Conclusion: bags break try-on in ways clothing doesn't. A bag has a true size, more than one way to be carried, and listings whose tags often say the wrong thing. Working with a luxury handbag reseller, Ello shipped a shopper height slider that draws a bag at its real size (median error 11–14% from true size, against 28% without it), every way a bag is carried from one try-on, hands that rest on a doubled chain bag the way people hold one (34 of 36 test views, up from 3), and a page where a store sees and fixes exactly what the try-on reads from each listing. All numbers are Ello's own test data; every change is on the public changelog.
A bag has a true size (a mini and a medium flap look identical in a cropped product photo), more than one way to be carried, and a listing that often disagrees with the bag, such as a Crossbody tag on a bag whose chain only reaches the shoulder.
| Problem | What we changed | Result in our tests |
|---|---|---|
| Every bag drew at roughly the same size on every shopper | Height slider; bag drawn from its listed measurements and the shopper's height | Median size error 11–14%, against 28% without height |
| Bags worn three or more ways showed only one or two | Every way a bag is carried, from a single try-on | Shoppers switch between views on the product photo |
| On doubled chain bags the arm hung behind the bag | Elbow bends, hand rests on top near the chain | Hand on top in 34 of 36 views, up from 3 of 36 |
| A chain bag on the shoulder could hang outside the arm | Chain runs down the inside of the arm, bag sits under it | Tucked under the arm in 7 of 7 test renders |
| Real shoppers' photos rejected as "no body detected" | A detector that looks for a person and never removes a photo | 0 real people missed in ~500 photos, against 1 in 5 before |
| Stores couldn't see why a bag rendered the way it did | What Ello reads page: see and fix every field the try-on uses | Matched the rendered views on 1,350 bags across two catalogs |
Two numbers: the bag's measurements, read from its listing, and the shopper's height, from a tape-measure slider in feet and inches with centimeters. The same bag then takes up less of a 6 foot shopper than a 5 foot one. The height is used for that try-on only and isn't saved by Ello.
A bag buyer reads the pose before the bag. A doubled chain bag with the arm hanging behind it looks pasted on. Shoulder carries now rest the hand where the strap meets the bag, and when one of two views comes out wrong, Ello redraws just that view instead of dropping it.
Yes, when it knows the bag's measurements and the shopper's height. In Ello's tests on a luxury reseller's bags, median size error was 11–14% with height, against 28% without (our data).
No. It's used for that try-on only and stays on the shopper's own device.
Yes. Crossbody, shoulder and handheld can all come from a single try-on; the shopper switches views on the product photo.