How to remove a lens glare that moves across the frame

Last updated 2026-09-17 · 8 min read

What is the short version?

Removing an object from a shot means rebuilding whatever was behind it. A glare has nothing hidden behind it at all — the picture is still sitting there underneath, simply tinted — so the job Vizard Agent has to do is subtraction rather than reconstruction.

1. Tell Vizard Agent the glare moves and roughly where it is.
2. Ask it to measure the colour and build a mask from that.
3. Ask to see the mask before any correction is applied.

What do you need before you start?

The clip and a description of the artefact. Vizard Agent needs to know it is dealing with a light artefact rather than an object, because the two call for opposite approaches and the wrong one wastes a lot of time.

• The clip. The original, not a re-encode.
• The colour. Green, magenta, warm white.
• Where it appears. Even loosely.
• Whether it moves. It almost always does.
• What shares that colour. A green jacket is a problem.

What do you type into Vizard Agent?

Describe it to Vizard Agent as a tint rather than as a shape. Calling it a mark or a spot sends the whole job towards patching, and a patch cannot follow something that changes its position, its size and its shape on every single frame.

Prompt

Variants worth knowing:

• "A light artefact rather than an object." Sets the approach.
• "Only that halo and not the bike." The false-positive risk.
• "Show me the mask first." Cheaper than a render.

What does Vizard Agent actually do?

Here is the sequence on a real clip with a drifting green flare. Almost half of it happens before any correction is applied, because a mask that catches the wrong thing produces a worse result than the glare did.

1. Builds an overview of the frames to see the glare move.
2. Reviews the movement frame by frame in more detail.
3. Pins down the glare's shape and trajectory at individual moments.
4. Measures the glare's colour to separate it from subject and background.
5. Zooms into the area where the halo is strongest.
6. Tests a mask that should touch only the coloured halo.
7. Checks the mask catches nothing else in the frame.
8. Builds a trial correction on the frame where the glare is worst.
9. Compares the corrected trial frame against the original.
10. Strengthens the suppression for the residual halo and re-renders.

Step four is what makes the mask possible. A flare has a colour signature that the scene underneath mostly does not share, and measuring it turns "the green bit" into a range the correction can target.

Step eight is a deliberate choice of test frame. Correcting the average frame tells you nothing about the moment the glare is at its strongest, and that is the moment anybody watching will notice.

Step ten is the second pass people skip. A first correction usually leaves a faint residue, and deciding how hard to push is a judgement made by looking at the trial rather than by running the whole render twice.

What does the result look like?

The shot without its coloured wash. The picture underneath is entirely intact, because it was never removed in the first place — only untinted — and Vizard Agent shows you the worst frame before and after rather than offering a general assurance that it worked.

When does this not work well?

Colour separation fails whenever the scene itself shares the colour. Green foliage under a green flare, a magenta sign under a magenta one, or a glare so bright that it has clipped to white will each defeat the method, and Vizard Agent will say so before wasting a render on it.

• The scene shares the colour. The mask catches both.
• A clipped, blown-out flare. There is no detail underneath to restore.
• Very large flares. They cover most of the frame.
• Flares across a face. Skin tones shift with the correction.
• Heavily compressed sources. The colour edges are blocky.

How do you fix a result that came back wrong?

Say whether the glare is still visible or whether something else in the shot has changed colour. Those are the only two directions this can fail in, and Vizard Agent narrows or widens the mask accordingly rather than starting the correction again.

• "There is still a halo." The suppression is strengthened.
• "The bike went grey." The mask was too wide; it is narrowed.
• "It works in one shot only." The colour is re-measured per shot.
• "The edges look smeared." The correction is softened at the boundary.

How does Vizard Agent compare to doing it yourself?

By hand you try to patch it, discover it moves, then try to track it, and discover it changes shape as well as position. The usual outcome is a blur that follows the glare around, which is more visible than the glare.

Common questions

Why not just patch it? Because it moves and changes shape, and the picture under it is fine. Vizard Agent untints instead.

What if something else is that colour? Then the mask catches it too. Vizard Agent checks for that before correcting.

Can it handle several flares? Yes, though Vizard Agent may need a separate colour measurement for each one.

Will the shot look different afterwards? Only where the glare was. Vizard Agent leaves everything else untouched.

What about a flare on a face? Riskier — skin shifts with the correction. Vizard Agent will show you the trade.

Can it do this in one pass? Usually two. Vizard Agent corrects, then makes a second pass for the residue.

What if the flare is white? Much harder. Vizard Agent has no colour to target, and often no detail underneath.

Does it work on drone footage? Yes, and Vizard Agent sees moving flares there more often than anywhere else.

Can I see the mask? Yes, and you should. It is the cheapest thing to check.

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

Does it need the original file? It helps. Compression blurs the colour boundaries the mask relies on.

Can it strengthen it selectively? Yes. Vizard Agent can push harder where the glare is worst and less elsewhere.

What about a lens smudge instead? A different problem. That is an object, so Vizard Agent reaches for a patch.

How do I describe it? By colour and movement. Those are the two facts the method needs.