Caterina Mandel
ZUM VORTRAG
Why try-on looks right and fits wrong - and what changed in 2026
Virtual try-on went mainstream in 2026. Zara Try-On passed 7 million sessions across 43 markets, Google moved try-on into Search. But people still don't use it to decide what to order.
That's because almost every try-on out there goes straight from two pictures to a third: a photo of you, a photo of the garment on a model, and a generated picture of you wearing the garment - without ever looking at actual measurements or fit. The garment sits on you the way it sat on the model, like it was made to measure, in every size you pick. Google Research and the University of Washington already said it in their FIT paper in April 2026: existing methods "fail to address a fundamental question for any user: how will this garment actually fit me?"
In 2026, the underlying technology moved far enough in three places to make it possible: how body measurements are calculated, how cloth can be simulated, and what diffusion models can be fine-tuned to do. I'll go through all three, what still doesn't work, and why the hardest part with a real brand is almost never the technology.
You'll leave able to tell a generated try-on from a simulated one, with three questions to ask any vendor, and knowing which data you need on your side to make it work.