How to remove a cast shadow without cutting into the person

Last updated 2026-09-22 · 8 min read

What is the short version?

There is a hard shadow on the wall behind the speaker. It moves whenever they move, so Vizard Agent cannot simply patch over it, and it touches them at the edges, so it cannot be masked carelessly either without taking a piece of their outline.

1. Tell Vizard Agent to remove the shadow on the background only.
2. Say the subject stays exactly as they are, hair included.
3. Ask to see the mask before the whole clip is rendered.

What do you need before you start?

The clip and a look at the wall. Vizard Agent checks whether any part of the background is ever clean — a moment when the shadow is out of frame gives it something to rebuild from, which changes the whole approach.

• The clip. The original file.
• Which shadow. There is often more than one.
• How much the subject moves. It sets the tracking.
• What the background is. Flat wall, texture, a room.
• Whether the audio matters. It is kept either way.

What do you type into Vizard Agent?

Say "behind" and say "only" in the same sentence. A bare request to remove the shadow reads to any general tool as a lighting correction, and a lighting correction lifts the subject's own shading along with the shadow that is on the wall behind them.

Prompt

Variants worth knowing:

• "On the wall behind." Which shadow, precisely.
• "Their own shading untouched." Rules out a global lift.
• "Evenly lit where the shadow was." What success looks like.

What does Vizard Agent actually do?

Here is the Vizard Agent sequence on a clip with a hard shadow on the wall behind a speaker. Separating the person from that shadow is most of the work in the list, and it gets attempted three different ways before one holds.

1. Checks the duration and reviews representative frames.
2. Locates the shadow on a contact sheet.
3. Extracts detailed frames of the subject and the shadow together.
4. Checks the audio before preparing any export.
5. Inspects a frame at a useful size to judge whether removal is possible.
6. Looks at the opening for a clean area of background to rebuild from.
7. Looks at the ending to see how the subject and the shadow move.
8. Checks for a local segmentation tool to isolate the subject.
9. Tests a subject mask and prepares a control view.
10. Builds a contrast-based mask, then checks whether it keeps the subject without including the shadow.
11. Looks for a person-segmentation model that holds the hair and shoulder edges.
12. Prepares a temporary version so the edit runs without dropping frames.
13. Generates a shadow-removal test with everything else untouched, and reviews the shadow and the face.
14. Repeats the test with a more precise instruction about the projected silhouette.

Step ten is where most attempts fail. A mask built on contrast treats the shadow as part of the subject, because it is dark and adjacent — so the removal either leaves the shadow or eats the person's outline.

Step eleven is the fix that works. A model built for people holds the hair and the shoulder line, which is exactly where a contrast mask breaks, and that accuracy is what makes the rest possible.

Step six matters because it decides the method. If the wall is ever visible clean, the shadow area can be rebuilt from real pixels; if it never is, the wall has to be recreated, which is a harder job with a different failure mode.

What does the result look like?

The same clip with an evenly lit wall behind the subject: the person unchanged, their own shading intact, no bright patch travelling behind them as they move, and the audio and the duration exactly as Vizard Agent received them.

When does this not work well?

Some shadows cannot be separated at all. A soft shadow that merges into the subject, one thrown across a textured or patterned wall, or one falling partly on the person makes a clean removal impossible however carefully Vizard Agent masks it.

• Soft-edged shadows. No boundary to work from.
• Patterned backgrounds. The pattern has to be rebuilt inside the area.
• Shadows on the subject. They are part of the lighting on them.
• Several overlapping shadows. Each needs its own treatment.
• Fast movement. The mask lags and the shadow flickers.

How do you fix a result that came back wrong?

Say whether you can still see the shadow or whether you can now see the repair. Vizard Agent adjusts the mask or the fill depending on which you name, and in either case the person's edge is the first thing worth checking.

• "His hair looks cut out." The person mask is rebuilt with a proper segmentation.
• "The wall has a bright patch." The fill is matched to the surrounding light.
• "The shadow flickers." The mask is stabilised across frames.
• "It is still there at the end." That range is re-tracked where the subject moves.

How does Vizard Agent compare to doing it yourself?

By hand you either brighten the background, which lifts the subject too, or you roto the person for every frame. Both are the same realisation arriving at different speeds: the shadow cannot be treated until the person is separated.

Common questions

Can it remove a shadow on the floor? Yes, and Vizard Agent treats it the same way as one on a wall.

What if the shadow falls on the person? That is their own lighting, and Vizard Agent will say it cannot be removed cleanly.

Will the wall look flat? Not if the fill is matched to the surrounding light, which Vizard Agent checks.

Can it keep a soft contact shadow? Yes. Tell Vizard Agent which shadow to keep and which to remove.

Does it work if the camera moves? Yes, though Vizard Agent has to track the background as well as the subject.

What about two people? Each casts a shadow, and Vizard Agent masks each person separately.

Will the audio change? No. Vizard Agent treats this as a picture-only correction.

Can it replace the background entirely instead? Yes, and where the shadow is complicated that is sometimes simpler.

How long does it take? Most of the time goes into the mask rather than the render.

Can I see the mask? Yes. Ask Vizard Agent for the mask view before it renders anything.

What if the wall is textured? Vizard Agent will tell you when the texture cannot be rebuilt convincingly.

Does it work on a still photo? Yes, and it is far easier — one mask rather than hundreds.

Can it soften rather than remove? Yes. Reducing a shadow often looks more natural than erasing it.

How do I check it? Watch the wall, not the person, and look for a patch that moves with them.