The short answer
Put the green screen clip on the timeline, put the new background on the track above it, select the green screen clip, open its video settings, turn on the chroma key, and sample the screen colour from an evenly lit patch. Then clean the edge, remove the green spill from the subject, and export at the resolution you actually intend to publish. That is the whole job, and it takes about two minutes once you know which control does what.
The rest of this page is that sequence written out in full — first for desktop, then the differences on mobile — followed by the starting values for each slider, a symptom-to-fix cheat sheet you can download, and the arithmetic that explains why some footage keys cleanly and some never will. If you only read one section, read the steps; everything after them exists to help when a step does not come out right.
The desktop steps, in order
Step 1 — Create the project and import both clips. You need two things: the footage shot against the screen, and the background you want behind the subject. Import both before you touch the timeline, so you are not hunting through folders halfway through the edit.
Step 2 — Stack them in the right order. Drag the green screen clip onto the timeline first, then place the background clip on the track above it. CapCut composites from the bottom up, so the keyed clip has to be underneath the new background. If you do this the other way round, the background simply hides the subject and no key will fix it.
Step 3 — Select the green screen clip and open its video settings. The key is a per-clip effect, not a project-wide switch. Select the clip on the timeline, then open the video settings panel for that clip. If the panel you opened belongs to the background clip, you will not find a key at all — that is the most common reason people think the feature is missing.
Step 4 — Turn on the chroma key and sample the colour. Enable the key, then use the colour picker on the screen itself. Sample from an evenly lit area — the middle of the screen, at roughly the height of the subject's shoulders. Do not sample a corner, and do not sample a hotspot near a lamp. The picker sets the centre point of everything that follows; a bad sample cannot be corrected by any slider.
Step 5 — Raise the strength only until the screen clears. Start low and increase until the last patch of green disappears, then stop. If the subject's edges begin to go see-through, you have passed the point where the key is still describing the screen and started describing the subject. Back it off a few points.
Step 6 — Bring the edge back, then despill. Hair, fingers and thin edges usually end up too transparent at first. The shadow or clean-up control restores the semi-transparent pixels along the boundary. After the alpha looks right, use despill to pull the green that bounced off the screen out of the subject's skin and clothing. Despill changes colour only — it does not change what is transparent.
Step 7 — Export at final resolution and bitrate. Judge the result on an exported frame, not on the preview canvas. Previews render at reduced resolution and hide edge errors that reappear at full size. Choose the resolution and bitrate you will actually publish at, export, and check one frame from the middle of a movement.
The mobile steps, and where they differ
The order is the same on a phone; what changes is where the controls live and how much precision you get.
- Start a new project, then add the green screen clip to the timeline first and the background clip above it.
- Select the green screen clip — not the background — and open its settings.
- Find the key control in that clip's panel and turn it on. The naming of this control changes between versions and between the iOS and Android builds, so look for the key inside the selected clip rather than memorising a menu path.
- Sample the screen with the picker, then adjust strength and clean-up exactly as on desktop.
- Zoom into the subject's edge before you judge it. On a phone screen a two-pixel fringe is invisible; on a laptop it is obvious.
- Export at the highest resolution the app will give you for that project, then check the result on a larger screen.
Two honest differences: the mobile build gives you less control over edge softness, and phone footage is normally 4:2:0, which stores colour at a quarter of the resolution — so the boundary between green and hair has the least colour information in the whole frame. That is not a slider problem, and the arithmetic section below explains why.
Starting values for the four controls
Use these to land in the right neighbourhood on the first attempt instead of sweeping a slider from zero until something looks acceptable — which is usually too wide.
| Control | What it moves | Start at | Sign it is wrong |
|---|---|---|---|
| Key colour | The centre of the radius | sample the lit screen | A grey patch that survives every tolerance |
| Strength / similarity | The radius | 30-40 | Subject edges going see-through |
| Shadow / clean-up | How much of the boundary survives | 10 | Hair turning into a solid silhouette |
| Despill | Green left on the subject | 40-60 | Skin going grey or magenta |
| Feather / edge blur | Softness of the alpha edge | 2-4 px | A visible cut-out line at the shoulder |
One rule covers all four: change one control at a time, and judge on a frame where the subject is moving. A still frame lets you get away with a radius that falls apart a second later.
When a step goes wrong
Eight symptoms, what causes each one, which control to touch and a starting value. The same content is available as a CSV you can keep next to the timeline.
| Symptom | Cause | Fix |
|---|---|---|
| Green fringe around hair | Radius reaching into 4:2:0 colour blocks at the edge | Strength 30-40, then re-sample |
| Subject loses a shoulder or hand | Skin or clothing sits close to the screen hue | Strength 25, Shadow 10 |
| A dark corner never disappears | Unevenly lit screen, one sample cannot describe it | Second key pass or garbage matte |
| Green cast on skin | Light bouncing off the screen onto the subject | Despill 40-60 |
| Subject looks pasted on | Edge too hard, light direction mismatched | Feather 2-4 px |
| Screen turns grey instead of clear | Key colour sampled from a shadow or hotspot | Re-sample the lit area |
| Only half the clip keys | Clip already carries a grade or LUT | Flatten the grade first |
| Export looks softer than the preview | Preview renders at reduced resolution | Key at source size, then downscale |
Download the same table as CSV (8 rows)
The arithmetic: why some footage keys and some does not
The figures below are calculated from the formulas shown, using the stated bitrates. They are estimates for comparison, not measurements of any particular export from your machine.
Resolution is a pixel count, and the counts are not close. 720p is 1280×720 = 921,600 pixels. 1080p is 1920×1080 = 2,073,600 pixels. 4K UHD is 3840×2160 = 8,294,400 pixels. So 4K carries exactly four times the pixels of 1080p, and 1080p carries 2.25 times the pixels of 720p. The key is computed per pixel, so at 4K there are four times as many samples along the same edge. That is why keying 4K and then scaling down to 1080p gives a cleaner edge than keying 1080p directly: you are not adding information, you are letting the key use what the sensor captured before the resample throws it away.
Bitrate turns into file size with one division. Size in bytes = bitrate in bits per second × duration in seconds ÷ 8. Assuming 12 Mbps at 1080p, one minute is 12,000,000 × 60 ÷ 8 = 90 MB. Assuming 45 Mbps at 4K, one minute is 45,000,000 × 60 ÷ 8 = 337.5 MB — about 3.75 times the 1080p file for four times the pixels. A ten-minute 4K clip at that assumed bitrate works out to roughly 3.4 GB. Change the assumption and the numbers change proportionally; what does not change is the formula.
Colour is stored at a quarter of the resolution, and that is where edges die. Phone footage is normally 4:2:0: brightness at full resolution, colour at half width and half height. In a 1080p 4:2:0 frame there are 2,073,600 brightness samples but only 518,400 colour samples. The boundary between green and hair is exactly where colour detail is thinnest, which is why a slightly over-wide tolerance produces a ragged fringe instead of a clean edge, and why despill alone never removes it.
What the other pages on this query actually cover
On 18 October 2026 I opened the top ten organic results for this query and recorded which of five things each one provides. This is a survey of published pages, not a test of CapCut itself — it says nothing about whether CapCut works, only about what the ranking pages hand you.
| What it provides | Pages | Readable |
|---|---|---|
| Step-by-step instructions | 6 | 6 |
| A downloadable table | 0 | 6 |
| A worked unit conversion | 0 | 6 |
| Comparison data carrying a date | 0 | 6 |
| Explicit limits for phone-only editors | 3 | 6 |
Survey run 18 October 2026, Google, hl=en, gl=us, pws=0. Four of the ten results were video or social pages that could not be read without an account, so the denominators above are six readable pages, not ten. Those four are recorded as unread rather than as empty.
Where this workflow stops working
- Phone footage gives you the least to work with. 4:2:0 is the default, so edge colour is already thin before you touch a slider. If the shot matters, key it on a desktop.
- Uneven lighting cannot be fixed with a wider tolerance. If the screen is two different greens, one pass cannot describe both. Two zones and two passes beat any amount of slider nudging.
- Green clothing, plants and signage will fight the key. The key cannot tell screen from subject. Move the object, change the wardrobe, or rotoscope that shot.
- Motion blur is not recoverable. A fast-moving hand smears green into the edge, and those pixels genuinely are part green. Shoot the keyed take with a faster shutter or accept the softness.
- Long-form edits are not what the phone build is for. Short social clips key fine; a twenty-minute timeline with several keyed layers wants a desktop editor.
Questions that come up
Why does the key look perfect in the preview and bad after export?
Usually because the preview renders at a lower resolution than the export. Edge errors occupying one pixel at preview scale occupy several at full scale. Export a single frame at final settings and judge that.
Should I shoot 4K if I am publishing at 1080p?
For a keyed shot, yes. You get four times the pixels to compute the alpha on, and downscaling afterwards averages away the residual fringe. The cost is storage: roughly 337 MB per minute at an assumed 45 Mbps versus 90 MB at 12 Mbps.
How do I know if my footage is 4:2:0?
Check the clip properties in your editor or with a media info tool. If it came off a phone and you did not deliberately choose otherwise, assume 4:2:0. That single fact explains most edge problems people blame on the keyer.
Can I key without filming against a screen at all?
Yes, with an AI segmentation model rather than a colour key. It handles hair and motion better and uneven lighting not at all, because there is no screen to be uneven. For a still image, a browser-based chroma key tool is the faster route.
Related reading: green screen editor for unevenly lit footage, how to light a green screen, removing a green screen from video, blue screen versus green screen, and replacing a green screen with an image.