Available now

BiRefNet V2 Video Background Removal: video background removal

BiRefNet V2 removes the background from a video and returns the subject alone, with an optional mask video.

Model ID

Input

Fields marked * are required. Review the price before generating.

URL of the video to remove background from

Uploading a file needs an account. Create account

MP4, WebM · up to 100 MB

Additional settings

Whether to output the mask used to remove the background

Whether to refine the foreground using the estimated mask

Price$0.83 / run

You sign up first. Your settings are kept.

About this model

Overview

Upload a video and pick one of six model variants: light, light 2K, heavy, matting, portrait or dynamic. The output comes as MP4, WebM, MOV (ProRes 4444) or GIF, and you can also get the mask it used.

A higher operating resolution helps with high-resolution footage; the largest one works only with the dynamic variant. Foreground refinement is on by default and cleans the edges around the subject.

What you can build

  • Cut-out presenters to place over slides or new backgrounds
  • Product clips with the background removed for compositing
  • Masks for color grading or effects in an editor

Prompting tips

  • Pick WebM or ProRes 4444 MOV when you need a transparent background; MP4 has no alpha channel.
  • Use the portrait variant for people and the matting variant for hair and other soft edges.
  • Start with the light variant; switch to heavy when edges look rough.

What to check

Play the whole result and mask when requested. Check edge flicker, lost details and compatibility with your editing software.

Inputs

  • Model (optional): Model to use for background removal. The 'General Use (Light)' model is the original model used in the BiRefNet repository. The 'General Use (Light 2K)' model is the original model used in the BiRefNet repository but trained with 2K images. The 'General Use (Heavy)' model is a slower but more accurate model. The 'Matting' model is a model trained specifically for matting images. The 'Portrait' model is a model trained specifically for portrait images. The 'General Use (Dynamic)' model supports dynamic resolutions from 256x256 to 2304x2304. The 'General Use (Light)' model is recommended for most use cases. The corresponding models are as follows: - 'General Use (Light)': BiRefNet - 'General Use (Light 2K)': BiRefNet_lite-2K - 'General Use (Heavy)': BiRefNet_lite - 'Matting': BiRefNet-matting - 'Portrait': BiRefNet-portrait - 'General Use (Dynamic)': BiRefNet_dynamic. Options: General Use (Light), General Use (Light 2K), General Use (Heavy), Matting, Portrait, General Use (Dynamic).
  • Operating Resolution (optional): The resolution to operate on. The higher the resolution, the more accurate the output will be for high res input images. The '2304x2304' option is only available for the 'General Use (Dynamic)' model. Options: 1024x1024, 2048x2048, 2304x2304.
  • Output Mask (optional): Whether to output the mask used to remove the background.
  • Refine Foreground (optional): Whether to refine the foreground using the estimated mask.
  • Video (required): URL of the video to remove background from.
  • Video Output Type (optional): The output type of the generated video. Options: X264 (.mp4), VP9 (.webm), PRORES4444 (.mov), GIF (.gif).
  • Video Quality (optional): The quality of the generated video. Options: low, medium, high, maximum.
  • Video Write Mode (optional): The write mode of the generated video. Options: fast, balanced, small.

Output and limits

  • Readable text inside the video, especially small text, is unreliable.
  • Scenes with several interacting characters can lose consistency.
  • Longer clips can show more motion drift or scene changes.
  • Low-quality inputs cause glitches: heavy compression, collages, screenshots and text overlays.
  • Some prompts or outputs can be refused by safety checks.

Price guide

  • up to: $0.83 / run