How to remove the same small part from every item in a shot
Ask Vizard Agent to find every appearance of the part before it removes any of them. A pan across a rack of identical units shows the same fitting a dozen times at a dozen sizes, and the ones you did not notice while typing are exactly the ones a single-pass removal leaves behind.
What is the short version?
You saw one white plug and asked for it to go. The camera moves, and there are nine of them — some large and obvious, some four pixels wide at the back of the rack, and a few bright spots on the plastic that are not plugs at all.
- Tell Vizard Agent what the part is, on every unit that shows it.
- Ask it to map when and where each one appears.
- Ask for a check on every location, not a look at the output.
What do you need before you start?
The clip and a name for the thing. Vizard Agent works far better from "the white drain plug on each box" than from "the white dot", because the first tells it what shape and material to look for when it starts separating real parts from reflections.
- The clip. As filmed, including the camera move.
- What the part is. A plug, a vent, a screw, a label.
- How many units. Roughly, so nothing is missed.
- What it must not touch. Edges, highlights, printed text.
- Whether the audio matters. It is kept either way.
What do you type into Vizard Agent?
Say "every" and mean it. A request naming one plug gets one plug removed perfectly, and Vizard Agent has no reason to go looking for the others unless the instruction tells it that the part repeats across the units in the shot.
Prompt
Variants worth knowing:
- "From every box." The instruction that turns one job into a census.
- "Map every appearance first." A list you can check before the render.
- "Keep the original audio." So the picture work never touches the sound.
What does Vizard Agent actually do?
Here is the Vizard Agent sequence on a product clip where the same white plug sits on each of several stacked boxes. Much of the work happens before anything is removed, and one of the early approaches is abandoned outright.
- Reviews the duration and samples frames.
- Checks which object-removal capabilities are available.
- Splits the clip into four parts and runs removals in parallel — the fast route.
- Inspects each part and finds one has lost its picture, then diagnoses why.
- Abandons the split approach and returns to the source file.
- Walks the clip second by second to find every plug position.
- Maps the appearances temporally — which plug is visible when.
- Zooms into each candidate, at six, ten, twelve, fourteen and nineteen seconds.
- Tests shape detection so that bright highlights on the plastic are not treated as plugs.
- Removes the plugs with a moving mask while preserving the original audio.
- Measures the residual positions after the render and widens the mask on the largest plug.
- Runs a final contact-sheet check on every plug location, plus motion and audio sync.
Step six is the difference between this and an ordinary retouch. Vizard Agent is not tracking one object through a clip; it is building a register of appearances, because a plug that is on screen for eleven frames still leaves a visible white dot if it is skipped.
Step nine is the one that stops the cure being worse than the disease. Moulded plastic throws specular highlights that are round, white and about the right size, and a detector tuned only on brightness would smear a dozen of them. Vizard Agent tests the shape separately before it trusts the detection.
Step three failed, and it is listed for that reason. Splitting a clip into parts and removing in parallel is genuinely faster when it works; here one part came back without picture, Vizard Agent diagnosed it rather than re-running it hopefully, and switched methods.
What does the result look like?
Every box in the pan reads as a clean moulded surface, including the ones at the back of the rack that are only on screen for a moment. The plastic still has its highlights, the camera move is as smooth as it was, and Vizard Agent has checked each location on a contact sheet rather than by watching a playback.
When does this not work well?
Some parts cannot be removed from every unit equally. A plug at the front of frame has enough surrounding surface to rebuild from; the same plug on a box twenty feet back is a handful of pixels sitting on an edge, and Vizard Agent will often leave it rather than paint a smudge there.
- Parts on the silhouette edge. There is no surface to rebuild.
- Very small instances. A few pixels, with no clean surround.
- Parts that are half-occluded. Behind a hand, a strap or another box.
- Strong reflections on the part itself. It reads as part of the lighting.
- A camera move with heavy motion blur. Each instance smears differently.
How do you fix a result that came back wrong?
Point at the one that is still there, by timestamp and position. Vizard Agent reopens the register of appearances and adds or widens that entry rather than re-running the whole detection, which would risk changing the ones that are already right.
- "The big one at 0:19 is still visible." That mask is measured again and widened.
- "You missed the yellow box." That appearance is added to the map.
- "A highlight has gone flat." Shape detection is retightened to exclude it.
- "There is a grey patch where it was." The repair is rebuilt from the surrounding surface.
How does Vizard Agent compare to doing it yourself?
By hand you fix the one you noticed. The client watches it, spots a second, then a third, and each round is another pass through the timeline finding where that particular plug enters and leaves the frame. The work is not hard; it is bookkeeping, done under a deadline.
| By hand | Vizard Agent | |
|---|---|---|
| Finding the instances | The ones you happened to see | A second-by-second register |
| Small instances | Usually skipped | Assessed, and skipped deliberately |
| False positives | Smeared with the rest | Shape-tested before removal |
| Fast route | Tried, then lived with | Tried, diagnosed, abandoned |
| Final check | Watch it back | Every location on one sheet |
Common questions
Why not just remove the one I can see? Because the camera moves and reveals the others. Vizard Agent maps them all first.
Will it find the ones I missed? That is the point of the mapping pass — Vizard Agent walks the clip rather than the description.
What about reflections that look like the part? Vizard Agent tests the shape so highlights on the plastic are not removed.
Does the audio survive? Yes. Vizard Agent keeps the original track through the picture work.
Can it do it in parallel to be quicker? Sometimes. Here the split render dropped a part's picture, so it went back to one pass.
How small is too small? When there is no clean surface around the part, a repair looks worse than the part.
Will the camera move still look smooth? Yes. Vizard Agent checks motion continuity and edit joins at the end.
Can I ask for only some of them? Yes. Name which units, and Vizard Agent leaves the rest alone.
What if the part appears again later? The register covers the whole clip, not the section you mentioned.
Does it change the colour of the box? No. The repair is built from the surrounding surface of that same box.
Can it handle a different part on each unit? Yes, though naming each one gives Vizard Agent a much better detection.
How do I check the result quickly? Ask Vizard Agent for the contact sheet of every location rather than the video.
Does it work on stills too? Yes, and it is far simpler — there is only one appearance to find.
What if a unit is only half in frame? It is treated if enough of the part is visible, and flagged if it is not.