# Photoswap: Personalized Subject Swapping in Images

**Authors:** Jing Gu, Yilin Wang, Nanxuan Zhao, Tsu-Jui Fu, Wei Xiong, Qing Liu, Zhifei Zhang, He Zhang, Jianming Zhang, HyunJoon Jung, Xin Eric Wang  
**arXiv:** [2305.18286](https://arxiv.org/abs/2305.18286)  
**Date:** 2023-05-29  
**Summary page:** https://www.summarizepaper.com/en/paper/2305.18286v1/

## Key points

- Photoswap enables users to personalize their visual content by seamlessly substituting one subject with another
- The technique works by learning the visual concept of the subject from reference images and swapping it into the target image using pre-trained diffusion models in a training-free manner
- Photoswap maintains the pose of the swapped subject and overall coherence of the image, ensuring a well-conceptualized visual subject can be transferred to any image with appropriate self-attention and cross-attention manipulation
- Experiments have shown that Photoswap is highly effective and controllable in personalized subject swapping as well as significantly outperforming baseline methods in human ratings across subject swapping, background preservation, and overall quality
- Photoswap has potential applications ranging from entertainment to professional editing.

## Summary

Photoswap is a novel approach that enables users to manipulate and personalize their visual content by seamlessly substituting one subject with another. In today's digital landscape where images are dominant, the ability to do this has become essential. With Photoswap, users can replace a tabby cat lounging on a sunlit window sill with their own playful puppy while preserving the original charm and composition of the image. The technique works by first learning the visual concept of the subject from reference images and then swapping it into the target image using pre-trained diffusion models in a training-free manner. This ensures that a well-conceptualized visual subject can be transferred to any image with appropriate self-attention and cross-attention manipulation while maintaining the pose of the swapped subject and overall coherence of the image. Experiments have shown that Photoswap is highly effective and controllable in personalized subject swapping as well as significantly outperforming baseline methods in human ratings across subject swapping, background preservation, and overall quality. This reveals its potential for applications ranging from entertainment to professional editing. The authors of Photoswap are Jing Gu, Yilin Wang, Nanxuan Zhao, Tsu-Jui Fu, Wei Xiong, Qing Liu, Zhifei Zhang, He Zhang, Jianming Zhang HyunJoon Jung and Xin Eric Wang who present an innovative solution for personalizing visual content by allowing users to swap subjects effortlessly while maintaining image integrity.