How to put text behind yourself in a talking-head video

Last updated 2026-09-01 · 8 min read

What is the short version?

The effect is three layers: a background, some large type, and you cut out and placed on top. It reads as expensive because the type genuinely goes behind you rather than sitting over your face like a caption.

1. Go to Vizard Agent and upload the talking-head footage.
2. Say what the words behind you should be.
3. Say what the new background should look like.

What do you need before you start?

Footage where you are clearly separable from what is behind you. Vizard Agent keys you out of the shot, so an even background helps enormously — and a phone recording that was held sideways will need sorting out first, which it can do.

• The footage. Talking head, reasonably lit.
• The words. Short — two or three at most.
• The setting. What should replace the room.
• The typeface. Or a description of the weight.
• The format. Vertical, usually.

What do you type into Vizard Agent?

Describe the three layers. Vizard Agent handles the keying and the compositing, so what it needs is the copy, the setting and the mood — and it is worth saying the background must contain no people, because generated rooms tend to.

Prompt

Variants worth knowing:

• Say "behind me". Otherwise it becomes a caption.
• "Empty of any people." Generated rooms come with occupants.
• Two or three words. Long lines lose the effect.

What does Vizard Agent actually do?

Here is the order Vizard Agent worked in on a real talking-head video. Steps twelve to thirty are a diagnosis nobody plans for: the first keying pass came back empty, and the reason was that the phone had recorded the video sideways.

1. Inspects the video's properties and transcribes the speech.
2. Extracts frames into a preview grid and inspects the speaker's position and background.
3. Checks the background-removal and image-generation options.
4. Removes the background and generates a studio background image.
5. Regenerates the background with a negative prompt because the first one contained people.
6. Inspects the keyed video and finds no visible content in it.
7. Composites a single frame over a red background to see whether it is rotated.
8. Checks a frame from five seconds in to confirm the orientation.
9. Scans the keyed file for any frames with actual content and reads the processing log.
10. Extracts a raw frame without auto-rotation and confirms the video is stored sideways.
11. Transcodes a test clip rotated ninety degrees and verifies it is upright.
12. Transcodes the full video upright at 1080x1920 and re-uploads it.
13. Runs the background removal again on the upright version and confirms clean keying.
14. Loads the text-behind-subject technique and selects a premium sans-serif from the installed fonts.
15. Composites a test frame with the type behind the cutout, then scales everything to match.

Step ten is the lesson. A background-removal pass that returns nothing looks like a broken tool and is almost always a rotation problem — the file's stored orientation and its displayed orientation disagree, and Vizard Agent tested that rather than assuming the keying had failed.

What does the result look like?

The source this page is written from measured 1920x1080, HEVC, 30fps and ran 127.74 seconds, and it turned out to be stored sideways — so the whole file had to be transcoded upright to 1080x1920 before the background removal would return anything at all.

The finished shot has three layers: a generated room with nobody in it, large type across the middle, and the speaker keyed and placed on top so the letters pass behind their shoulders. Nothing about the speaker was altered.

When does this not work well?

The whole effect rests on getting a clean cutout of you, and cutouts have honest limits that no amount of processing removes. Vizard Agent shows you the keying composited over a flat test colour before it builds anything, rather than presenting a finished shot and hoping the edges hold up.

• Busy backgrounds. The key picks up edges it should not.
• Hair and glasses. The hardest edges to cut cleanly.
• Matching clothes. A dark top on a dark wall will not separate.
• Long words. They disappear behind the subject.
• Sideways files. They must be fixed before keying.

How do you fix a result that came back wrong?

Say which of the three layers is wrong — the background, the type or the cutout. Vizard Agent keeps the keyed video, the generated background and the type layer as separate files, so a new background or different wording re-composites without the keying needing to be run again.

• "The edge around my hair is rough." Key refined and re-composited.
• "There are people in the background." Regenerated with a negative prompt.
• "The words are hidden behind me." Repositioned or shortened.

How does Vizard Agent compare to doing it yourself?

By hand this means either a green screen at the shoot or rotoscoping afterwards, and then three separate layers assembled in a compositing tool. Vizard Agent keys from ordinary footage, and when the key comes back empty it goes and diagnoses the cause rather than simply running it again.

Common questions

Do I need a green screen? No. Vizard Agent keys from an ordinary room, though an even background helps.

How many words work? Two or three. Vizard Agent keeps them large enough to read around you.

Can I keep my real background? Yes. Vizard Agent can put the type behind you over your own room.

Why did my background removal come back empty? Usually rotation. Vizard Agent checks whether the file is stored sideways.

Will the background have people in it? Not if you say so. Vizard Agent generates it with a negative prompt.

Can it match my brand typeface? Yes. Vizard Agent selects from installed fonts or uses one you supply.

Does the text move? It can. Vizard Agent can animate it, though static usually reads as more premium.

What about my hair edges? The hardest part. Vizard Agent shows the key over a test colour so you can judge.

Can I use this on existing footage? Yes. Vizard Agent works from anything you have already recorded.

Does it change how I look? No. Vizard Agent cuts you out and moves you; it does not alter you.

Why does the type going behind me matter? Because a caption over your face reads as a caption, and type that passes behind your shoulder reads as a set someone built. It is the same words doing completely different work.

Can it do this for a whole series? Yes. Vizard Agent reuses the background and the type treatment across every episode.

Does it check the result? Yes. Vizard Agent composites test frames and inspects the keying before rendering.