How to change specific text in a video while keeping the animation

Last updated 2026-08-18 · 6 min read

What is the short version?

A finished video with one wrong word in it is a common and infuriating problem. The word is usually a name, and the name is usually rendered into the picture, which means it is not text any more. It is pixels that happen to look like letters.

1. Go to Vizard Agent and upload the video.
2. Say the exact old wording and the exact new wording.
3. Say that the animation and everything else must stay as it is.

What do you need before you start?

The video and both spellings. Vizard Agent finds the text in the picture itself and works out its size and position, so what only you can supply is the exact new wording, including its capitalisation, since it will be rendered exactly as you write it.

• The video. The finished one, at the best quality you have.
• The old text, exactly. So Vizard Agent can find the right instance of it.
• The new text, exactly. Capitalisation included. It goes in as written.
• Whether the text moves. If the camera or the object moves, say so.
• Where it appears. "On the phone screen", "in the lower third". It narrows the search.

What do you type into Vizard Agent?

Quote both words exactly as they should read. Vizard Agent locates the text visually rather than needing coordinates from you, and giving it the precise replacement wording is what lets it match the letter spacing and the size rather than approximating them.

Prompt

Variants worth knowing:

• A longer replacement. Say so; a word that is wider than the original needs the spacing rebuilt.
• Several instances. If the word appears more than once, say whether all of them change.
• A different language. Replacing with a different script needs a font that covers it.

What does Vizard Agent actually do?

Here is the order Vizard Agent worked in on a real text replacement. What stands out is how much of it is measurement: it does not eyeball the position, it draws a grid and reads the numbers off it.

1. Extracts frames and looks for the text in the picture.
2. Pulls a full-resolution frame and zooms into the screen the text sits on.
3. Probes the video's properties and downloads the source locally.
4. Analyses the frame-to-frame movement to see whether anything shifts.
5. Checks for scene changes across the clip.
6. Draws a coordinate grid over the text area and reads the exact position off it.
7. Draws a second grid over the whole object to place it in context.
8. Searches the installed fonts and renders a comparison to match the original type.
9. Measures the letter size and bounding box, tests the replacement on one frame, then compares four different replacement methods before committing to one.

Steps 6 and 9 are the craft. A coordinate grid turns "roughly there" into a number, and testing four methods on one frame costs almost nothing compared to rendering the whole video the wrong way.

What does the result look like?

From the run this page is written from, probed on the delivered file: 576x1024, H.264, 30fps, 24.05 seconds, AAC audio. The same dimensions and the same length as the original, with one word different and the animation around it untouched.

Vizard Agent replaced only the region the text occupies, so the movement, the lighting and the rest of the frame are still the original pixels rather than a re-render of the whole video.

Twenty-four seconds of video with one word changed is a job whose whole value is in what did not change. Vizard Agent treats the original as the thing being preserved and the replacement as the smallest possible intervention in it.

When does this not work well?

Text baked into a picture has no underlying editable layer, so every replacement is a reconstruction. Vizard Agent measures carefully and matches the type as closely as the installed fonts allow, and a close match is not the same as the original file.

• The font may not exist on the system. Vizard Agent renders a comparison and picks the closest, and closest is sometimes visibly not identical.
• A longer word breaks the layout. Replacing a four-letter name with an eight-letter one changes the spacing of everything around it.
• Moving text is much harder. Text on a static screen is reliable; text tracking with a handheld camera is not.
• Textured or angled surfaces betray the patch. Text on a curved or lit surface has to be relit to match.
• The original project file beats all of this. If you still have the project that produced the video, changing the word there is always cleaner than any reconstruction.
• Low resolution limits the match. A small or heavily compressed video gives Vizard Agent less to match the letterforms against.

How do you fix a result that came back wrong?

Say what looks off. Vizard Agent keeps the measured coordinates, the font comparison and each of the tested methods, so trying a different typeface or nudging the position works from the measurements rather than starting the analysis again.

• "The font is not quite right." Another candidate from the same comparison.
• "It sits slightly too high." A position change against the measured grid.
• "The edges look patched." A different replacement method from the four already tested.

How does Vizard Agent compare to doing it yourself?

By hand this is masking out the old word, matching a font by eye, matching the size by eye, matching the colour by eye, and then discovering the whole patch drifts two seconds later because the object it sits on moves.

Common questions

Does it work if the text is part of the picture? Yes, that is the case this is for. Vizard Agent reconstructs the region rather than editing a text layer.

Can it match the original font? It renders a comparison of the installed fonts and picks the closest. An unusual typeface may not have a match.

What if the new word is longer? Say so. Vizard Agent rebuilds the spacing, and a much longer word changes the layout around it.

Can it change the same word everywhere it appears? Yes. Say whether you want every instance or just one.

Will the video be re-encoded? Only the region that changes. The rest of the frame keeps the original pixels rather than being rebuilt by Vizard Agent.

Can it change text in more than one place? Yes. Name each one and Vizard Agent measures them separately before replacing.

What if the text moves in the shot? Vizard Agent checks the movement first. Static text is reliable; tracked text is harder and worth previewing.