Week 23: Building a "Clueless" AI Wardrobe Assistant

The Experiment

I had this vision of creating my own version of the iconic computerized wardrobe from the movie "Clueless"; the one where Cher digitally tries on outfits before school? I wanted to be able to input parameters like weather and occasion, and have AI suggest outfit combinations from my actual wardrobe.

The first step was obvious: I needed to catalog what I already owned before any kind of magical outfit generation could happen. My goal was to mix up my style and see if AI could suggest creative combinations I wouldn't think of myself. I'm a fan of the Blackbird Spyplane newsletter and Japanese Cityboy aesthetic, so I was curious if AI could work within those style parameters with my existing clothes.

Could AI help me wear more of my clothes and look better wearing them? Let the adventure begin 🎒

The Process

Here's how I tackled it:

1. Photographing My Wardrobe

• Took flat-lay (i.e. put them on the floor) photos of 77 clothing items and shoes

2. First Attempt: Gemini

• Uploaded photos to a Google Drive folder

• Asked Gemini to classify my wardrobe items

• Hit immediate roadblocks - Gemini essentially refused to engage with the task as I wanted

3.Pivoting to ChatGPT

• Transferred my approach to ChatGPT

• Created a project where I could upload images in batches of about 10 items

• Outlined the classification system I wanted: type, color, warmth, formality, style vibe, season, and notes

• Told ChatGPT about my fashion influences (Blackbird Spyplane newsletter, Japanese Cityboy aesthetic)

4. The Classification Process

• ChatGPT analyzed each batch of clothing items

• It created detailed entries with impressive attention to detail:

• Type (top, bottom, etc.)

• Color

• Warmth (light, medium, heavy)

• Formality (casual, smart casual, formal)

• Style Vibe (minimalist, heritage, outdoors)

• Season

• Additional notes on styling

5. Memory Issues Emerge

• Worked through all 77 items across multiple batches

• Initially seemed to be working well

• When I asked for the complete database export, discovered major gaps

• Found that shoes and jackets had been classified during our session but mysteriously vanished from the final export

• ChatGPT had essentially hallucinated having these items in the final database

The Outcome

This experiment landed somewhere between success and failure. On the positive side:

• ChatGPT did an excellent job with the actual classification when viewing individual images

• The style categorization was accurate and useful

• The process itself gave me better insights into my wardrobe than I expected

On the negative side:

• Memory issues with ChatGPT meant the final database was incomplete

• Items classified early in the process disappeared from the final export

• When I asked for wardrobe recommendations, ChatGPT suggested I buy items I already owned (but it had forgotten about)

The most valuable lesson came from the process itself - simply photographing and reviewing my wardrobe gave me enough insights that I began questioning whether I actually needed an AI to suggest outfits at all.

Key Takeaway

When working with AI on multi-stage projects involving many items, export your data after each batch. Don't trust the AI's memory to maintain the full dataset, even within a single project or conversation.

Additionally, sometimes the process of systematically reviewing something (like your wardrobe) provides enough value on its own, even if the AI component isn't fully successful.

Pro Tips for Beginners:

1. Export As You Go: Don't wait until the end of a long session to save your data - export after each batch is processed.

2. Consider Notion Integration: If I were to redo this, I'd use Notion to store each item alongside its classification for easier visual reference.

3. Start Small: Test your approach with a small subset before committing to photographing your entire wardrobe.

4. Be Realistic About Goals: Sometimes simply going through the process of cataloging provides enough insight, without needing the AI to make complex recommendations.

Want to Try It Yourself?

• Start with a smaller category (just shirts or pants) to test the process

• Photograph items with consistent lighting on a neutral background

• Process in batches of 10 or fewer items

• Use a structured storage system like Notion or a spreadsheet

• Save your data after each batch to avoid memory issues

I may return to this project eventually, with the goal of creating a simple GPT that could say "The weather is cool and rainy, and you need to be slightly formal today" and then suggest a complete outfit. But for now, I've realized that simply going through the cataloging process gave me most of the wardrobe insights I was looking for.