EGO HERO / BACKGROUND LAB
Lose the background.
Keep the edge.
Four specialist models. One beautifully simple studio. Pick for speed, hair, graphics or native 2K detail.
01 / CHOOSE YOUR SPECIALIST
What kind of edge are we keeping?
02 / ADD AN IMAGE
Drop it. We’ll do the delicate part.
The dependable all-rounder and the best first try for mixed image types.
Clicking Remove background uploads this image to EGO HERO for temporary server processing. Do not submit sensitive or unauthorised imagery. See our Privacy Policy. By continuing, you also agree to the Terms.
03 / COMPARE THE MODELS
Same task. Different superpowers.
Measured end-to-end on this CPU service with a 2560 × 1440 image. Queueing or post-restart model warm-up can add time.
| Model | Best for | Measured 2K wait | Max inference support | Download size | Speed | Fine detail |
|---|---|---|---|---|---|---|
| MODNet | Portraits, full body, clean studio shotsSoft hair alpha | 10–15 sec | 2048 px long edge | Original dimensions | ||
| ISNetRECOMMENDED | People, products, pets, stickersReliable hard edges | 7–10 sec | 1024 × 1024 | Original dimensions | ||
| Lucida | Logos, glow, art, translucent detailsComplex graphic edges | 34–42 sec | 1024 × 1024 | Original dimensions | ||
| BiRefNet Lite-2K | Messy hair, tiny gaps, intricate full bodyBest fine-detail retention | 50–60 sec | 2560 × 1440 · rotates for portrait | Original dimensions |
Resolution promise: inference support is the detail the model actually examines. Download size is always restored to your original pixel dimensions, up to the 24-megapixel upload limit.
Will the result keep my original resolution?＋
Yes. The downloaded result always uses the source image’s exact width and height. Each model computes its mask at the inference resolution shown in the table, and EGO HERO restores that mask to the original canvas.
Which model should I use for long, messy hair?＋
Start with BiRefNet Lite-2K. It is the slowest option, but it retains the finest strands and small gaps most reliably. MODNet is the faster portrait alternative.
Are images processed locally on my phone?＋
No. These four specialist models run on EGO HERO’s isolated inference service. Uploads and results are scheduled for deletion about 30 minutes after the job ends and are not used for training.
Why can JPEG not be transparent?＋
JPEG has no alpha channel. PNG and WebP preserve transparency; JPEG downloads place the cutout on the background colour you choose.