How to replace the background behind you on camera

Last updated 2026-08-16 · 5 min read

What is the short version?

You do not need a studio and you do not need to key anything yourself. What you do need is to say what should be behind you, because "a better background" is not an instruction anyone can act on.

1. Go to Vizard Agent and upload the footage.
2. Describe the background: the room, the mood, the colour.
3. Say whether anything sits behind you — text, a logo, a lower third.

What do you need before you start?

Footage, and a description of what should replace what is behind you. There is no green screen to set up, which is the point — but keying works off the edge of your silhouette, so the things that decide the result are how you were lit and how much you move.

• Footage where you hold still-ish. Keying works from the edge of your silhouette. Fast movement blurs that edge, and blurred edges are where composites look fake.
• Even light on your face and shoulders. A backlit subject loses their outline.
• A background described in words. "A cosy studio, warm lamps, no people" beats "something professional". Say if it must be empty of people — that is a common correction.
• Where the text goes, if there is text. Behind your shoulder is a different job from over your chest, and Vizard Agent will place it where you say.

What do you type into Vizard Agent?

One instruction covers both halves of the job: what goes behind you, and what goes on top. Naming the second is what stops Vizard Agent placing a caption across your face, and it is the clause people leave out most often.

Prompt

Variants:

• Colour only. "Put me on a flat deep-green background." Fastest and least likely to look generated.
• Match a brand. Name the hex codes and say what must not appear. Vizard Agent honours negative instructions like "no red elements".
• Keep the room, change the mood. "Same room, evening light." This regrades instead of keying, and holds up better on moving footage.

What does Vizard Agent actually do?

Here is the order it worked in on a real piece of footage. Most of it is checking rather than doing — Vizard Agent looked at what it had produced, found a problem, and went back. This is one run, not a typical one; the shape holds, the effort varies:

1. Probes the video for resolution, duration and audio format.
2. Transcribes the speech so any text it places can follow what you are saying.
3. Extracts frames into a grid and looks at it to find where you sit in frame and what is behind you.
4. Reads its own tool options for background removal and image generation before choosing.
5. Keys the footage and generates the background in parallel, then opens the generated image to check it.
6. Regenerates the background with a negative prompt because the first one had people in it.
7. Composites, then tests the composite by putting a single frame over a flat red plate and looking at it.

Step 7 is the one to notice. Compositing over a colour nothing in the scene contains is how you see a bad edge immediately.

What does the result look like?

A finished video at the aspect ratio you asked for, with your original audio intact. The delivered file from this run is H.264 with the speech untouched — background replacement changes only the picture, and Vizard Agent keeps both the keyed footage and the generated background in the project.

Expect this category to take longer than a straight cut, and note that it was not separately measured. Comparable work runs a median of 28 to 38 minutes end to end, and across all projects the median cost by tier is Flash 47, Pro 55, Max 242, Ultra 263 credits.

When does this not work well?

Everything here is decided by the edge of your silhouette — that is the only thing a key has to work with. Vizard Agent will composite whatever it manages to separate, so the failures below are all cases where the separation itself is unreliable, and most were fixed at the camera rather than afterwards.

• Footage filmed sideways on a phone is the reliable failure. Phones record landscape and store an "actually rotate this" flag alongside it. Background removal reads the stored picture and ignores the flag, so it keys a sideways person and returns what looks like an empty file. In the run behind this page, Vizard Agent found this by scanning for frames with any content at all, then re-encoded the source upright before trying again. If your clip is from a phone and the result comes back empty or rotated, say so — the fix is one instruction, not a re-shoot.
• Fast movement and motion blur. Hands crossing your face, quick turns, hair moving fast. The edge fails there first.
• Fine detail at the silhouette. Curly hair, glasses frames, a thin microphone. Some of it will be cut off.
• It cannot relight you. If you were lit by a window and the new background is a night scene, you will look pasted on. Ask for a background whose light matches your footage, or ask for a regrade too.
• Text burned into the original footage stays. Replacing the background does not remove a caption that was already baked into the picture.

How do you fix a result that came back wrong?

Reply in the same Vizard Agent conversation instead of starting over. The footage, the key and the generated background all stay in the project, so a new background is a regeneration rather than a fresh start — and the keying, which is the expensive half, is not redone.

How does Vizard Agent compare to doing it yourself?

By hand this is a green screen, a light on it, and a keyer in an editor — plus a background somebody has to source or build. Without the green screen it becomes rotoscoping, which is frame-by-frame work and the reason nobody does it for a routine talking-head video.

Neither replaces the other. Vizard Agent wins on ordinary footage shot in an ordinary room; a proper key on a lit green screen still wins when the edge has to survive close inspection.

Common questions

Do I need a green screen?

No. Vizard Agent separates you from an ordinary room, which is the whole point of doing it this way. A green screen still gives a cleaner edge, so use one if you already have one set up.

Can the text sit behind me rather than over me?

Yes, and it is worth asking Vizard Agent for by name. Text behind the subject is what makes a shot read as a set rather than as an overlay, and it is the single change that most makes these look deliberate.

Will my audio change?

No. This is a picture-only operation, and Vizard Agent leaves the speech untouched. If you also want the audio cleaned up, ask for it explicitly — it is a separate thing rather than something bundled in.

Can I use my own photo as the background?

Yes. Upload it with the footage and tell Vizard Agent to use that image. This is also the most reliable route when the background has to be a real place, because nothing is being invented.

Does it work if I move around a lot?

Less well. Vizard Agent keys off your silhouette, and motion blur softens that edge — so a piece to camera holds up far better than a walking shot, and asking for a flat colour background hides more of the damage than a detailed scene will.