How to mirror a video without reversing the text on screen

Last updated 2026-08-30 · 7 min read

What is the short version?

Flipping a video horizontally is a single command that anything can do. Flipping it without turning every caption into mirror writing is an entirely different job, because the text has to be lifted off first, the picture flipped underneath, and the text laid back on top.

1. Go to Vizard Agent and upload the video.
2. Say to mirror it horizontally.
3. Say the on-screen text must stay readable.

What do you need before you start?

The video and nothing else. Vizard Agent locates the text itself, including stickers and symbols that are not really words at all, so there is no need to list where anything appears or to dig out the original caption graphics.

• The video. With its captions burnt in.
• The instruction. Mirror, except the text.
• Any exceptions. A logo that should also stay.
• The output shape. Usually unchanged.
• Symbols too. Arrows and stickers, if they matter.

What do you type into Vizard Agent?

Say what to flip and what to exempt in one sentence. Vizard Agent treats the exemption as the harder half of the job — finding and lifting the text — and stating it up front is what makes the whole approach different from a one-line flip.

Prompt

Variants worth knowing:

• Symbols exempt too. Question marks, arrows, subscribe buttons.
• A logo kept. Branding that must not read backwards.
• Only part of the video. If the flip applies to a section.

What does Vizard Agent actually do?

Here is the order Vizard Agent worked in on a real mirrored edit. The approach becomes clear around the twenty-first step: it built a transparent layer containing only the text characters, with none of the footage behind them.

1. Looks at the real frames and the video's parameters before flipping anything.
2. Checks the shot list to understand where the captions sit and how they move.
3. Works out exactly where the text and symbols appear so they are not flipped.
4. Reviews an extended storyboard of the whole clip before processing.
5. Examines the stickers and text close up to see what has to be preserved.
6. Checks the opening sticker frame, then the subscribe caption and the question mark.
7. Determines the exact boundaries where each text panel appears so nothing is left behind.
8. Isolates the text regions at source resolution and verifies each crop by eye, twice.
9. Builds a transparent layer with only the text characters, mirrors the footage, and composites the text back in its normal orientation.

Step 7 is the one that prevents a visible seam. If the boundary of a text panel is off by a few pixels, the mirrored footage shows through beside the restored caption, and that edge is far more obvious than mirrored writing would have been.

What does the result look like?

The finished file measures 360x640, H.264, 25fps, 52.44 seconds, AAC audio. The footage is flipped — subjects face the other way, the composition is reversed — while every caption, the subscribe prompt, the question mark and the stickers all read exactly as they did before.

The effect is that nothing looks edited. A viewer sees a normal video, which is the point: mirrored text is the one artefact that immediately announces a flipped clip.

When does this not work well?

Lifting text off a video means reconstructing whatever sat behind it, and that gets harder as either the text or the background starts moving. These are the cases where Vizard Agent runs into the limits of the footage itself.

• Moving text is harder. Captions that travel need tracking, not a fixed region.
• Semi-transparent captions blend. The background shows through the letters.
• Text over faces is awkward. The area behind it has detail to rebuild.
• Some things should flip. Text within the scene itself, on a sign, usually should.
• Mirroring changes meaning. Handedness, road direction and writing in shot all reverse.

How do you fix a result that came back wrong?

Point at the caption that is wrong and say what should have happened to it. Vizard Agent keeps all the located text boundaries, the isolated regions and the transparent layer, so one element can be re-cut or exempted without redoing the flip.

• "There is an edge around the caption." The boundary is re-measured and re-cut.
• "The sticker got flipped." Added to the exempt layer and recomposited.
• "That sign should have flipped." Removed from the exemptions.

How does Vizard Agent compare to doing it yourself?

By hand this means flipping the video, then rebuilding every caption from scratch on top of it — matching the original typeface, size, position and timing by eye. Most people either accept mirrored text or give up on the flip.

Common questions

Why would I mirror a video at all? Composition, platform reuse, or matching a shot to its neighbour. The reason does not change the method.

Will the captions be retyped? No. Vizard Agent reuses the original pixels, so the typeface and styling are exactly right.

Does it find stickers as well as words? Yes. Vizard Agent treated the stickers and the question mark as text alongside the captions here.

What about text on a sign in the scene? That usually should flip. Say so and Vizard Agent leaves it in the mirrored layer.

Will there be a visible edge? Not if the boundaries are right, which is why Vizard Agent measures them at full resolution.

Why lift the text off instead of flipping it back in place? Because flipping a region twice leaves the background inside that region mirrored the wrong way. Lifting the characters onto their own layer keeps the picture and the words independent.

Does the audio change? No. Vizard Agent mirrors the picture only.

Can it flip just part of the video? Yes. Say which section and the rest is left alone.

What if the captions move around? Harder, but possible. Vizard Agent tracks them rather than using a fixed region.

Can it mirror only the background? Yes, in principle. Vizard Agent separates layers to do this job anyway, so exempting a person rather than a caption is the same technique applied to a different part of the frame.

Will the mirrored footage lose quality? No. Flipping is a lossless rearrangement of the same pixels, so the only quality question in this job is how cleanly Vizard Agent lifts the text off and puts it back.

Does this work on a video I did not make? Technically yes, but the rights do not change. Mirroring somebody else's video is still using their video, and platforms that detect reuse are not fooled by a flip.

Does Vizard Agent check the result? Yes. It reviews the composited text against the mirrored footage before delivery.