VIDEO WORKFLOW / RECORDED TESTS
Remove green screen from video
Upload your video, let the tool pick the screen color, check the preview, then export a WebM with a clear background. No account or file upload is needed. If the backdrop is not detected, click Pick and sample it yourself.
Best for talking-head clips, product demos and people against an evenly lit, solid-color backdrop. Our recorded 1080p export kept 1920 × 1080 and 9.426083 seconds. Hair edges can retain a slight green fringe; very fine individual strands are not preserved separately.
Remove the backdrop and check the result
- Choose the original video you want to use. The browser must be able to decode the MP4, MOV or WebM file. Prefer the website’s HD download over its small player preview.
- Check the automatically selected screen color. Sampling happens once when the file loads. If it cannot find a clear screen, use Pick on an uncovered part of the backdrop, away from clothing and skin.
- Preview against light and dark backgrounds. Check the face, arms, clothing and outline. Use Original and Matte to inspect the key. Refine tolerance, edge softness and spill cleanup as needed; those changes do not trigger another automatic color sample.
- Export and inspect the downloaded file. Choose Original size to retain the source resolution within the tool’s supported size range. Video exports are silent. Use a player or editor that supports WebM alpha; a black player background does not by itself mean the key failed.
If the entire green backdrop remains as a faint film
Check the selected color before adding edge softness. A green preset is not a measurement of your actual screen. In our real 1080p clip, changing only the selected color cleared the sampled backdrop points while leaving the fully opaque total unchanged.
| 3.3-second PNG frame | Default #00B140 | Auto #4EA955 |
|---|---|---|
| Four corner alpha values | 59 / 59 / 59 / 59 | 0 / 0 / 0 / 0 |
| Background-point alpha | 59 | 0 |
| Fully opaque pixels | 450,313 | 450,313 |
| Intermediate alpha | 78.28% | 2.17% |
| Fully clear pixels | 0 | 1,578,281 |
That reverse check supports clearing this background film. Equal opaque totals do not prove every pixel stayed in the same place, so we inspected the sampled pictures too. PNG counts are kept separate from decoded WebM counts.
How close was the automatic color?
The automatic selection was #4EA955 (78, 169, 85). A separate border-sampling diagnostic recorded #4BAA55 (75, 170, 85). For this green key, the channel-excess measure is G − max(R, B): the selected color gives 169 − 85 = 84, and the diagnostic gives 170 − 85 = 85, a difference of 1. This is a difference in that measure, not a one-unit RGB distance or a ground-truth color calibration.
Do not copy this clip’s hex color onto your own footage. Let the tool sample your backdrop or use Pick.
What happened to a real 1080p video?
We used the website-provided MP4 from Mixkit’s presenter clip, under its Stock Video Free License. It was downloaded directly, not upscaled or transcoded before testing. Settings were tolerance 28, softness 12, spill cleanup 65, Original size and 30 fps. Automatic color: #4EA955.
Input and export were both 1920 × 1080, 9.426083 seconds. The WebM export was 18,454,755 bytes.
| Time (s) | Alpha = 0 | Intermediate alpha | Opaque pixels |
|---|---|---|---|
| 1.41 | 77.30% | 2.71% | 414,453 |
| 3.30 | 76.22% | 2.63% | 438,718 |
| 5.18 | 77.81% | 2.91% | 399,907 |
| 7.07 | 77.37% | 1.80% | 431,885 |
| 8.48 | 77.30% | 1.94% | 430,435 |
Alpha 0 is fully clear, 255 fully opaque; values between them are intermediate. The subject moves between these times, so different opaque totals across frames do not by themselves indicate lost foreground. This table is not a paired before-and-after test.
We reviewed all sampled head crops. There were no obvious large holes in the visible face, main hair shape or arm; a fine green fringe remained. The separate same-frame PNG comparison above is the reverse check, with 450,313 opaque pixels on each side. It does not establish fine-strand preservation.
Downloaded footage looks rough: is the source good enough?
Check the actual downloaded dimensions, not the website’s HD label or the player’s size on screen. Our early Hair Detail download from Hollywood Camera Work was a compressed 352 × 196 player preview lasting 2.669333 seconds, not the original HD plate. Source credit: Hollywood Camera Work.
We excluded it from fine-hair conclusions. Enlarging that preview did not restore missing detail, and uneven dark corners made it unsuitable for judging a clean backdrop. We kept its measurements in the record below, but stopped using them as a quality score.
- Choose the actual HD download when available; do not upscale a small preview and call it a higher-resolution source.
- Inspect the unprocessed file at its native size. If fine edges are already blended into the screen or blocky, an enlarged output cannot demonstrate recovery of the missing detail.
- Look for an evenly lit backdrop. Dark corners, reflections and compression artifacts can remain visible after the main screen clears.
- Check the person in the result, not just the amount of empty background. The supported use here is the main subject in a solid-screen video, not individually retained fine strands.
Questions while removing green screen from video
What if no screen color is detected?
The status reads: “No clear screen color found. Click Pick and sample the backdrop manually.” The original stays visible and export remains disabled until you choose a color. A clear patch around the frame edges helps automatic detection; tight crops or an obscured backdrop may need Pick.
Will changing settings replace my picked color?
No. Automatic sampling runs on a new file load. Our functional check kept a manually selected #347A42 after parameter changes and Reset. Pick remains available whenever you want to sample again.
Does a higher clear-pixel percentage mean a cleaner result?
Not on its own. It can mean more backdrop removed or more subject removed. In our same-frame pair, intermediate alpha fell from 78.28% to 2.17%, but the useful reverse evidence was the background-point alpha changing from 59 to 0 with the opaque total unchanged at 450,313. Inspect the outline as well.
Why does the exported video look black behind the person?
Some players do not display WebM alpha. Open it in a supporting browser or editor, or use a PNG sequence when your editing workflow cannot read the video’s alpha channel. Video and frame-sequence exports are silent.
Will every hair strand survive?
No. Hair edges may retain a slight green fringe, and very fine individual strands are not preserved separately. Use the preview to check the main outline before exporting.
The full test record behind this workflow
The earlier measurements below remain available so successful samples do not erase unsuccessful tests. None of the following percentages is a standalone quality score.
Earlier 720p background-clearing run
| Time (s) | Before: clear pixels | After: clear pixels | Before: intermediate | After: intermediate |
|---|---|---|---|---|
| 1.41 | 0.00% | 76.71% | 80.38% | 3.78% |
| 3.30 | 0.00% | 75.57% | 79.24% | 3.82% |
| 5.18 | 0.00% | 77.15% | 81.15% | 4.20% |
| 7.07 | 0.00% | 76.83% | 79.63% | 2.94% |
| 8.48 | 0.00% | 76.85% | 79.59% | 2.86% |
In a separate 3.3-second PNG probe at #50AA56, the opaque count was 197,295 and all sampled corner alpha values were 0. A sampled #56B058 probe retained the same 197,295 opaque total but corner alpha was 4. These probes and the exported-video measurements use different output paths and are not mixed.
Small website previews, retained but excluded from fine-hair judgment
Show the five Hair Detail preview samples
| Time (s) | Before: clear | After: clear | Before: intermediate | After: intermediate |
|---|---|---|---|---|
| 0.40 | 41.74% | 40.31% | 16.67% | 17.02% |
| 0.93 | 41.97% | 40.66% | 14.17% | 15.00% |
| 1.47 | 42.04% | 40.80% | 15.60% | 15.73% |
| 2.00 | 41.70% | 40.79% | 18.00% | 18.09% |
| 2.40 | 42.06% | 41.03% | 15.83% | 15.91% |
Show the five additional fabric-preview samples
| Time (s) | Before: clear | After: clear | Before: intermediate | After: intermediate |
|---|---|---|---|---|
| 1.75 | 25.79% | 25.25% | 37.58% | 37.96% |
| 4.08 | 28.70% | 28.92% | 36.81% | 36.94% |
| 6.40 | 39.46% | 38.50% | 27.99% | 28.93% |
| 8.73 | 34.20% | 32.71% | 34.14% | 35.56% |
| 10.48 | 37.01% | 34.41% | 32.47% | 35.03% |
Historical observation: visible layers and folds remained, with a missing segment at the thinnest outer edge. This recorded lower-bound result is not a separate product capability or a guarantee for other materials.
Early implementation fixture
The first algorithm checks used a constructed flower-and-stem fixture, not a human subject. They checked connected shapes and export consistency, and cannot establish human hair quality.
Show all early trial percentages and export sizes
| Time (s) | Baseline | Trial 1 | Trial 2 | Trial 3 |
|---|---|---|---|---|
| 0.36 | 3.69% | 5.72% | 6.23% | 6.23% |
| 0.85 | 1.44% | 3.63% | 4.21% | 4.21% |
| 1.33 | 1.59% | 3.63% | 4.27% | 4.27% |
| 1.81 | 1.51% | 3.66% | 4.17% | 4.17% |
| 2.17 | 1.32% | 3.51% | 4.13% | 4.13% |
| Run | Size | Duration (s) | Export bytes |
|---|---|---|---|
| baseline | 640 × 480 | 2.416666 | 622,623 |
| round1 | 640 × 480 | 2.416666 | 779,474 |
| round2 | 640 × 480 | 2.416666 | 790,660 |
| round3 | 640 × 480 | 2.416666 | 786,214 |
Automatic-color fallback checks
Constructed detector fixtures checked control behavior, not video quality. A grey backdrop with a green center item triggered the manual message; parameter changes preserved it, and export stayed disabled. A green clothing fixture at the bottom edge also did not count as a detected screen. Manual Pick then enabled export. Those fixtures do not guarantee correct classification in every composition.
All displayed measurements come from our saved browser runs. Download the complete recorded counts, input identifiers, settings and code hashes. Percentage formula: count / (width × height) × 100, rounded to two decimals. No new video tests were run for this guide.