AdClo
Adaptive clothing that works the way you do.
Tell AdClo what makes dressing hard (limited grip, one hand, a carer helping, sensory needs, a brace,
a medical device or sitting all day). It ranks real adaptive clothes from IZ Adaptive, Silverts and
BILLY Footwear for you, shows them on your own photo with YouCam AI Clothes Virtual Try-On, and an
AI stylist you can talk to builds whole outfits and talks back. Every Buy link opens that exact product.
| Your photo | Tried on with YouCam (IZ Adaptive open-back polo) |
|---|
| |
About 1.3 billion people live with a significant disability (WHO): limited grip, using one hand,
needing help to dress, sensory sensitivity, a prosthetic or brace, a medical device, or sitting all day.
Most shopping filters only cover size and colour. AdClo starts from the problem
("buttons are hard for my hands", "I sit all day") and ranks garments by whether they solve it,
then uses YouCam AI Clothes Virtual Try-On to show the piece on the shopper's own photo.
Screenshots
1. Fit profile. The shopper ticks what makes dressing hard. Everything else is ranked from this.
2. Ranked with reasons. Every product page says why it works for you, and marks your usual size.
3. Your look. One piece per slot, tried on together, each bought from its own store.
4. Full-outfit try-on with YouCam. Bottoms, top and shoes are tried on in a chain. The chinos are
fitted to Marcus's real body, so the try-on never adds a leg he doesn't have.
5. Works on phones. Grace's shop with one need ticked: the magnetic blouse ranks first at 100% fit.
How it works
AdClo is a shopping app built around a fit profile: what makes dressing hard (sitting all day,
limited grip, one-handed dressing, assisted dressing, sensory needs, a prosthetic or brace, medical
access), who you shop for, your usual sizes and a photo of you.
- Shop by category: tops, bottoms, jackets, shoes and dresses. Every piece is ranked by how well
it fits your needs, with plain reasons ("No back pockets to press into your skin") and warnings.
- Product page: sizes (your usual size is marked), why it works for you, Try it on me, and
Complete the look, which suggests matching pieces from other categories with a reason
("Charcoal lets the sky blue stand out · both smart").
- AI stylist: type or say what you need ("my cousin's wedding is Saturday, I dress with one hand,
nothing bright, about $200"). AdClo shows what it understood, then builds an outfit and explains each
pick. Voice input helps people who find typing hard, and the stylist answers out loud in a natural voice.
- Quick style: pick an occasion and a budget, no AI needed.
- Your look: the pieces you picked, one per slot. Try them all on together, then buy each one.
- Full-outfit try-on: try-ons are chained. Bottoms first, then top, jacket and shoes, each step
using the previous result as the photo. Your own shoes are kept unless you try on new ones.
How the AI stylist stays trustworthy
- Understand (Groq,
openai/gpt-oss-120b): the request becomes a structured plan (occasion, budget,
needs, colours, specific pieces). Unknown values are thrown away.
- Shortlist (AdClo's own rules): only real catalog pieces that meet the person's needs, avoid
their colours and fit the budget. A tight budget drops the jacket or shoes before the essentials.
- Choose and explain (Groq): the AI picks from the shortlist and writes why. Picks outside the
shortlist or over budget are replaced by the rules-based choice, so it cannot invent clothes.
The prompts require person-first language, and a filter replaces phrases like "wheelchair-bound"
in anything the AI writes. Voice input uses Groq's whisper-large-v3-turbo.
The try-on never invents a body
Generative try-on draws trousers on two full legs, so on a photo of an amputee it adds a leg that isn't
there. For shoppers who tick the limb-difference need, AdClo keeps the try-on true to their real body
(public/keep-body.js):
- Shoes are tried on first, while the photo still shows the person's own legs.
- MediaPipe image segmentation finds the person's outline in their original photo. The model sees
256×256, so it also runs on overlapping tiles for sharp edges around legs and feet.
- The YouCam result is kept only inside that outline, and the original photo is put back everywhere else.
- If segmentation can't run on the device, trousers and dresses are left off the try-on and the app
says why. Tops, jackets and shoes still try on, and the fit ranking still covers every piece.
This runs in the browser and only for shoppers who tick the limb-difference need.
The stylist talks back
Alongside the written answer, the AI writes a short, casual spoken reply, the way a friend would talk
you through the outfit, not a read-out of the page. It's spoken with Groq's
canopylabs/orpheus-v1-english (six voices to pick from). It plays on its own when "Talk back to me"
is on, and there's a Listen / Stop button on every answer. Replies are split into sentences under 200
characters (the model's limit) and played back to back. The Groq org admin must accept the Orpheus
terms once in the Groq console before voice output works.
Core logic lives in lib/: agent.js (AI stylist), groq.js (AI client), match.js (needs to features), stylist.js (pairing and outfits),
outfit.js (chained try-ons) and youcam.js (API client).
YouCam API used
AdClo uses YouCam AI Clothes Virtual Try-On (cloth-v4) through the server in server.js, so the
API key never reaches the browser.
| Step | Endpoint |
|---|
| Upload image | POST /s2s/v2.0/file/..., then PUT to the signed URL |
| Start try-on | POST /s2s/v2.0/task/cloth-v4 with garment_category upper_body, lower_body, outer, shoes or full_body. Later steps pass the previous result as src_file_url. change_shoes: false keeps the person's own shoes. |
| Get result | GET /s2s/v2.0/task/cloth-v4/{task_id} |
- Speed: about 20 seconds per garment, so a three-piece look takes about a minute. The app shows
progress for each step and gives up with a plain-language message after 2 minutes.
- Tested on: seated and standing photos, including a person in a wheelchair. Tops, trousers, shoes
and chained outfits all work, with the face, hands and any mobility aid left unchanged.
- Serverless: try-on is stateless (the browser holds the task id and polls), so it runs on Vercel.
Run it
Needs Node 20 or later.
npm install
cp .env.example .env # add YOUCAM_API_KEY and GROQ_API_KEY
npm start # http://localhost:3000
Without an API key the app runs in demo mode: everything works, but the "try-on" returns your own
photo, so you can build the UI without spending API units. Without a Groq key, the AI stylist is
hidden and quick style still works. The live app at https://adclo.vercel.app runs with both keys.
Catalog
data/catalog.json lists real products from adaptive brands (IZ Adaptive, Silverts, BILLY Footwear).
Each Buy button opens that exact product on the brand's own store. Photos in public/garments/ are
the brand's product shots of the garment alone (no model). Prices are shown in each store's currency
(IZ Adaptive: CAD, BILLY: USD), so outfit totals are approximate.
To add a product, use a front-facing photo of the garment alone. Items without a photo still appear,
but their try-on button is disabled. The logic tests use a separate sample catalog in test/fixtures/.
Tests
npm test
44 tests cover needs matching, outfit building, the stylist guardrails and the YouCam client.
Accessibility
The app works with a keyboard and a screen reader, uses large touch targets (48px or more), supports
dark mode and reduced motion, announces try-on progress through a live region, and explains errors in
plain language.
Trademarks and product images
- AdClo is an independent hackathon project. It is not affiliated with, sponsored by or endorsed by
IZ Adaptive, Silverts or BILLY Footwear.
- Brand and product names are used only to identify the real products the app recommends and links to.
- Product photos in
public/garments/ belong to their brands. They are used only to show each product
and link to its page on the brand's own store. No product is sold through AdClo, and AdClo takes no
commission. Each photo's source page is listed in CREDITS.md.
- Any brand can ask for its images to be removed, and we will remove them straight away.
- People shown in the demo ("Marcus" and "Grace") are Pexels stock photos used under the Pexels License.
The background music in the demo video is an original composition, synthesized in code and released CC0.
Hackathon notes
Built for the YouCam API Skin AI & eCommerce VTO Hackathon (Devpost, 2026) by Victor Kiprotich
(@chocle254), an individual entrant based in Kenya.
- Eligibility: the submitter is over 18 and of legal age in Kenya.
- If selected as a winner: happy to take part in the exit interview and to be featured in a Perfect
Corp blog article.
- Built during the hackathon: the first commit is on 24 October 2026.
License
MIT for the code. Product photos, brand names and stock photos are not covered by this license; see
Trademarks and product images.
