BiRefNet — complete background removal on Replicate

The full BiRefNet model zoo behind one endpoint: all 10 official variants plus ToonOut for anime/stylized art. Returns a transparent cutout or a raw matte. MIT-licensed — free for commercial use.

Highlights

• 11 models, one input — switch with variant; one warm container serves them all.
• Anime / illustration — variant=toonout handles hair wisps, line art, and translucency that general removers smear.
• Animated GIF & WebP — multi-frame inputs are matted frame-by-frame and returned as an animated WebP with full transparency (frame timing + loop preserved).
• Real controls — cutout (RGBA) or raw mask; grow / shrink / blur the matte; refine foreground colours; pick inference resolution.
• MIT / commercial-OK — unlike BRIA-based removers, the output is free to use commercially.

Variants

Inputs

• image — any image; animated GIF/WebP supported (→ animated WebP output).
• variant — which model (table above). Default general.
• resolution — square inference size; 0 = the variant’s native resolution.
• output_format — cutout (RGBA) or mask (raw alpha matte).
• mask_blur / mask_offset — soften / grow / shrink the matte.
• refine_fg — FB blur-fusion to strip background bleed on soft edges.

Credits

Unofficial community packaging — not affiliated with the model authors; we earn nothing (compute fees go to Replicate).

• BiRefNet — Peng Zheng et al. · MIT · https://github.com/ZhengPeng7/BiRefNet
• ToonOut — Matteo Muratori & Joël Seytre · MIT · https://arxiv.org/abs/2509.06839

Source: https://github.com/sprited-ai/birefnet · Anime-only endpoint: https://replicate.com/sprited/birefnet-toonout