Hair, fur and semi-transparent edges are where most background removers fall apart. BiRefNet keeps the strands.
Ordinary background removal produces a hard binary mask: a pixel is either subject or background. Hair breaks that assumption, because a single pixel at the edge of a curl is genuinely part subject, part background. The result is the familiar halo, or hair chopped into a helmet-shaped blob.
Matting models solve this by predicting a continuous alpha value per pixel. ImageLabs offers BiRefNet for exactly this case — portraits, pets, glassware, motion blur — alongside a cheaper one-credit pass for clean-edged product shots.
Upload the portrait: Higher input resolution gives the matting model more strand detail to recover.
Pick the engine: BiRefNet (2 credits) for hair, fur and glass. The standard pass (1 credit) for hard-edged products.
Export a transparent PNG: Alpha is preserved exactly — no flattening, no white fringe baked in, no downscale.
Full-resolution free tier: No 0.25-megapixel preview cap. What you upload is what you download.
Batch mode: Drop a folder of portraits, choose the engine once, and process the set.
Alpha preserved end to end: Exports keep true transparency so composites do not show a grey halo.
Which model handles hair best? BiRefNet. It predicts soft alpha rather than a hard mask, which is what strand-level edges require.
Do I get a full-resolution PNG for free? Yes — every plan, including the free daily credits, downloads at original resolution with no watermark.
Can it handle glass or motion blur? Yes. Both are soft-alpha problems, the same class BiRefNet is built for.