The AI Wardrobe Playbook — Run the App Sam Altman Called 'a Whole Startup' Tonight, Then Steal Its Pattern for 10 Businesses
The companion to the video on Wardrobe (github.com/tandpfun/wardrobe) — the open-source, MIT-licensed app by Thijs (@cdngdev) that Sam Altman quote-tweeted: 'this would have been a whole startup not too long ago.' Drop camera-roll photos into a local-first React app and a vision model (gpt-5.4-mini) detects every garment with bounding boxes, hex colors, and tags under a strict JSON schema; gpt-image-2 redraws each piece as a clean catalog cutout on a mathematically-chosen chroma-key background the code keys out pixel-by-pixel with spill removal; then images/edits combines your model-reference photo with each cutout into an editorial photo of YOU wearing it. THIS PLAYBOOK: the full 15-minute setup with every exact command and .env variable, the model-reference photo rules that decide 80% of your render quality, the green-screen trick explained simply (image models can't output transparency — so the app builds its own chroma screen and composites in code), the honest cost table across all three quality tiers with per-garment math (about 2 generations per garment — 40-60 cents at high, ~10 cents at medium, pennies at low) and the two-pass move that cuts the bill, the Outfit skill's three styling rules (tonal harmony, one statement piece, silhouette balance), 10 ways to point the extract-and-rerender pattern at businesses beyond your closet — reseller listings, e-commerce catalogs, furniture staging, food menus, packing lists, capsule audits, insurance inventories, sneaker catalogs, rental gear, try-before-you-buy — each with a concrete copy-paste prompt pattern, and the troubleshooting page for the three walls everyone hits first. Independent field notes from Hyperautomation Labs — not affiliated with the repo.
Free. No spam. Unsubscribe anytime.