AI background remover
Removing backgrounds from images and video with AI — where the edge quality is decided, why video is harder, and what to check before delivery.
Background removal isolates a subject from what is behind it. AI does this by segmentation — deciding per pixel what is subject and what is not — rather than by colour keying, which means it works on ordinary footage with no green screen.
Images versus video
The same problem, one dimension apart, and the difference is large:
- Images need one correct mask. Fast, and quality is now high enough for most uses.
- Video needs a correct mask on every frame and consistency between them. An edge that shifts slightly frame to frame reads as crawling or shimmer even when each individual frame looks fine.
Where quality is won or lost
Almost entirely at the edges, and in predictable places:
| Hard case | Why | What helps |
|---|---|---|
| Fine hair, fur | Thousands of sub-pixel boundaries | Contrast between hair and background |
| Motion blur | The edge genuinely is ambiguous | Faster shutter when shooting |
| Transparency — glass, smoke | Partly subject and partly background at once | Manual cleanup; segmentation handles it poorly |
| Subject matching background tone | Nothing to separate on | Separate them with light or colour when shooting |
| Thin gaps — between arm and body | Easily filled in by mistake | Poses with clear separation |
Shooting for it
If you control the capture, four choices make removal near-perfect and cost nothing: light the subject separately from the background so tones differ, keep a contrasting backdrop, avoid heavy motion blur, and keep limbs clear of the torso so there are no thin gaps to resolve.
Checking before delivery
- Place it on a contrasting background — dark subject on white, light on black. Fringing that is invisible on grey appears immediately.
- Zoom to 100% on hair and edges.
- For video, step through frame by frame at the moments of fastest motion.
- Look for halo — a thin rim of the old background surviving around the subject.
Common questions
Why is video background removal worse than image?
Every frame needs a correct mask and the masks must agree with each other. Small per-frame differences read as crawling edges even when each frame looks fine alone.
Do I need a green screen?
No. AI segmentation works on ordinary backgrounds. A contrasting background still improves edge quality substantially.
How do I fix a halo around the subject?
It is background pixels surviving in the edge. Contract the mask slightly or, when shooting, increase separation between subject and backdrop.