Credits
VTryItOn doesn't train or own any of the models it uses. Generation runs on a Hugging Face Space built on the open-source models below — listed as used by that Space at the time of writing.
Models
- LeffaMITby Zijian Zhou et al.
The core virtual try-on and pose-transfer model that generates the result image.
- Stable Diffusion InpaintingCreativeML Open RAIL-Mby Runway
Diffusion base model Leffa builds its inpainting on.
- by Peike Li et al.
Generates the human-parsing masks (ATR/LIP) used to isolate body regions.
- OpenPose (body pose estimation)See upstream projectby CMU Perceptual Computing Lab / community re-implementations
Estimates body keypoints used to align the garment to your pose.
- DensePoseApache 2.0by Meta AI Research (Detectron2)
Generates dense body-surface correspondence used by the demo pipeline.
- by sayeed99 (fine-tune of NVIDIA SegFormer)
Optional garment-region parser used during preprocessing.
Built with
- Diffusers & TransformersApache 2.0by Hugging Face
Model-running code foundation used by the Space.
- GradioApache 2.0by Hugging Face
Powers the Space's API and this site's @gradio/client connection to it.
- MediaPipe Tasks Vision (Pose Landmarker)Apache 2.0by Google
Runs entirely in your browser to check your photo shows a full body before submitting.
- by Hugging Face
Client-side "Sign in with Hugging Face" OAuth helper.
Leffa (MIT License)
Copyright (c) 2024 Zijian Zhou. Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files, to deal in the software without restriction, subject to the inclusion of the above copyright notice in all copies. See the full license text at github.com/franciszzj/Leffa/blob/main/LICENSE.