MirrorProof

• 

  MirrorProof makes the purchase gap visible: a YouCam-generated look, clearly labelled, followed by independent buying checks.

Inspiration

Virtual try-on helps shoppers visualise a look, but the generated image alone cannot prove physical fit, comfort, fabric behaviour, or return eligibility. MirrorProof adds an honest decision layer after a YouCam Clothes V3 preview so the shopper can act without hiding those unknowns.

What it does

MirrorProof accepts a shopper image, garment image, and garment category; maps them to YouCam Clothes V3; polls the asynchronous task; labels the generated result; and asks which visible concerns were resolved. Physical size, fabric information, and return-policy acceptance remain independent checks. The final receipt is a reversible buy, compare, or skip decision. A preview can look positive and still return compare until every external purchase fact is confirmed.

The public deployment runs in safe demo mode. The unchanged official YouCam sample inputs replay only their matching verified result, while other inputs receive a no-image explanation. This gives judges free access without exposing the API key or consuming the hackathon allocation.

How we built it

MirrorProof is a dependency-free Node.js 20 application. Its server validates HTTPS image URLs, keeps Bearer authentication private, sends only the documented Clothes V3 fields, captures the task identifier, and polls until success. The decision engine runs separately from the image generator so visual preference cannot silently become a fit-confidence score.

A real Clothes V3 run using Perfect Corp's official sample assets returned HTTP/API 200 and reached task_status: success after one poll. The repository includes a redacted evidence record, output hash, working safe replay, and twelve focused tests covering input mapping, URL safety, live polling, missing-key protection, decision behaviour, and secret redaction.

Challenges we ran into

• Proving a real API integration without publishing secrets or allowing anonymous visitors to spend units.
• Preserving uncertainty instead of inventing a confidence percentage.
• Making asynchronous processing understandable to a first-time shopper.
• Giving judges a complete, testable experience with no login or payment.

Accomplishments that we're proud of

• Verified a real YouCam Clothes V3 result and preserved independently inspectable evidence.
• Built a decision boundary that defaults to compare when independent purchase facts remain unresolved.
• Added explicit AI-generated-image and AI-assistance disclosure.
• Published the product, source, evidence, and a 2:45 end-to-end demo.

What we learned

A trustworthy virtual try-on should make the boundary between visual confidence and physical-product uncertainty visible. Integration evidence can also remain auditable without exposing credentials or enabling unlimited public API use.

What's next

We will test the receipt with small apparel retailers and frequent online returners, measure comparison and checkout-handoff rates, add retailer-configured size and return-policy evidence, and validate whether the decision layer reduces avoidable returns without reducing qualified purchases.

Built With

• javascript
• node.js
• vercel
• youcam-api
• youcam-clothes-v3