A frog sits in the middle of the shot. Two seconds later it's gone, and the moss and water behind it are still there, moving with the camera. No blur patch, no wobble at the edges. That's the clip above, original on the left.
Removing an object from a still photo is solved. Removing one from video is a different problem, and that's why so many "free AI object remover" results look smeared.
Why video is harder than a photo
A photo remover only has to invent one plausible patch. In video the patch has to be:
- In the same place on every frame, even as the object and camera move.
- Consistent over time, so the rebuilt background doesn't shimmer.
- Correct where the object overlapped something: a hand passing in front of it, a shadow it cast, a reflection.
If any of those is off, the eye catches it instantly, even when every single frame looks fine on its own.
The workflow that works
1. Track the object, not just one frame
Draw the mask once and let tracking carry it across the shot. Then scrub through it. The frames to check are the ones where the object turns, is partly hidden, or leaves frame. Fix the mask there. Per-frame mask control is the difference between a usable shot and a redo.
2. Include the shadow and reflection
The object is gone but its shadow is still on the floor? That's the most common giveaway. Grow the mask to cover the shadow, and any reflection in glass or water.
3. Remove and rebuild the background
Run the removal on the tracked region only. The background behind the object is rebuilt across the whole shot, so it moves with the camera instead of sitting on top of the image like a sticker.
4. Check edges at full size
Play the result at 100% around the old edge of the mask. Halos and soft edges show up there first. If a few frames are off, adjust the mask on those frames and rerun just that range.
Asking an agent to do it
In Shoty you can do the steps above by hand, or ask the agent in the chat:
Remove the frog from shot 2. Track it through the whole clip, include its shadow, and keep the water and moss moving naturally. Show me the result before replacing the original.
The agent tracks the object and runs the removal. With the run mode set to review, it waits for your approval before anything replaces the original. You still decide if it's clean enough to ship.
Common removal jobs
- Crew, gear and boom mics that drifted into frame
- Logos and signs you don't have clearance for
- Passers-by in street shots
- Wires and rigs in stunt or product shots
- Modern objects in period scenes
The same tracked-mask approach covers all of them. What changes is how big the hole is. The more of the frame you remove, the more the background has to be invented, so check those shots extra carefully.
Want to try it on your own footage? Download the free Shoty editor. If you're editing with an AI agent, see how to edit video with Claude.