Photoswap: Learn subjects for personalized image swapping

Understand the project

Practical overview

Use it when developing or studying a subject-specific image-swapping workflow and you can prepare the concept-learning model and checkpoint.

Learn a subject as a token in a diffusion model.
Use a trained diffusion-model checkpoint in an attention-based subject-swapping workflow.
Explore concept-learning approaches such as Text Inversion, DreamBooth and Custom Diffusion.

Before you choose

The README directs users to contact the project for Photoswap checkpoints.

The README recommends more source images for a better learned concept and swap result.
DreamBooth encoder fine-tuning is described as potentially improving results, especially for human faces, with higher memory requirements.
The README calls for installing the package requirements and placing the trained checkpoint in the checkpoints folder; the supplied instructions do not show the remaining execution steps.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Use a trained diffusion-model checkpoint with Photoswap’s attention-swap notebook to swap a subject in images.

A trained diffusion-model checkpoint; the README says to contact jgu110@ucsc.edu for Photoswap checkpoints.
A single GPU with 16 GB of memory.
The repository’s requirements.txt.

Place the trained diffusion-model checkpoint in the checkpoints folder.
Install the packages from requirements.txt with the command below. The README notes that the concept-learning environment is not suitable for attention swap.
Documented command
```
pip install -r requirements.txt
```
Run real-image-swap.ipynb for subject swapping.
Tune the swapping step and text prompt for the concept being used.

The notebook performs subject swapping using the trained checkpoint; tuning the swap step and prompt can improve performance.

Different learned concepts may require different swap steps.
A concept model with tuned weights may be less capable of general concept generation.
The README specifies a single GPU with 16 GB of memory.

Evidence and transparency

Source record details

Source availability: Active
Sources checked: 18 October 2026
Original website title: GitHub - UCSB-AI/photoswap: Official implementation of the NeurIPS 2023 paper "Photoswap: Personalized Subject Swapping in Images" · GitHub · Official homepage title · retained in its source language