remove-background

Remove background from images using AI (rembg / U²-Net). Use this skill whenever the user asks to: remove background, cut out object, make transparent background, delete background, extract subject, r

SKILL.md

GitHub

Remove Background

AI-powered background removal from images using rembg (U²-Net neural network).

How It Works

The skill uses remove_bg.py (in the same directory as this SKILL.md) that wraps the rembg library. The script auto-installs dependencies if missing.

Quick Start

For most requests, just run:

python <skill-dir>/remove_bg.py "<input_image>" "<output_image>"

Replace <skill-dir> with the actual path to this skill's directory (the folder containing this SKILL.md).

Available Models

Model Speed Quality Best for
u2net (default) Medium High General purpose — objects, products, people
u2netp Fast Good Quick previews, batch processing
u2net_human_seg Medium High People, portraits, full-body shots
isnet-general-use Medium High Complex scenes, fine details

Use --model flag to select: --model u2net_human_seg

Output Formats

Format Flag Result
Transparent PNG --format png (default) Subject on transparent background
White background --format white Subject on solid white
Black background --format black Subject on solid black

Advanced Options

  • --alpha-matting — Enables alpha matting for cleaner edges on hair, fur, and fine details. Slower but better quality.

Workflow

  1. Identify the input image path (from user upload or workspace)
  2. Decide on model — use u2net by default, u2net_human_seg for people
  3. Run the script:
    python <skill-dir>/remove_bg.py "/path/to/input.jpg" "/path/to/output.png" --model u2net --format png
    
  4. Save output to the workspace folder so the user can access it
  5. Show the result to the user with a computer:// link

Batch Processing

For multiple images, run in a loop:

for img in /path/to/images/*.jpg; do
    python <skill-dir>/remove_bg.py "$img" --format png
done

Output files are saved as <original_name>_nobg.png by default.

First Run

The first run downloads the U²-Net model (~170MB). Subsequent runs are instant since the model is cached. If rembg is not installed, the script auto-installs it.

Troubleshooting

  • Slow first run: Model download, normal — subsequent runs are fast
  • Poor edges on hair/fur: Add --alpha-matting
  • Memory issues on large images: Resize first with Pillow, then remove background
  • Wrong subject detected: Try isnet-general-use model