Vizard Agent

How to change the label on a product in a video you already have

Last updated 2026-09-04 · 7 min read

Tell Vizard Agent what the label says now and what it should say. It finds every shot the product appears in, maps the coordinates on the jar, tracks the label as it moves, flattens it to measure exactly how much space the old text occupied, and repaints the new name with shading that matches the curve.

What is the short version?

A label is not text on a screen. It is text printed on a curved surface, lit from one side, moving through the shot — and swapping it convincingly is a tracking problem before it is a typography one.

  1. Go to Vizard Agent with the footage.
  2. Say what the label says now and what it should say.
  3. Ask it to check the wide shots too, where the old text is small.

What do you need before you start?

The footage and the exact wording, spelled the way it is printed. Where you still have the real artwork for the new label, send that too — matching a typeface from a photograph of a bottle is something Vizard Agent can do, and matching the actual design file is simply better.

What do you type into Vizard Agent?

Say both names and ask about every appearance. The failure mode here is not the hero shot — it is the two seconds in the background of a wide shot where the old name is small but still legible if somebody pauses.

Prompt

The product is called [new name] now, not [old name]. Can you edit the label in this video to say the new name? Check every shot it appears in, including the ones where it's small.

Variants worth knowing:

What does Vizard Agent actually do?

Here is the order Vizard Agent worked in on a real product video whose label had to carry a different brand name. Steps six and seven are the pair that make the result sit on the object rather than float over it.

  1. Scans the product clips for the shots the item appears in.
  2. Zooms in on the product shots and picks the exact spans.
  3. Reads the labels as they currently are.
  4. Extracts the label frames across the whole clip.
  5. Checks label visibility across both halves of the video.
  6. Maps the coordinates on the jar and on the label itself.
  7. Tracks the label and swaps the text, then checks frame one.
  8. Runs a hardened tracker across both product shots.
  9. Tiles the label checks and reviews the swap across the shot.
  10. Rectifies the label — flattens the curve — to measure the text area.
  11. Measures the old text's extent so the new name fits the same space.
  12. Repaints the label with matched shading, and reruns it.
  13. Checks the label in the finished video, in context.
  14. Checks whether the old text is readable anywhere else.
  15. Patches the label in the tail shot it found that way.
  16. Verifies the label holds up in motion, not just on stills.

Step ten is the one worth understanding. A label wrapped round a jar is distorted by the curve, so measuring the old text on the curved image gives you the wrong width. Flattening it first — then measuring, then painting, then re-curving — is why the new name occupies the same visual space as the old one.

Step fourteen is the check that saves you. Having fixed the hero shots, it went looking for the old name elsewhere in the video and found it in the tail, small but legible.

What does the result look like?

A video where the product carries the new name, on the curve, with the same lighting falling across it, holding steady while the jar moves through the shot — and no surviving instance of the old name anywhere else in the cut.

Nothing else about the footage changes. The presenter, the grade, the cut and the audio are the ones you already had; only the printed text on one object is different.

When does this not work well?

Repainting a label only works when there is enough of a label there to repaint. Where the text is too small in frame, too motion-blurred or too briefly visible for Vizard Agent to resolve it, the honest options are to hide that shot or to re-film the product.

How do you fix a result that came back wrong?

Point at the specific shot rather than at the video. Vizard Agent keeps the tracking data, the rectified flat label and the measured text extent from the first pass, so correcting one shot re-runs the repaint only there without re-tracking the whole video again.

How does Vizard Agent compare to doing it yourself?

By hand this is a rotoscoping job: track the jar, corner-pin a replacement, match the grain and the light, and do it again for every shot. It is entirely doable and it is a day's work for a competent compositor.

By hand Vizard Agent
Tracking the label Manual, per shot Tracked, then hardened where it slips
Sizing the new text Eyeballed on the curve Measured on the flattened label
Lighting Matched by eye Shading matched from the frames
Finding every instance You remember The whole video re-checked
A second shot The same work again The same treatment applied

Common questions

Can it match my real label artwork? Yes, and it is better. Send the file and Vizard Agent uses it.

What if the jar rotates? Vizard Agent tracks it and re-curves the text as it turns.

Will it look painted on? Not once Vizard Agent matches the shading — that is what step twelve is for.

Can it change a logo as well as text? Yes. Vizard Agent handles it the same way, provided it is legible enough.

What about the small background shots? Vizard Agent checks them and tells you which cannot be fixed.

Is hiding a shot an option? Yes, and sometimes the right one. Vizard Agent will suggest it.

Does anything else in the video change? No. Vizard Agent touches only the label pixels.

Can it do several products at once? Yes. Vizard Agent gives each one its own tracking pass.

What if hands cover part of it? Vizard Agent treats the occluded frames separately.

Will it hold up at full screen? Vizard Agent verifies in motion, which is a harder test than a still.

Can it change the label colour too? Yes, though a full redesign is better done on the artwork.

How long does it take? Longer than a text overlay and much less than a reshoot.

Should I just reshoot? If you can, often yes. This is for when the product no longer exists in that form.

Why flatten the label first? Because a curve makes the old text look narrower than it is, and text sized against the curved image comes back the wrong width.