remove person from video aierase person from videoremove passerby from videoAI VFXvideo inpainting

Remove Person From Video With AI: No Blur, No Leftover Shadow

How to remove a person from video with AI: no blur patch, no leftover shadow. A tracked-mask workflow for erasing a passerby or crew member from your shot.

I ChelI Chel
October 25, 2026
Before
After
Chameleon removal, made in Shoty. Original footage on the left, result on the right.

In the clip above, a chameleon disappears from the branch while the leaf, the light and the camera move all keep going exactly as before. No smeared patch, no wobble where it used to sit. That's the same tracked-mask-and-rebuild technique this guide uses for a messier job: erasing a person who walked into your shot.

A person is a bigger, busier hole than most objects. Get the same technique wrong on a person and it shows up fast.


Why a person is harder to remove than most objects

Removing a small object is mostly about the hole it leaves. A person adds three problems on top of that:

  • A bigger area to rebuild. Arms, legs and a torso cover more of the frame than a bottle or a sign, so there's less nearby background left to copy from.
  • A shadow and sometimes a reflection. A person casts a longer, more visible shadow than most props, and it moves as they walk.
  • Overlap with things that matter. A passerby crossing in front of a shopfront sign, a crew member stepping past the lead actor, a photobomber walking through a wedding shot behind the couple.

None of that is a reason to rule out AI removal. It's the reason to track the whole clip before you run anything.

The workflow

1. Track the person for every frame they're in

Draw the mask on the clearest frame, where the person is fully visible, and let tracking carry it forward and back from there. Scrub the result before running anything. The frames to check are where they turn, get partly blocked by something else in the scene, or step out of frame.

2. Grow the mask to cover the shadow

A person's shadow is the giveaway that gets missed most often. If the shadow falls across the floor, the ground or a wall, extend the mask to cover it too, or the shot will still read as "something was there."

3. Remove and rebuild in passes if the hole is large

When the person covers a wide stretch of the frame for several seconds, running the removal on the whole range at once asks the model to invent more than it has reference for. Splitting a long pass into shorter, overlapping ranges usually holds up better than one long run.

4. Check edges and anything with a straight line behind it

Play the result at 100% around the old edge of the mask. Fences, window frames, signage with text and tiled floors are where a rebuilt background most often goes slightly crooked or blurry. If a short stretch is off, fix the mask on just those frames and rerun that range only.

Asking the agent to do it

Remove the person walking through the back of shot 5. Track them through the whole clip, include their shadow on the sidewalk, and rebuild the shopfront behind them. Show me the result before replacing the original.

With the run mode set to review, the agent tracks the person, runs the removal, and waits for your approval before anything replaces the original footage.

This crew member steps into frame for about two seconds near the end of shot 2. Remove just that range and leave the rest of the shot untouched.

Naming the exact range keeps the agent from reworking frames that were already clean.

Common jobs this covers

  • Crew and boom operators who drift into frame during a take
  • Passers-by crossing a street shot
  • Photobombers walking through event or wedding footage
  • A stand-in or body double left in a frame meant for the final actor only
  • Tourists or bystanders in the background of travel footage

The approach is the same for all of them. What changes is how long the person is in frame and how much they overlap with something the background needs to keep looking right.

Where this still needs a person checking it

  • Multiple overlapping people. If two people cross paths while one is being removed, fix one at a time and recheck the frames where they overlapped.
  • Unique backgrounds. A sign with readable text, a face in the background, or a patterned surface behind the removed person is harder to rebuild convincingly than a plain wall or foliage. Check those shots at full size before you ship them.
  • Very long holes. The longer a person blocks a part of the frame, the less clean background the model has to work from. If a shot has them standing still in one spot for a long stretch, consider reshooting a clean plate instead.

Download the free Shoty editor to try it on your own footage. For the same technique applied to objects instead of people, see how to remove an object from video with AI.

FAQ

Can AI remove people from the background of a video?

Yes, as long as the person is tracked across the whole shot and not just erased frame by frame. The mask has to follow them as they move, and the background behind them has to be rebuilt across time so it doesn't shimmer once they're gone.

How do I erase a person from a video without a blur patch?

The blur patch comes from filling the hole one frame at a time from nearby pixels. Track the person first, grow the mask to include their shadow, then run the removal on the tracked region across the whole shot so the rebuilt background moves with the camera instead of smearing.

Can I remove a passerby who walked through my shot?

Yes. A passerby crossing the frame is usually easier than a subject who lingers, because they only cover a small range of frames. Track them for just that range, check the frames where they overlap something in the background, and rebuild only that stretch.

Is it free to remove a person from a video with AI?

The Shoty editor and its timeline tools are free to download and use. The AI removal itself is a generation and uses credits. Trimming the clip down to just the frames where the person is in shot keeps a test run cheap.

Try it in Shoty

The editor is free on Mac and Windows. Bring Claude, ChatGPT or Gemini on the plan you already pay for.

Download the free editor