Skip to content

TryOffDiff

Rebuild a standard garment image from one photo

Visit official site

What TryOffDiff is

The model is described as a diffusion-based, high-fidelity garment reconstruction technique that produces a standardized garment image from a single photo of a dressed person. It differs from virtual try-on by aiming to extract a canonical garment view.

What you can do with it

  • Extract a canonical garment image from one photo
  • Keep the shape, texture and complex patterns of the garment
  • Use Stable Diffusion for garment image generation
  • Apply SigLIP-based visual conditioning to the reconstruction

Model notes

  • The material reports experiments on the VITON-HD dataset
  • The approach is compared with pose-transfer and virtual try-on baselines
  • Fewer pre-processing and post-processing steps are described

Pros & cons

✓ What we like

  • Only one photo of the wearer is required as input
  • The canonical image keeps garment detail

! What to watch out for

  • The task of capturing shape, texture and complex patterns is described as a distinct challenge
  • The method is framed around canonical garment images rather than virtual try-on

FAQ

What does it produce?

A standardized canonical garment image from a single photo of a person wearing it.

Which models are used?

Stable Diffusion with SigLIP-based visual conditioning.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review