracker jumps. Then you keyframe around the handling, and the retouch breathes in and out for the two seconds that matter most.

By hand Vizard Agent
The unit of work One continuous mask One window per state
Handling moments Where the track breaks Treated as their own problem
The survey The obvious appearances Every frame it could show in
Checking At playback speed Frame by frame, per window
What the surface becomes Invented Matched to a reference

Common questions

Why split it into windows? Because each state is a different problem. Vizard Agent solves them separately.

Do I need a reference? It helps a great deal. Otherwise Vizard Agent has to invent the surface.

What about the lid moments? Those are the hard ones, and the reason Vizard Agent surveys before retouching.

Will it flicker? Not if each window is checked frame by frame, which is why that step exists.

Can it do the whole clip in one pass? Sometimes, for a static shot. Vizard Agent will say when one pass is enough.

What about reflections? They show the feature too. Tell Vizard Agent to include them.

Does the rest of the product change? No. Vizard Agent restricts the correction to the feature and its immediate surface.

Is this honest? For a concept or design preview, yes. For an advert of the product as sold, Vizard Agent will say no.

What if the product rotates? Harder. Vizard Agent will tell you where the rebuild cannot keep up.

Can it show me the states it found? Yes, and it is worth reading before the retouch starts.

How long does it take? Mostly survey time. The retouch itself is quick once the windows are set.

Can it change the feature instead of removing it? Yes — the same method applies to replacing it with a different one.

Does resolution matter? Yes. Vizard Agent needs detail to rebuild a convincing surface.

What about shadows the feature casts? Vizard Agent removes those too, or the shot reads as wrong without anybody knowing why.