sprited/birefnet-video

BiRefNet video background removal — every variant + ToonOut, applied per frame. Transparent WebM output by default (also MP4/MOV/GIF). MIT-licensed for commercial use.

Public
528 runs
License
GitHub
Weights
Paper
Playground API Examples README Versions

Examples

View more

Run time and cost

This model costs approximately $0.025 to run on Replicate, or 40 runs per $1, but this varies depending on your inputs. It is also open source and you can run it on your own computer with Docker.

This model runs on Nvidia L40S GPU hardware. Predictions typically complete within 26 seconds. The predict time for this model varies significantly based on the inputs.

Readme

BiRefNet — video background removal

The full BiRefNet variant zoo + ToonOut, applied frame-by-frame to a video. Sister model to the stills endpoint sprited/birefnet. MIT-licensed — free for commercial use.

Highlights

  • Transparent output by default — VP9 .webm with full alpha (MP4 can’t carry transparency). Also ProRes 4444 .mov, .mp4 (composited), or .gif.
  • Anime / illustration — variant=toonout handles hair wisps, line art, and translucency that general removers smear.
  • The whole zoo, one input — general / HR / portrait / matting / dynamic / lite / 2K, swapped per request via variant.
  • Controls — cutout or raw mask; refine foreground; grow/shrink/blur the matte; quality and write-mode encoding presets.

Per-frame note: BiRefNet has no temporal model, so fast motion can shimmer slightly between frames — the tradeoff for variant choice + ToonOut quality. For still images and animated GIF/WebP, use sprited/birefnet.

Inputs

  • video — mp4 / mov / webm / m4v
  • variant — default general
  • output_format — cutout or mask
  • video_output_type — webm (default, alpha) / mov (alpha) / mp4 / gif
  • background_color, resolution, refine_fg, mask_blur, mask_offset
  • video_quality (low/medium/high/maximum), video_write_mode (fast/balanced/small), precision (fp32 default / fp16)

Credits

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

Model created