Background removal

The technique is image segmentation. A model classifies each pixel as subject or background, producing a mask that the editor uses to cut the subject out. Video adds a temporal requirement: the mask must stay stable frame to frame, otherwise edges shimmer, and the better systems reference nearby frames rather than treating each one independently.

For creators it removes the setup cost of a physical green screen. A subject can be placed on stock footage, a screenshot, a solid colour, or a stacked layout without lighting a backdrop, and product shots can be isolated for thumbnails and slideshow frames. It is what makes reaction and commentary formats shootable in any room.

The nuance is that edges remain the hard part. Fine hair, glasses frames, transparent objects, motion blur, and low contrast between subject and wall all produce artefacts, and a busy background makes it worse. Even lighting, a plain wall, contrasting clothing, and slower movement improve the mask far more than adjusting settings after the fact.

In workflow terms, removal happens before compositing and cropping, since the replacement background needs to be sized for the final vertical frame rather than the original horizontal one. Choosing a replacement that contrasts with the subject also hides small mask errors that a similar coloured background would put on display.