How is this different from the RVM background remover?

MODNet is an image-matting model applied per frame, which gives sharper edges (especially hair) on portraits. RVM (Robust Video Matting) is a recurrent video model that carries information between frames, so it stays more temporally stable on motion and full-body shots. For a steady portrait, MODNet often looks crisper; for movement, RVM usually looks smoother.

What is Temporal smoothing?

MODNet has no memory between frames, so its cutout edge can flicker. Temporal smoothing blends each frame’s mask with the previous one (an exponential moving average) to reduce that shimmer, at the cost of a little sharpness on fast motion. It is optional — leave it off for the crispest per-frame result.

MODNet is trained for portrait matting, so it works best on videos with a person, ideally fairly steady. Objects, animals and scenery may not separate cleanly.

Is my video uploaded?

No. The MODNet model runs in your browser via ONNX Runtime Web (WebGPU, or a WASM fallback), with WebCodecs for decoding and compositing — the file never leaves your device.

Remove Portrait Background Online (MODNet)

Remove the background from a portrait video using MODNet (github.com/ZHKKKe/MODNet), an image-matting model that produces especially crisp edges around hair and fine detail. MODNet runs right in your browser via ONNX Runtime Web (WebGPU, with a WASM fallback) and mattes each frame independently — so it shines on portraits and talking-head shots where the subject is fairly steady. Because it has no temporal memory, fast motion can make the cutout edge shimmer slightly between frames; turn on Temporal smoothing to damp that. For full-body or high-motion footage, the RVM-based remover stays more stable frame to frame. Everything runs locally in your browser using WebCodecs — your file is never uploaded to a server, so it stays private and conversion starts instantly.

How to remove portrait background

Drop a portrait video onto the box above.
Click Remove background — the MODNet model mattes each frame on your device.
If the cutout edge shimmers, enable Temporal smoothing and re-run.
Preview the cutout, then click Create MP4 to download the result.

Frequently asked questions

How is this different from the RVM background remover?: MODNet is an image-matting model applied per frame, which gives sharper edges (especially hair) on portraits. RVM (Robust Video Matting) is a recurrent video model that carries information between frames, so it stays more temporally stable on motion and full-body shots. For a steady portrait, MODNet often looks crisper; for movement, RVM usually looks smoother.
What is Temporal smoothing?: MODNet has no memory between frames, so its cutout edge can flicker. Temporal smoothing blends each frame’s mask with the previous one (an exponential moving average) to reduce that shimmer, at the cost of a little sharpness on fast motion. It is optional — leave it off for the crispest per-frame result.
What works best?: MODNet is trained for portrait matting, so it works best on videos with a person, ideally fairly steady. Objects, animals and scenery may not separate cleanly.
Is my video uploaded?: No. The MODNet model runs in your browser via ONNX Runtime Web (WebGPU, or a WASM fallback), with WebCodecs for decoding and compositing — the file never leaves your device.