People & Portraits · 5 min
Removing Backgrounds from Portraits and Headshots
Portraits are where most background removers fall apart. Hair is the usual failure point — frizzy strands get cut off, or the background bleeds through in a gray halo. Getting it right matters for team pages, about sections, LinkedIn photos, and editorial work.
Pick the portrait model first
The portrait model is tuned for people and tends to preserve hair detail better than general-purpose models. Start there.
If the result is decent but a few locks of hair are cut, switch to the precise model and compare — they often produce different but equally valid interpretations of the same strands.
Refining hair edges with the brush
Hair is where the brush tool earns its keep. Use the restore brush at a medium size and soft hardness to recover individual strands the AI missed.
Avoid painting aggressively — restore only what the original image actually shows. Over-restoring brings background pixels back into the cutout.
Privacy matters for people photos
For anything sensitive — internal team photos, personal portraits, work under NDA — use the on-device model. The image never leaves your browser.
On a modern laptop with WebGPU it runs fast enough that you barely notice the difference vs. the cloud model.
Common portrait gotchas
A few edge cases to watch for:
- Dark hair on dark background: the AI may merge the two. Try the precise model, which tends to respect value gradients better.
- Glasses: lenses can be over-cut. Refine with the restore brush if the lens becomes partially transparent.
- Earrings and jewelry: thin metal often gets partially cut. Easiest to fix with the restore brush.
- Head wraps and hats: the AI generally handles these well, but confirm the brim or edge wasn’t over-cut.
FAQ
Does it handle group photos?
Yes — all subjects in the foreground are kept, the background is removed. Brush refinement may be needed between closely adjacent people.
Will my face be recognized or stored?
No. The on-device model runs fully in your browser and nothing is sent anywhere. The cloud models only process the image in memory to produce the output — no storage, no training.
Try it now.
Free. Instant. No signup.
Keep reading