background-remover

name	background-remover
description	Remove backgrounds from images using rembg (IS-Net / U2Net). Outputs transparent PNGs or composites onto a solid colour or custom backdrop image. Works on portraits, products, and general subjects.
metadata	{"openclaw":{"emoji":"✂️","requires":{"bins":["uv"]}}}

knockout — Background Removal

Removes image backgrounds using rembg (IS-Net by default). The foreground can be kept as a transparent PNG or composited onto a custom background.

The skill directory (where this SKILL.md lives) is referred to as $SKILL_DIR below.

---

When to Use

Use this skill when the user wants to:

• Remove the background from portrait or product photos
• Place subjects onto a branded or custom background
• Prepare clean cutouts for other tools (e.g. clawimig padding, clawdepth)

---

Setup (first run only)

cd "$SKILL_DIR" && uv sync

Model weights are downloaded automatically on first use (~180 MB for IS-Net).

---

Agent Workflow

1. Ask the user

Before I process the images, I need to know:

✂️ Background
  - none          — keep as transparent PNG
  - #rrggbb       — solid colour (e.g. "#ffffff" or "#1a1a2e")
  - /path/to/img  — composite onto a background image

🧠 Model (optional, default: isnet-general-use)
  - isnet-general-use  — best for most subjects
  - u2net_human_seg    — optimised for people
  - isnet-anime        — optimised for illustrated/anime characters

⚙️  Device
  - auto  — GPU if available, else CPU (default)
  - cpu   — force CPU (~3s/image via ONNX CPU provider)

📁 Input / output directories  (default: ./input and ./output)

2. Edit config.json

3. Run

cd "$SKILL_DIR" && uv run python scripts/rembg_batch.py --config config.json

4. Report results

---

Config Reference

Key	Values	Default	Description
input_dir	path	./input	Source image folder
output_dir	path	./output	Destination folder
model	see models below	isnet-general-use	Segmentation model
bg	null / hex / path	null	Background (null = transparent)
output_format	png, webp	png	Output format
device	auto, cpu, cuda	auto	Inference device

Available models: isnet-general-use, isnet-anime, u2net, u2net_human_seg, u2netp, birefnet-general

---

Common Invocations

# Transparent PNG
cd "$SKILL_DIR" && uv run python scripts/rembg_batch.py

# White background
cd "$SKILL_DIR" && uv run python scripts/rembg_batch.py --bg "#ffffff"

# Dark brand background
cd "$SKILL_DIR" && uv run python scripts/rembg_batch.py --bg "#0d1117"

# Custom backdrop image
cd "$SKILL_DIR" && uv run python scripts/rembg_batch.py --bg ./brand-backdrop.jpg

# Force CPU
cd "$SKILL_DIR" && uv run python scripts/rembg_batch.py --device cpu

# Portrait-optimised model
cd "$SKILL_DIR" && uv run python scripts/rembg_batch.py --model u2net_human_seg

---

Output

Processed images written to output_dir as PNG (or WebP), same stem as input.