Vizard Agent

How to remove a sweat stain from a shirt in a video

Last updated 2026-09-22 · 8 min read

Ask Vizard Agent to remove only the stain and name anything on the shirt that must survive. A stain is a patch of darker fabric that moves and stretches with the cloth, so the fix is a measured brightness correction inside a tight mask rather than a patch pasted over it.

What is the short version?

The take is good and there is a sweat mark on the shirt. Replacing that area of fabric always looks like a patch, whereas lifting the darkness back to the tone of the cloth around it looks like the shirt — which is why Vizard Agent measures before it corrects.

  1. Tell Vizard Agent to remove the stain and nothing else.
  2. Name what is on the shirt — a logo, a print, a seam.
  3. Ask to see the correction on one frame before the whole clip.

What do you need before you start?

The clip and a look at what the shirt is doing. Vizard Agent measures the fabric itself, but plain cotton, a printed tee and a striped shirt are three different jobs, and the striped one is the one that will not work.

What do you type into Vizard Agent?

Keep the scope tight in the words you use. Asking Vizard Agent to clean up the shirt invites a general tidy that flattens the fabric texture across the whole garment, whereas naming the stain itself keeps the correction inside the area that is actually wrong.

Prompt

Just remove the wet patch on the front of his shirt in this clip. Do not touch the logo, keep the fabric texture and the folds, and leave the rest of the shirt exactly as it is. Show me one corrected frame before you render the whole thing.

Variants worth knowing:

What does Vizard Agent actually do?

Here is the Vizard Agent sequence on a clip with a sweat mark on a printed shirt. Six different approaches get tried on single frames and compared against each other before anything at all is applied to the full video.

  1. Samples frames and looks at the stain's shape and movement.
  2. Inspects a full-resolution close-up of the shirt across the clip.
  3. Builds a reference frame and a candidate mask for the stain.
  4. Previews a clean-shirt reconstruction and checks the logo stays intact.
  5. Tests a tone-preserving correction that removes only the darkening.
  6. Tightens the mask so it stays inside the shirt and stops picking up the edge.
  7. Measures the shirt's brightness inside and around the stain to tune the strength.
  8. Tries matching the stained area to the clean fabric immediately beside it.
  9. Quantifies the remaining contrast in the centre and adjusts the target.
  10. Tests a texture patch as an alternative, and checks it for seams and folds.
  11. Tries a tight content-aware fill pulling texture from the shirt's own boundary.
  12. Tunes the final correction, confirms it on a frame, then applies it across the clip with the audio preserved.
  13. Reviews the render at several moments for seams and motion consistency.

Step seven is what separates this from a guess. The correction's strength is set from the measured difference between the stained fabric and the fabric beside it, so it lands on the right tone rather than the tone that looked right on one frame.

Steps four to eleven are worth reading as a set. Each approach is tested on a single frame and rejected for a specific reason — the reconstruction loses the print, the patch shows a seam at the fold — which costs seconds rather than a render each.

Step six is the classic failure mode. A mask that spills past the shirt edge corrects a strip of the background too, and that shows as a bright halo moving around the person's body.

What does the result look like?

The same take with a shirt that looks dry: the fabric keeps its texture and its folds, the print or logo is untouched, there is no rectangle or smudge travelling with the cloth, and the audio is the original.

When does this not work well?

Some stains cannot be corrected convincingly. A pattern running through the stained area, a very dark or strongly coloured mark, or fabric that changes shape dramatically from frame to frame all leave Vizard Agent with a correction that reads as a patch rather than as cloth.

How do you fix a result that came back wrong?

Say whether you can still see the stain or whether you can see the repair instead. Those two complaints need opposite adjustments, and Vizard Agent moves the correction strength or the mask boundary accordingly rather than starting the whole job again.

How does Vizard Agent compare to doing it yourself?

By hand you would mask the stain and lift the brightness, which works on the frame you are looking at and drifts as the shirt moves. The alternative is a clone patch, which is stable and shows a seam the moment the fabric folds.

By hand Vizard Agent
Approach One method, applied Six tested on single frames
Strength Set by eye Measured against the fabric beside it
The mask Drawn generously Tightened to stay inside the shirt
Prints and logos Corrected along with it Named and protected
Checking On playback Frames sampled across the clip

Common questions

Will the fabric still look like fabric? Yes, if the correction is tone-based — Vizard Agent tests that first.

Can it handle a patterned shirt? Sometimes. Vizard Agent will tell you when the pattern cannot be rebuilt.

What about stains under the arms? Vizard Agent treats them the same way, and plainer fabric makes it easier.

Does it work on a moving person? Yes. Vizard Agent tracks the area rather than fixing one region of the frame.

Can it protect a printed logo? Yes. Name it and Vizard Agent keeps the mask off it.

Will the audio change? No. This is a picture correction and the original track is kept.

Can it fix several stains at once? Yes, each with its own mask and strength.

What if the shirt is white? Easier for Vizard Agent, because there is less colour to match.

How long does it take? Most of Vizard Agent's time goes on the tests rather than the render.

Can I see the options? Yes. Ask Vizard Agent to show you the candidate corrections side by side.

Will it look edited at full resolution? That is exactly what the frame checks are for.

Can it do the same on a photo? Yes, and it is much simpler with one frame.

What if the stain spreads during the clip? Vizard Agent tracks its shape rather than assuming it stays the same.

Is a reshoot ever better? If the mark crosses a pattern or a seam, Vizard Agent will say so.