/product-image-processor

What it does

Type /as:product-image-processor in Claude Code and it reads the image URLs already in your project’s product library. The skill downloads every image, resizes to a consistent max dimension, and removes backgrounds using AI — saving clean, transparent PNGs ready for spec books, presentations, or procurement documents.

Three output folders, one per stage: originals/, resized/, nobg/.

Install

Claude Desktop:

1. Open Customize → Browse plugins
2. Click + → Add marketplace from GitHub
3. Enter AlpacaLabsLLC/skills-for-architects
4. Install Architecture Studio

Claude Code (terminal):

claude plugin marketplace add AlpacaLabsLLC/skills-for-architects
claude plugin install as@skills-for-architects

Or symlink just this skill:

git clone https://github.com/AlpacaLabsLLC/skills-for-architects.git
ln -s $(pwd)/skills-for-architects/skills/product-image-processor ~/.claude/skills/product-image-processor

Dependencies

Requires Python 3.9+ with Pillow, rembg, and onnxruntime. The skill auto-installs missing packages on first run. The u2net background removal model (~170MB) downloads once and is cached.

Usage

/as:product-image-processor

No arguments needed. The skill finds your project’s product-library.csv, validates it, and reads the Image URL and Product Name fields by name. It asks only for an output location if it can’t infer one, suggesting ./product-images-YYYY-MM-DD/. The library itself is never modified.

/as:product-image-processor

Sheet: 1FMScYW9guezOWc_...
Image URL column: AC
Name column: C (Product Name)
Output: ~/Documents/Work-Docs/product-images-2026-03-04/

Processing pipeline

Stage	What happens	Tool
Download	curl -L each URL, preserve original format	curl
Resize	Scale to max 2000px longest edge, convert to PNG, skip upscaling	Pillow
BG Remove	AI background removal via u2net, output transparent PNG	rembg

Files are named sequentially with slugified product names: 001-eames-lcw.png, 002-vardagen-vaso.png, etc.

How it handles failures

The skill never stops a batch on a single failure:

• Download failures (404, timeouts) — logged and skipped, batch continues
• Resize failures (corrupt files) — logged and skipped in subsequent stages
• rembg failures (vectors, icons) — logged, original kept in resized/
• Library validation errors — stops before reading and reports the header or row error, leaving the file untouched

After every batch: success/failure counts per stage with reasons.

Pairs with

Use /as:product-spec-bulk-fetch to extract product specs and image URLs from vendor pages into the product library, then run /as:product-image-processor against it. Fetch specs → process images → spec-ready assets.