Strip image and video backgrounds without leaving the terminal

What it does BackgroundRemover is a Python command-line tool that segments foreground subjects from images and video using PyTorch-based u2net models. It can emit transparent PNGs, alpha-channel ProRes .mov files, matte keys for Premiere, or composite a subject onto a new background image or video. It runs locally, auto-detects a CUDA GPU for speed, and falls back to CPU if necessary.

The interesting bit The video pipeline is the unusual part: it processes frames in parallel, then re-encodes them into professional codecs like ProRes 4444 or VP9 with alpha, which is the kind of output you usually expect from a compositor, not a terminal one-liner. It also exposes an HTTP API server mode, so you can turn the CLI into a local microservice without writing a wrapper.

Key highlights

• Supports batch processing for both image folders and video folders.
• Offers alpha matting with tunable erosion thresholds for sharper or softer edges.
• Swaps between three segmentation models: general u2net, human-optimized u2net_human_seg, and a faster u2netp variant.
• Reads from stdin and writes to stdout for image pipeline integration.
• Auto-detects CUDA and falls back to CPU; the README cites roughly 5–10× speed-ups on GPU.

Caveats

• Transparent video output defaults to ProRes 4444, which most web browsers and VLC struggle to play correctly; the authors recommend mpv, QuickTime, or professional editors like DaVinci Resolve.
• Docker users need to manually increase shared memory (--shm-size=2g) for video processing, or multiprocessing crashes with an OSError.
• HEIC/HEIF image support requires an extra dependency (pillow-heif), and pipe mode works only for images, not video.

Verdict Video editors, data-augmentation pipelines, and developers building local media tools should keep this handy. If you only need the occasional selfie cutout, a browser tool is probably less fuss.