Video Background Removal
Use on-device segmentation to estimate a subject mask across the clip, then export over a selected chroma-key color for further compositing. Review the mask because results vary with the scene, motion, lighting, and source quality.
No green screen needed
Neural segmentation estimates a subject mask on processed frames; hair, motion blur, and low-contrast boundaries can need cleanup after review.
Chroma-key ready
Exports place the subject over a selected solid backdrop for chroma-key workflows in compatible editors.
On-device
The model runs through WebAssembly on your device. Supported video contents are not uploaded for processing.
How to remove a video background
- Open the tool and add your video Select a compatible file for Video Background Removal.
- Review and adjust Choose a replacement color, run segmentation, then review the subject mask across the clip and adjust the source or workflow if edges are missed before export.
- Export and download Review the exported result from Video Background Removal, then download it.
Frequently asked questions
Why a colored background instead of transparency?
The current opaque video export has no alpha channel, so the tool uses a selected solid key color for later compositing in a compatible editor. Edge quality and re-keying results vary with the mask, color choice, and editor.
Why the 30-second cap?
Segmentation evaluates processed frames with a neural model, which can use substantial memory and compute. Trim longer clips first; runtime and practical limits vary by browser, device, frame count, and resolution.
Which model is used?
The workflow can select a portrait-matting model (MODNet) when a face is detected and a general u2net segmentation model otherwise. Model selection and execution depend on the supported on-device WebAssembly path.
Can I get a transparent background instead of a color?
Not in the current opaque-video workflow. The export uses a selected solid key color for re-keying in compatible editors; transparent video output is not currently offered.
Does it work without a person in the shot?
It can attempt general-subject segmentation when no face is detected. Mask quality varies by subject separation, movement, lighting, and background, so review the full clip before export.