Reference / Effects / Utility
Updated
Remove background
Remove background is a Utility effect.
What it does
Section titled “What it does”Reads the shot with a trained model and files how much of each pixel is the subject, one matte a frame. Another layer reads that matte through its Matte row, a Set matte or a track matte.
Robust Video Matting reads the shot in order, carrying what it made of one frame into the next. BiRefNet reads each frame on its own, is slower by a long way, and is not trained on people alone.
The model is an addon, installed from Settings ▸ Addons. Without it the effect wears a badge and leaves the picture alone. The badge names the pack the Model row asks for.
Parameters
Section titled “Parameters”| Parameter | Control | Range | Default | Unit |
|---|---|---|---|---|
| Analyse | Button | - | - | - |
| Cancel | Button | - | - | - |
| Model | Dropdown | Robust Video Matting · BiRefNet | Robust Video Matting | - |
| View | Dropdown | Composite · Matte | Composite | - |
| Invert | Checkbox | On or off | Off | - |
| Detail | Dropdown | Portrait · Full body | Portrait | - |
| Frames | Dropdown | All frames · Layer in to out · Work area | All frames | - |
| Save to folder | File | None | - |
In the Effect Controls panel:
- Detail is editable while Model is set to Robust Video Matting.
What each control does
Section titled “What each control does”- Analyse: read the shot and file its matte. Only one analysis runs at a time.
- Cancel: stop a run and keep the frames already read. A later Analyse carries on from them on BiRefNet, and reads the shot from the top on Robust Video Matting.
- Model: which trained model reads the frames.
- View: the layer with its background cut away, or the matte itself as a grey picture.
- Invert: keeps the background and cuts the subject away.
- Detail: how much of the frame Robust Video Matting works at. Portrait is head and shoulders; Full body is a whole person at a distance.
- Frames: which frames Analyse reads. All frames is the whole footage however the layer is trimmed, Layer in to out is the frames between the layer’s in and out points, and Work area is the frames inside the composition’s work area.
- Save to folder: a folder each matte is also saved to as an 8-bit greyscale PNG,
named by source frame (
matte_000123.png). On BiRefNet, frames already in the folder are read back rather than analysed again. Robust Video Matting reads every frame and writes over them.
Frames the analysis has not reached render untouched.