---
license: cc-by-nc-2.0
base_model:
- black-forest-labs/FLUX.1-Fill-dev
pipeline_tag: image-to-image
tags:
- tryon
- vto
- tryoff
---
# LoRA Model Card for CAT-Tryoff-Flux (ComfyUI)

This is a **LoRA adaptation** of the original [CAT-Tryoff-Flux](https://huggingface.co/xiaozaa/cat-tryoff-flux) virtual try-off model. It is optimized for use in **ComfyUI** and has been tested specifically with the **FLUX.1-Fill-dev** base model.

## Overview

The LoRA version enables efficient garment try-off capabilities using ComfyUI workflows. It reconstructs a clear front view of clothing from input images of people wearing the garments, closely following the methodology of the original CATVTON-FLUX.

- **Originally Developed by:** [X/Twitter: Black Magic An](https://x.com/MrsZaaa)
- **Repository:** [github](https://github.com/nftblackmagic/catvton-flux)

## ComfyUI Usage

This LoRA is designed to be used **only** in ComfyUI environments. It integrates seamlessly into custom workflows for try-off image generation.

The recommended setup includes:
- Base model: `FLUX.1-Fill-dev`
- LoRA applied: `CAT-Tryoff-Flux-LoRA`
- LoRA strength: Adjust as needed based on garment complexity and fit accuracy.

> ⚡ **Performance Note:** The reference workflow uses **Nunchaku nodes** to significantly improve processing speed. These can be replaced with standard ComfyUI nodes if Nunchaku is not available.

## Workflow and Setup

You can download and explore the full ComfyUI workflow here:

👉 [Download the CAT-Tryoff-Flux LoRA Workflow](https://huggingface.co/theunlikely/cat-tryoff-flux-lora/blob/main/workflow.json)

## Output Examples

Below are a few example outputs generated with the LoRA model. On the left is the input, on the right is the output:

### Example 1
<img src="results/1.png" alt="result 1" width="512"/>

---

### Example 2  
<img src="results/2.png" alt="result 2" width="512"/>

---

### Example 3  
<img src="results/3.png" alt="result 3" width="512"/>

---

## Notes

- This LoRA model is **experimental** and has only been tested with `FLUX.1-Fill-dev`.
- For best results, remove the background and use high-quality garment input images.
- Accuracy may vary based on lighting, occlusion, and clothing type.