</p><h2>What Wardrobe Does and How It Works</h2><p>Wardrobe uses OpenAI's Responses API for garment detection and the Images API (gpt-image-2) for extracting clean, transparent-background cutouts of each piece of clothing. The process is straightforward: upload a photo containing multiple garments, and Wardrobe identifies each item, separates it, categorizes it by type (shirt, pants, dress, shoes, accessory), detects colors and patterns, and stores everything in a local JSON database at <code>data/library.json</code>. The original photos, generated images, and metadata all stay in a local <code>data/</code> directory, which means no external storage dependency and no cloud lock-in beyond the OpenAI API itself.</p><p>The tool offers two modes of operation. The web UI provides drag-and-drop import for manual use, while the built-in Codex skills automate bulk workflows. The import skill, <code>$import-clothes</code>, accepts a folder of photos, extracts and catalogs each garment, generates modeled editorial previews using a provided reference photo, and writes everything to the local data store. The outfit-generation skill, <code>$generate-outfits</code>, creates complete modeled lookbooks from the cataloged wardrobe, enabling a user to see themselves wearing different combinations without actually trying anything on.</p><p>Getting started takes three commands and an API key:</p><pre><code>git clone https://github.com/tandpfun/wardrobe.git cd wardrobe npm install cp .env.example .env

Add OPENAI_API_KEY to .env

npm run dev</code></pre><p>The importer stays disabled until you add an OpenAI API key to the environment and place a PNG reference photo of yourself at <code>data/model-reference.png</code>. The reference photo is used by gpt-image-2 to generate modeled images of you wearing the cataloged garments, turning a flat clothing catalog into a virtual fitting room.</p><h2>Why Wardrobe Exploded: The GPT Vision for Everything Trend</h2><p>Wardrobe's rapid growth is not just about clothing. It is a signal that GPT-4o's vision capabilities have reached a threshold where practical consumer applications are trivially easy to build. Two years ago, building a clothing cataloging tool required training a custom convolutional neural network, sourcing thousands of labeled images, and deploying a model inference pipeline. Today, it requires a JavaScript file, one API call, and a weekend. The commoditization of computer vision is happening faster than most founders realize.</p><p>The same pattern Wardrobe uses for clothing can be applied to inventory management, pantry tracking, hardware parts identification, plant species recognition, or any domain where visual classification of physical objects is useful. The code changes are minimal: point GPT-4o at a different category of objects, adjust the prompt, and you have a new product. For solo founders and indie hackers, this represents a massive surface area for new applications. The barrier to entry for AI vision products has dropped from six figures and six months to a few dollars in API costs and a weekend of coding.</p><p>The GitHub star trajectory confirms market demand. One thousand stars in two days is not viral by GitHub standards, but it is strong validation for a niche utility. For comparison, similar GPT-vision-powered tools in adjacent categories have followed a similar curve: fast initial growth driven by novelty and utility, followed by sustained adoption as users integrate the tool into daily workflows. Wardrobe's early traction suggests that AI-powered organization tools have genuine consumer demand beyond the novelty spike.</p><h2>Technical Architecture and Design Decisions</h2><p>Wardrobe's architecture is deliberately simple. The front end is a Vite-based JavaScript app with drag-and-drop upload. The backend is a Node 22 server that proxies requests to OpenAI's API. The data layer is a flat JSON file. That is the entire stack. There is no database, no authentication system, no cloud storage integration. The simplicity is a feature: it means Wardrobe can be deployed on any machine with Node installed, works offline after initial API calls, and requires zero ongoing infrastructure costs beyond the OpenAI token usage.</p><p>The configuration exposes several tunable parameters. Users can set the vision model (defaults to gpt-5.4-mini), the image generation model (gpt-image-2), and the image quality (high or standard). The model reference for generating modeled photos is configurable via <code>WARDROBE_MODEL_REFERENCE</code>. This flexibility allows users to trade off between cost and quality based on their use case, from casual wardrobe browsing to professional e-commerce catalog generation.</p><p>The MIT license means anyone can fork, modify, and redistribute the code without restriction. For founders, this is significant: Wardrobe is not just a tool but a starting point. A forked version with multi-user support, a SQLite database, and Stripe subscriptions could become a SaaS product targeting fashion resellers or boutique owners. The core vision pipeline is ready; the business model is the only missing piece.</p><h2>Comparison to Alternatives</h2><p>Wardrobe occupies a specific niche that existing tools do not fully address. Commercial wardrobe apps like Cladwell and Pureple focus on outfit recommendations and style analysis but require manual input and do not auto-catalog from photos. AI fashion search tools like Vue.ai and Syte are enterprise products targeting retailers, not individuals. Wardrobe sits in the middle: it provides AI-powered auto-cataloging for individuals, with the flexibility of open-source customization.</p><p>The closest technical alternative is probably a self-hosted CLIP-based pipeline using Facebook's CLIP model for zero-shot image classification. While CLIP offers more privacy (no API calls) and lower ongoing costs (no per-token fees), it requires Python, GPU access, and significantly more setup effort. Wardrobe trades those requirements for a dead-simple JavaScript setup and the superior accuracy of GPT-4o's vision model. For most users, especially non-technical ones, Wardrobe's tradeoff is the right one.</p><p>The key limitation is the dependency on OpenAI's API. Every photo upload incurs token costs for vision processing, and every modeled image generation incurs image generation costs. For casual use, these costs are negligible, but for bulk cataloging of hundreds of garments, the API bills add up. A future improvement could add support for local vision models like LLaVA or Florence-2, giving users a zero-cost option for basic cataloging while reserving GPT-4o for high-quality modeled previews.</p><h2>Who This Is For</h2><p>Wardrobe is for three distinct audiences. First, individual consumers who want to catalog their own clothes for easier outfit planning and wardrobe management. The web UI and simple setup make it accessible to anyone comfortable with a terminal and an API key. Second, indie hackers and solo founders looking for a template to build their own GPT-vision-powered products. The codebase is clean, well-documented, and MIT-licensed, making it an ideal starting point for derivative projects in inventory, retail, or logistics. Third, fashion resellers and small boutique owners who need to generate product catalogs from unsorted inventory photos. A forked version with multi-user support and payment integration could serve this market directly.</p><p>The deeper lesson for founders is that Wardrobe validates a product pattern that is replicable across dozens of verticals. Pick a category of physical objects that people own in volume, point GPT-4o's vision at it, build a simple cataloging UI, and ship. The demand exists, the AI capability is ready, and the code is already open-source. The question is not whether this approach works, but which category you will build for first.</p>