Skip to content

IMAGDressing

A modular research model for controllable virtual clothing generation

Visit official site

What IMAGDressing is

The repository presents IMAGDressing as an interactive, modular clothing generation model intended to give virtual try-on systems flexible and controllable customisation. The described architecture combines CLIP semantic features with VAE texture features via a hybrid attention module inside a denoising UNet, which the authors link to user-controlled editing. The project also introduces IGPair, a dataset of more than 300,000 clothing and worn-image pairs with a documented assembly process.

What you can do with it

  • Generate garment effect images in different styles
  • Build an online try-on feature in a fitting application
  • Study clothing generation and virtual try-on methods

Who it is for

  • Garment designers producing effect images, per the source scenarios
  • Try-on app developers implementing online fitting
  • Researchers studying clothing generation and virtual try-on

What to watch out for

  • Setup requires Python with PyTorch 2.0+ and CUDA 11.8
  • Region-specific garment replacement is labelled experimental

Pros & cons

✓ What we like

  • The architecture and the feature sources are described explicitly
  • A large paired dataset with a documented assembly process is published
  • Plugin compatibility is named concretely

! What to watch out for

  • Local Python, PyTorch and CUDA setup is required
  • No hosted service, price or quota is published
  • Region-specific replacement is described as experimental

FAQ

What is IGPair?

The repository describes a dataset of over 300,000 clothing and worn-image pairs.

Which plugins are named?

ControlNet, IP-Adapter, T2I-Adapter and AnimateDiff.

Is LoRA training required?

The source describes fast customisation without extra LoRA training.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review