How to remove an object from video — locally, not in the cloud

The short answer

To remove an object from video, mark it and let an inpainting model rebuild what it covered from nearby frames. Use dynamic tracking when the object moves and static removal when it stays put. Results depend on what the camera saw behind it: crowds, repeating tiles, and thin poles stay hard. Locally, use AI Object Remover.

Start from the thing you want gone in the shot you filmed. A walker, a cyclist, or a car that crosses the frame is a dynamic job: put points on the target in one frame and the tracker carries the mask. A bin, a parked sign, or other clutter that does not travel is static: stamp the region and let the fill rebuild what was behind it. If the job is a swap rather than a hole, that is Object Background Replace, not a second erase pass.

Pick dynamic or static first

• Walker, cyclist, or car crossing the frame → Dynamic Tracking Removal. Points on one frame, then the tracker carries the mask.
• A parked bin, sign, or fixed clutter → Static Object Removal. Stamp the region and let the fill rebuild what was behind it.
• Quick, Standard, and High Quality sit on the same rail. A higher tier spends more time on that mark. Preview before you queue an export.

The mark has to be an object in the footage — the person or the prop you can point at. Put the points on that subject, then watch the free 1s, 3s, or 5s preview where it crosses the mark. If the fill holds on that pass, queue the export on your GPU.

What the fill can borrow

• The fill paints texture that neighboring pixels support, or that a nearby frame showed once the object moved aside
• Crowds, repeating tiles, and thin poles remain hard: the mask can hold while the fill has nothing stable to borrow
• A long stretch where the object hides the background is harder than a brief cross

If someone stands in front of a one-of-a-kind sign for the whole shot, there is no clean patch to copy, and the hole stays a guess. Watch the free 1-second, 3-second, or 5-second preview where the subject crosses the mark. A single still can look closed while the next frame shows the hole crawling. If it crawls, regenerate the mask or switch tier before you export. Do not treat that still as a before-and-after.

How to erase the nuisance

1. Open the matching module and import the clip you filmed
2. Mark with points, regenerate if the mask crawls, and preview 1, 3, or 5 seconds on the worst pass
3. Export on the machine only if the fill holds while the subject crosses the mark

Sign-in and a paid plan are needed to export, convert, repair a file, or download a file from a link. Judge the motion, not one frame. If the background behind the mark was never in the shot, the higher quality tier still has nothing extra to copy. The erase runs on your machine, with no upload needed.

When to switch modules

Switch modules if the object only looks still. A car that rolls a few feet is dynamic. A sign that stays in the corner is static. Tracking a parked bin, or stamping a walker on one frame, is how the mask loses the target and the fill smears the background you wanted to keep. If the mask slips, regenerate before you change tier, and preview that same crossing again. If more than one object should go, mark them as separate jobs and preview each crossing. A crowd is not a stack of easy walkers. Dense overlap and thin poles stay hard on either module. Choose the module by the motion, not after the object category.