ref-rip

Multi-source image reference scraper. Hand it a text query and it returns 100+ download options pulled in parallel from ten sources, deduped, quality-scored, and saved locally with a manifest.

Source pools (all hit in parallel)

Every run picks a fresh time-based seed unless --seed is given, so re-running the same query gives new options. Hits are scored on pixels + filesize + source-bias and the top N are kept. Sources without dimension metadata get a neutral default so they aren't unfairly outranked.

Quick use

python3 ~/.claude/skills/ref-rip/scripts/ref_rip.py "Breaking Bad still"

Drops 100 images into ./refs/Breaking_Bad_still_<seed>/ plus manifest.json. Default sources is all ten.

Common flags

# 200 results + contact-sheet preview
ref_rip.py "vintage gas station" -n 200 --contact-sheet

# Only Pinterest + Film-Grab + Tumblr (moodboard mode)
ref_rip.py "monochrome brutalist architecture" --sources pinterest,filmgrab,tumblr

# Only eBay + tshirtslayer (merch / product mode)
ref_rip.py "metallica vintage tour shirt" --sources ebay,tshirtslayer

# Reproducible run
ref_rip.py "found polaroids" --seed 42

# Manifest only (no downloads)
ref_rip.py "rusted machinery" --no-download --json

# Custom output dir
ref_rip.py "cathedral interior" -o ~/Refs/cathedral

Output

refs/<query>_<seed>/
├── 000_filmgrab_There_Will_Be_Blood_a1b2c3d4.jpg
├── 001_ebay_metallica_kill_em_all_e5f6g7h8.jpg
├── 002_tshirtslayer_iron_maiden_77_9c10857f.jpg
├── 003_ia_kennel_records_1923_...
├── ...
├── manifest.json         # full metadata for every hit
└── contact_sheet.png     # if --contact-sheet

manifest.json schema per hit: url, source, title, width, height, bytes, page_url, license, score, saved_path.

Source-specific notes

Pinterest, Tumblr, Flickr — install once for best results:

pip3 install --user gallery-dl

Without it, these three fall back to DDG site:DOMAIN queries — still works, just less direct.

Unblock Pinterest pinimg.com / login walls — --cookies BROWSER:

ref_rip.py "y2k fashion" --cookies chrome

Pulls cookies from your logged-in browser (chrome, safari, firefox, edge, brave, chromium) and passes them to gallery-dl. This is the only way to bypass Pinterest's CDN block + Tumblr/Flickr login walls. Quit the browser first — Chrome/Safari lock the cookie DB while running.

Film-Grab — uses film-grab.com directly. Best query style is movie-aware: "Drive 2011 Refn", "Blade Runner 2049 cinematography". Generic queries find fewer matching post pages.

FrameSet — uses DDG site-restricted search since frameset.app is a JS SPA. May return fewer hits than other sources.

eBay — only listing photos (i.ebayimg.com). Auto-upgrades thumbnail size suffix from s-l500 to s-l1600 for highest non-watermarked res. Great for product/merch refs.

tshirtslayer.com — search HTML → user-submitted shirt gallery photos. Strips /styles/THUMB/public/ from URLs to get originals where possible.

Tuning

• -n / --count — total kept (default 100)
• --per-source — how many each source fetches before global scoring/dedup (default count // 2)
• --workers — parallel download threads (default 8)
• --sources — comma list, subset of: ia, ddg, bing, pinterest, filmgrab, frameset, tumblr, flickr, ebay, tshirtslayer

When to use

• Building moodboards / reference packs for an edit, photo shoot, design comp, or AI image prompt
• Hunting public-domain assets for black-mirror, feverdream, sedition-haze, or any video edit skill
• Source material for mood-board, palette-rip, sticker-rip downstream
• Hunting band-tee designs, vintage merch, product photos for design/branding

Notes

• Internet Archive results carry license info in the manifest; other sources don't (check before redistributing)
• HTML-scrape sources (Bing, Film-Grab, eBay, tshirtslayer) can shift if those sites change their markup; the four other sources stay solid
• Pure stdlib + optional Pillow (for contact sheet) + optional gallery-dl (for Pinterest/Tumblr/Flickr)