Removing a background from video without a green screen is now genuinely usable for most social and product work, and it still fails in the same three places it always did: fine hair, motion blur, and anything semi-transparent. Knowing where the failures are is the whole skill, because the difference between a clean result and an obviously cut-out subject is almost always in what you shot rather than what you processed.
What decides whether it works
Contrast between subject and background. The single biggest factor and the one you control on the day. A dark jacket against a dark wall is hard for every tool ever made.
Motion blur. Fast movement smears the edge, and a smeared edge has no correct answer. Shoot at a higher shutter speed if you know you will cut the subject out.
Hair, fur, and mesh. Where results still range from good to obviously wrong. Tied-back hair cuts cleanly; loose hair against a busy background does not.
Transparency. Glass, smoke, water, and anything the background shows through. No current tool handles these well without manual work.
Even lighting on the subject. Underexposed edges disappear into the background and cannot be recovered.
The practical workflow
1. Shoot for the cut if you know it is coming. Separation between subject and background, faster shutter, and even light on the edges. Five minutes on set saves an hour later.
2. Remove at the highest resolution you have. Edges are pixels; fewer pixels means worse edges. Do not downscale first.
3. Check on motion, not on a frame. A still frame always looks acceptable. Play it back: edge flicker between frames is the usual defect and it is invisible in a screenshot.
4. Composite with matching light. A subject cut from cool light and dropped onto a warm background reads as fake even with a perfect edge. Grade the subject to the plate, not the reverse.
5. Finish with a slight edge softness. Perfectly sharp edges look cut out. A pixel of feather sells it.
What this costs
Canon 8frame prices, a credit is $0.01 in a top-up pack. Background removal runs 12 credits per operation on images. For video, the practical route on a canvas is to generate the replacement environment rather than to roto a plate: an 8-second Veo 3.1 Lite clip silent is 33 credits, and putting a real product into a generated environment is often faster and cleaner than removing a real one.
That is the shift worth noticing. Historically you removed a background because you were stuck with the one you shot. Now, for product and context work, it is frequently cheaper to shoot on a clean surface and generate the world than to extract the subject from a bad one. See how to plan a product shoot.
When to shoot on green anyway
Fine hair with movement. Still the case for a green screen.
Anything semi-transparent. Glass, liquids, smoke.
Long-form with a person talking. Small per-frame errors accumulate across two minutes and become distracting.
When the composite has to be perfect. Broadcast, or anything on a large screen.
Green screen is not obsolete; it is now the tool for the hard 20% rather than the default for everything.
FAQ
Can AI remove a video background without green screen? Yes, well enough for social, product, and most web use. Fine hair, motion blur, and transparency remain the limits.
Why does my cutout flicker? The edge is being solved slightly differently per frame. More contrast in the source, higher resolution, and less motion blur all reduce it; there is no fix in post for a source that is too soft.
Is it better to remove the background or generate a new scene? For products, generating the environment around a cleanly shot subject usually beats extracting one from a messy scene. For people, removal is still the route.
Shoot for the cut, check on motion, and grade the subject to the plate. The 8frame canvas is free and unlimited, and generation is paid from $19/month.