AMATI
An online store for Italian dresses, bags and shoes with an AI photo studio and three languages.
- Client
- AMATI, an Italian fashion boutique
- What we did
- Design, store, admin, AI photo studio, three languages
- Year
- 2026, launch in preparation
- Technology
- Next.js, Payload CMS, Postgres, BiRefNet

The task
The boutique brings dresses, bags and shoes from Italy, often only a few of each. The photos were shot every which way: on a mannequin, on a hanger, on the floor.
It needed a storefront that feels like a boutique rather than a marketplace, and an admin where the owner runs the catalogue in three languages herself.
Above all, any photo from a phone had to look as if it came from one studio.
Storefront and catalogue
The home page works like a boutique window: one piece across the screen, “New from Italy” and a look builder where you flip the dress, bag and shoes separately and see the whole look.
- Filters by size, colour, price, availability and discount
- A Dress · Bag · Shoes look builder and “Add the look to cart”
- Cart, checkout and a message to the manager in Telegram
- Booking a fitting at the boutique




AI photo studio
The owner uploads any photos in a batch. The system removes the background, centres the piece in a 4:5 frame with equal margins and scale, and makes two versions: one for the site and one for social media.
- The open BiRefNet model on the store’s own server: about 8–10 seconds per photo with no monthly fee
- For the site, the piece without a background on branded sandy fields; for social media, on a background with a soft shadow
- Close-ups and photos on a model are recognised and only cropped, not cut out
- The original is kept and any photo can be reprocessed


Admin and speed
The owner runs the catalogue, copy and translations herself in Payload CMS. The site serves pages from cache, so it handles thousands of products and loads fast even on mobile data.
- Ukrainian, Russian and English in every admin field
- Page caching: saving a product refreshes only the related pages
- WebP photos in three sizes and Product markup for Google
- A Docker server with HTTPS and nightly backups of the database and photos for 14 days


In numbers
3
interface and admin languages
≈10
seconds to process one photo
2
versions of every photo: site and social
14
days of nightly backups kept