A video background remover tried on four seconds of a serum ad
The templates above show what to type. This section shows one real run. We made a sample for this page from the first four seconds of a showcase film on FlowPix, the “A 15-second UGC talking ad for a serum” film, and ran it through the video background remover. The clip came back twice: as a transparent WebM, and as an MP4 with the person laid on a flat colour, efe7dc, with the original sound kept.
The sample is a talking-presenter clip; the tool assumes the subject is a person by default, and the request for the sample set it to a person. The brand in the film, LÛMA, is made up, like the serum.
Where the clip came from
The source film was made with Seedance 2.0 Mini. The request asked for a 15-second vertical (9:16) talking-to-camera ad, in English, as one single camera shot like a real UGC review. One presenter holds the serum bottle and speaks lines that the request gave word for word, in a natural and friendly tone, with no claims about results. The film's spend line lists a 15-second talking video on Seedance 2.0 Mini with a ready-made presenter.
The agent's reply says it shot a single take to camera with the bottle in hand and had the presenter say the lines exactly as written. It also offered captions for sound-off viewing, a 1:1 or 16:9 reshape, and a 6-second cut. None of those were taken up in the replay. The credits charged for the film itself were not recorded, so we give none here.
What went in and what came out
The sample is the first four seconds of that film: 720 by 1280 pixels, 24 frames per second, 96 frames, with sound. The call asked for a VP9 output, with the subject set to a person and edge refinement off. The only charge recorded for the sample is 48 credits, which is the 96 frames at half a credit each. A separate charge for the colour version is not recorded.
Real figures from the sample:
| Step | What went in | What came out |
|---|---|---|
| Source film | 15-second vertical request, Seedance 2.0 Mini, a ready-made presenter, exact spoken lines | One 15-second vertical take; credits not recorded |
| Slice for the sample | First 4 seconds of that film | 720 x 1280, 24 fps, 96 frames, with sound |
| Background removal | The slice, VP9 output, subject is a person, edge refinement off | A transparent WebM; 48 credits recorded for the sample (96 frames x 0.5) |
| On a colour | The same slice, colour efe7dc | An MP4 with the person on that colour and the original sound; separate charge not recorded |
What the request leaves for you to decide
The skill the agent follows has four settings, and the sample used them like this. Background: saying nothing, or “transparent”, gives the WebM with an alpha channel; a colour you name becomes six hex digits, so green is 00ff00, white ffffff and black 000000. Subject: it is taken to be a person unless you say it is a product, an object or an animal. Edges: refinement is off unless you ask for cleaner outlines or complain about the first result. Calls: the agent makes one call per request.
Refinement is billed at three quarters of a credit a frame, rounded up, against half a credit without it. On the 96 frames of this sample that arithmetic would give 72 credits instead of 48, though we did not run that version.
What the requests in these runs did, and a limit worth knowing
The practical lesson from the serum film is the one the request made: it named the format, the length, the language, the single shot, and gave the spoken words in quotes. For the cutout, the sample did the same in its own terms. It used a clip that was already a short, single-subject shot, said the subject was a person, and named the colour as a hex value rather than a word. A fixed camera and one presenter were part of the source request, though we cannot say how much that helped the cut.
One limit comes from how the agent works. It never sees the picture, so it does not judge how clean a cutout is. The skill tells it to say nothing about what the video shows or how good the cut is, and to offer edge refinement or a different colour once if you are unhappy. The judgement of the edges is yours.
Before you use the result
Play both files through. Look at the outline of the person against a background like the one you plan to use, since a transparent WebM plays over anything and the colour MP4 only over the colour you chose. Listen to the MP4, which is the version recorded as keeping the original sound; we did not record whether the transparent WebM carries audio, so check that file yourself if you need it.
Check the limits before you upload. The video can be up to 60 seconds and 60 frames a second; a longer one is refused before anything is charged, so trim it and upload again. Billing is by the frame, so a short clip costs less than a long one. The result is another video in your list, kept for 7 days like the original, with a download button on the page. Download it soon after it appears.

