How to empty a room behind someone in a video
Upload the video and tell Vizard Agent what should go and what must stay. Vizard Agent cuts the person out across the whole clip, checks the outline, hair, hands and feet survived, generates a clean empty version of the room from one of its own frames, and composites them back together.
What is the short version?
Emptying a room is not removing objects one at a time. It is easier to lift the person out, build the room they should be standing in, and put them back — which also means the furniture, the clutter and the door frame all go at once.
- Go to Vizard Agent and upload the video.
- Say what should go — the furniture, the fittings, the door.
- Say the person must stay exactly as they are.
What do you need before you start?
The video and a clear list. Vizard Agent generates the empty room from the footage itself, so no reference photograph is needed, but stating what stays matters as much as stating what goes because everything in shot is otherwise a candidate for removal.
- The video. As filmed, not re-exported.
- What goes. Furniture, fittings, clutter.
- What stays. The person, and anything else.
- The wall. Plain, or matching what is there.
- The lighting. Whether it should stay as filmed.
What do you type into Vizard Agent?
Say "keep the person exactly the same" in the same breath as the removal. Vizard Agent treats that as the fixed point of the whole job, and without it the safest interpretation is that everything in the frame is open to being rebuilt.
Prompt
Variants worth knowing:
- A plain wall. Rather than a rebuilt version of the same room.
- Some fittings kept. A window, a skirting board.
- The original audio. Untouched throughout.
What does Vizard Agent actually do?
Here is the order Vizard Agent worked in on a real room-emptying job. The order is the insight: it removes the person first and builds the empty room second, rather than trying to erase furniture from around a moving subject.
- Analyses the duration, resolution and representative frames.
- Reviews those frames to understand the scene and how the person moves.
- Checks which tools can edit the setting while keeping the person.
- Confirms the output format of the background removal so it can composite properly.
- Extracts a larger frame to build the clean wall from, and inspects the original framing to keep the floor, ceiling and lighting consistent.
- Cuts the person out of the entire video, preserving their original movement.
- Checks the cutout's format, alpha and synchronisation, then extracts frames from it.
- Confirms visually that the cutout kept the outline, the hair, the hands and the feet.
- Generates a clean plate of the room with a continuous wall and no furniture, reviews it, composites the person onto it with the original audio, then investigates the visual error in the first render and checks the cutout's transparency pixels.
Step 8 is where these jobs are won or lost. Hands and feet are the parts a cutout most often clips, and they are also the parts a viewer watches, so checking them specifically rather than glancing at the whole frame is the difference.
What does the result look like?
From the run this page is written from, probed on the delivered file: 1920x1080, H.264, 29.97fps, 44.57 seconds, AAC audio. Forty-five seconds of the same person, moving exactly as they did, standing in a room that now has a continuous plain wall, no furniture and no doorway — with the original audio untouched.
The person is not regenerated. That distinction matters: they are the same pixels from the same take, composited onto a new background, which is why they still look like themselves.
When does this not work well?
Separating a moving person from a room and then rebuilding everything that was behind them has real limits in both halves of the job, and they compound. These are the situations where Vizard Agent runs into the footage rather than the brief.
- Contact with objects is hard. Someone leaning on a desk cannot lose the desk cleanly.
- Shadows belong to the old room. They have to be removed or rebuilt too.
- Hair and hands are the weak points. Fine detail is where a cutout fails first.
- Lighting must agree. A person lit by a lamp that no longer exists looks wrong.
- Reflections give it away. Mirrors and windows show the room you removed.
How do you fix a result that came back wrong?
Point at the moment where it breaks down. Vizard Agent keeps the cutout, all the alpha checks, the generated plate and the composite tests, so the background or the separation can be redone without repeating the other half.
- "His hand disappears." The cutout is refined and re-checked at that frame.
- "You can see the old wall." The plate is regenerated and recomposited.
- "The shadow is still there." Removed from the plate or rebuilt to match.
How does Vizard Agent compare to doing it yourself?
By hand this means rotoscoping a moving person for forty-five seconds, then painting out furniture frame by frame, or building a set and reshooting. Both are days of work, which is why most people just tidy the room and film again.
| By hand | Vizard Agent | |
|---|---|---|
| The person | Rotoscope every frame | Cut out across the whole clip at once |
| The room | Paint out object by object | A clean plate generated from a frame |
| Checking | Watch and hope | Outline, hair, hands and feet verified |
| Failures | Start over | The render error diagnosed at the pixel level |
Common questions
Does the person change at all? No. Vizard Agent composites the original pixels rather than regenerating the person.
Can it remove everything at once? Yes, and Vizard Agent finds that easier. Building an empty room beats erasing objects one by one.
What about the door? Vizard Agent makes the doorway part of the wall if you ask it to.
Will the lighting still match? Vizard Agent checks the original framing to keep the floor, ceiling and lighting consistent.
Does the audio survive? Yes. Vizard Agent preserves the original track through the composite.
Why remove the person first? Because erasing furniture around a moving subject means solving a new problem on every frame, whereas lifting the person out once turns it into a single clean-room generation.
What if they touch the furniture? That is the hard case. Contact points betray the removal, and Vizard Agent flags them rather than hiding them.
Can it keep some of the room? Yes. Name what stays — a window, a skirting board — and it is preserved.
Does this work for a stationary subject? Yes, and more easily. Less movement means a simpler cutout.
Is this the same as replacing the background? Not quite. Vizard Agent rebuilds the room the person is actually standing in rather than dropping in an unrelated backdrop, which keeps the floor, the perspective and the light consistent with the take.
Does Vizard Agent check the result? Yes. It reviews the composite at several points across the clip before delivery.