Skip to main content

TryOnDiffusion Architecture

Detailed explanation of the dual UNet architecture.

Dual UNet Structure​

The model consists of two parallel UNets:

  1. Person UNet: Generates final output
  2. Garment UNet: Processes garment features

Key Components​

  • Cross-attention mechanisms
  • Self-attention with pose conditioning
  • FiLM layers for feature modulation
  • Attention pooling for pose embeddings

See TryOnDiffusion README for complete architecture details.