Steal the look

Inspiration

Discovery happens on video now. You see an outfit in a reel or a makeup look on TikTok and think "would that actually work on me, and where do I buy it?" Then nothing happens. Existing tools make you screenshot a still and upload it, so the video, the moment, and the "on me" are all lost. Perfect Corp's YouCam APIs are the missing piece: they render an outfit or a full makeup look onto your own photo, photorealistically. We built the bridge from inspiration, to "on my face," to "in my cart," across both fashion and beauty, on one backend.

What it does

Two experiences, both powered by YouCam Virtual Try-On, both shoppable.

1. StealTheLook (fashion). Drop a screen-recorded reel or short (a screenshot works too). We read the frames, pick the clearest view of each garment, and use it as the reference for YouCam AI Clothes Virtual Try-On, so the outfit renders on your photo. Tap any garment to try it on and shop it, with a tagged Amazon link for every piece. It also runs as a real Android app: a floating chip rides over Instagram, and one tap captures the reel you are watching and shows a shoppable "Shop this look" panel.

2. Makeup Atlas (beauty). Add your face once, by uploading a photo or using your camera. Swipe the globe and your same face wears each country's trending makeup, rendered by YouCam AI Makeup Virtual Try-On (real lip and eye color composited on, identity preserved). France, Spain, South Korea, Brazil and more, each with its trending look described and its products linked on Amazon. Looks are tailored to your age range.

How we built it

Next.js and TypeScript on Render, with one API backend serving both experiences, plus a native Android client in Kotlin.

• Video to best frame: a reel is a moving target, so instead of taking one arbitrary still we work out which moment shows each garment most clearly, and use that frame for both the try-on and the shopping image. Render quality tracks that choice more than anything else in the pipeline.
• YouCam Apparel Try-On: the garment reference is a crop from the user's own reel rather than a retailer's catalogue photo, which is the non-obvious part - most try-on runs against a fixed product library, and we point it at the open web.
• YouCam Makeup Try-On: each country's signature look is mapped to real makeup effects, so only the makeup changes and identity is preserved.
• Android overlay: a floating chip draws over whatever app you are in and reads the current screen only when you tap it. We draw only our own views and never read or inject into other apps, with a persistent capture notice for trust.
• Shopping: what the detector describes becomes tagged, shoppable links, with a fallback chain for finding specific listings.
• Privacy and safety by design: detection describes clothing only, never the person. No face or body analysis of anyone in a reel. Selfies are processed per call and never stored on our servers. Makeup Atlas is 13 and up, collects an age range only, and lets you clear your data anytime.

Challenges we ran into

• Undocumented API contracts. The task shapes we needed differ from the published examples, so we mapped them empirically rather than guessing - and did the whole discovery pass without spending a single render credit.
• Video is hostile input. A still from a moving clip is a lottery, so best-frame-per-garment selection is not a nicety, it directly determines render quality. We tested against a clip where most frames were blurred or occluded and the model still picked the clean one per item.
• Capturing the right instant is harder than it sounds. Edge cases around when a frame is actually fresh cost us a day; a moving reel never trips them, but we handled them anyway.
• Cost discipline. Every render costs real units, so we added per-country caching, swipe debouncing, a session cap, and per-IP rate limits.

Accomplishments that we're proud of

• Both of Perfect Corp's flagship try-on APIs, apparel and makeup, working live in one product, with identity preserved and results that look real.
• A video-native flow (best-frame selection) that goes a step beyond screenshot-and-upload tools.
• A real Android overlay that shops the reel you are already watching, proven on live Instagram.
• We deliberately did not build the creepy version. No analyzing people's faces or bodies from their reels. It is about the clothes and the makeup, never judging a person.
• Shipped, hardened, and live: security headers, input validation, rate limits, sanitized errors, and no exposed secrets.

What we learned

Virtual try-on is not a feature, it is a conversion mechanic. The distance from inspiration to purchase is mostly "will it look good on me," and rendering it on the actual shopper collapses that distance. Perfect Corp sells this to retailers, and it works just as well pointed at the open web, across both fashion and beauty.

What's next

• Unify the two: one face wearing both the outfit and the makeup from a single look, shopped as one list.
• Live per-country and per-age beauty-trend data, and richer makeup textures.
• Sustainability: accounts with credit packs (built) plus affiliate revenue (live) so heavy use pays for itself.
• iOS via a ReplayKit broadcast session and a share-sheet path.

Built With

• ai-clothes-virtual-try-on
• ai-makeup-virtual-try-on
• amazon-associates
• android
• kotlin
• next.js
• perfect-corp
• react
• typescript
• youcam-api