A small animal sits in frame, then it's gone, and the ground underneath keeps tracking naturally with the camera instead of smearing into a patch. That's the clip above. Removing a person from a CapCut video is the same job with a harder subject: bigger than the animal in the clip, usually walking instead of sitting still, and dragging a shadow along wherever it goes.
What CapCut actually gives you
CapCut's removal and background tools aren't the same tool everywhere. The phone app, the desktop app and the web version ship different feature sets, and what's there changes between updates. Rather than follow a menu path written for a different copy, open yours, look for an object or background removal tool, and test it on a short clip of the person before you commit a full project to it.
Why a walking person is the hard case
Whichever removal tool you find, it's almost certainly built to fill a gap from nearby pixels in a short run of frames, which works fine on something small and still. A person changes that math:
- They're bigger, so more of the frame has to be rebuilt behind them on every frame they cover.
- They move, changing shape with each step instead of sitting in one pose.
- They cast a shadow, which a removal that only masks the body leaves sitting on the ground after the person is gone, usually the first thing anyone notices.
- They can cross paths with someone else, which is where a quick tool's tracking is most likely to jump to the wrong subject.
A one-tap removal built for a dropped phone or a stray object will show a smear, a flicker, or a leftover shadow the moment any of those apply.
The AI pass for a moving person
1. Mask on the clearest frame, then track the whole clip
Start where the person is fully visible and sharp, not mid-stride or half out of frame, then track forward and backward through the shot. Scrub the result and fix the mask only where it drifts, usually at a turn, a partial block, or a crossing with someone else.
2. Grow the mask for the shadow and any reflection
Catching the body but missing the shadow or a reflection in glass or a puddle is the most common tell. Extend the mask to cover both before the removal runs.
3. Rebuild the background across the whole shot, not frame by frame
Run the removal on the tracked region so what's behind the person is rebuilt over time and moves with the camera, instead of sitting on the image like a sticker.
4. Round-trip back if your project lives in CapCut
If you're finishing in CapCut, the result drops back onto your timeline there. If you want to finish somewhere else, export as EDL, FCPXML or OpenTimelineIO instead of a flattened clip.
Asking an agent to do the AI pass
Remove the person walking through this shot. Track them through the whole clip, including the shadow they cast, and keep the background moving naturally with the camera. Show me the result before replacing the original.
With the run mode set to review, the agent masks and removes the person and waits for your approval before touching the original clip. You decide if it's clean enough to ship.
When CapCut's own tool is enough
If the person barely moves, the camera is locked off, and the clip is only a couple of seconds, whatever removal tool your version of CapCut has will probably get there faster than setting up a separate pass. Reach for the tracked-mask route once they're walking, turning, or getting crossed by someone else in frame.
Download the free Shoty editor for the tracked-mask route. For the same job on an object instead of a person, see remove object from video in CapCut, and for the masking step on its own, AI rotoscoping.