How to make a semi-transparent garment fully opaque in a video

Last updated 2026-08-20 · 6 min read

What is the short version?

A thin fabric that shows what is underneath on camera is a lighting problem you cannot re-shoot. Fixing it in the edit means isolating that fabric specifically — not darkening the shot, not touching the face — and holding that selection steady as the person moves.

1. Go to Vizard Agent and upload the clip.
2. Name the garment and say it should be fully opaque.
3. Say what must not change — the face, the rest of the outfit, the background.

What do you need before you start?

The clip and a clear boundary. Vizard Agent finds the garment itself and works out the mask, and the thing worth stating plainly is everything that must survive untouched, because a fix that also darkens the skin or the background is not a fix.

• The clip. Any resolution. This run was 4K vertical.
• The garment. Named plainly: the headscarf, the shirt, the dress.
• The change. Fully opaque, darker, a different shade.
• What stays. The face, the other clothing, the background.
• The audio. Usually untouched. Say if it should change.

What do you type into Vizard Agent?

Say what must not change in the same breath. Vizard Agent treats that as a constraint on the mask rather than a note, and stating it up front is what keeps the correction confined to the fabric instead of spreading across the frame.

Prompt

Variants worth knowing:

• A different shade. Say the colour and Vizard Agent tunes towards it.
• Only part of the clip. Give the timecodes if the problem is not throughout.
• A test first. Vizard Agent renders a short section before committing anyway.

What does Vizard Agent actually do?

Here is the order Vizard Agent worked in on a real opacity fix. Nearly all of it is building and proving a mask; the actual correction is a handful of steps near the end, which is the correct proportion for this kind of work.

1. Probes the clip and extracts frames to look at the garment closely.
2. Crops into the problem area and inspects the fabric where it is thinnest.
3. Measures the actual colour values across different parts of the frame.
4. Builds a coordinate grid to fix the garment's position precisely, then checks how it moves over time.
5. Loads a clothing segmentation model, tests it live and visualises the result to judge the mask.
6. Measures the model's speed and tunes it so a full-resolution pass is affordable.
7. Smooths the mask edges with a guided filter and combines it with colour-based detection.
8. Tests three darkness and opacity curves, looks at each result, then examines the hardest area — a translucent flap at the edge — separately.
9. Renders a three-second test to check the fix holds steady across frames, validates it at four points in the clip, then processes every frame in parallel at full resolution and remuxes the original audio.

Step 9 first is what stops a wasted render. A correction that looks perfect on one still can flicker across frames, so Vizard Agent proves temporal stability on a short section before spending a full-resolution pass on 1,180 frames.

What does the result look like?

From the run this page is written from, probed on the delivered file: 2160x3840, H.264, 60fps, 19.66 seconds, AAC audio. Full 4K vertical, twenty seconds, every frame processed, the garment opaque, and the original audio carried across untouched.

Keeping the delivery at the source resolution matters more here than in most edits. Downscaling would hide the fix and the flaw together, so Vizard Agent processed at full size in parallel rather than taking the cheaper route.

When does this not work well?

This is frame-by-frame correction against a moving subject, and the quality of the mask is where the whole job succeeds or fails. Vizard Agent tests on stills before committing to a render, and some material simply makes a clean selection impossible.

• Fast motion blurs the edges. A blurred boundary between fabric and skin resists a clean mask.
• Similar colours confuse the selection. A dark garment against a dark background is the hard case.
• Very thin fabric may still read as fabric. Vizard Agent increases opacity; it does not repaint what is behind.
• Long clips cost time. Every frame is processed at full resolution, which is why this run was twenty seconds.
• It will not change the person. The face and body are explicitly outside the mask.

How do you fix a result that came back wrong?

Name the moment and the area. Vizard Agent keeps the mask, the tested curves, the still comparisons and the validation strips, so tightening the selection or easing the correction is another test on a frame and a re-render of the affected range.

• "It flickers at the shoulder." The mask smoothing is retuned and re-validated.
• "Too dark now." A gentler curve, shown on a still first.
• "The edge of the sleeve is missed." Handled as its own area, the way the flap was.

How does Vizard Agent compare to doing it yourself?

By hand this is rotoscoping a garment across hundreds of individual frames, feathering the edge until it stops crawling from one to the next, and then rendering the whole clip at 4K just to find out whether the correction holds.

Common questions

Will it change my face? No. Vizard Agent builds the mask around the garment and leaves the person outside it.

Does it work at 4K? Yes. This run delivered at 2160x3840 with every frame processed at full resolution.

How long can the clip be? Short clips are the comfortable case. Every frame is processed, so cost scales with length.

Can it change the colour too? Yes. Say the shade and Vizard Agent tunes the curve towards it.

Is the audio affected? No. Vizard Agent remuxes the original audio into the processed video at full quality, so nothing about the sound changes.

What if the fix flickers? Tell Vizard Agent where. It re-tunes the mask smoothing and re-validates before re-rendering.

Does it darken the whole shot? No. The correction is confined to the mask, which is why Vizard Agent spends most of the job building and proving that mask.

How does it know which part is the garment? A clothing segmentation model does the first pass, and Vizard Agent refines the result with colour-based detection and edge smoothing.

Can it fix only part of the clip? Yes. Give the timecodes and Vizard Agent processes that range instead of the whole clip, which keeps a long video affordable.

How does it check the fix holds? It renders a short test section first and inspects consecutive frames, then validates the result at several points across the clip before the full pass.