|
|
--- |
|
|
license: apache-2.0 |
|
|
tags: |
|
|
- qwen |
|
|
- qwen-edit |
|
|
- qwen-edit-lora |
|
|
- try-on |
|
|
- lora |
|
|
base_model: |
|
|
- Qwen/Qwen-Image-Edit |
|
|
pipeline_tag: image-to-image |
|
|
library_name: diffusers |
|
|
--- |
|
|
|
|
|
## Qwen-Edit Try-On Lora (Alpha version) |
|
|
This is an alpha version Lora that is trained to perform **multi-reference try-on editing** with [Qwen Image Edit](https://huggingface.co/Qwen/Qwen-Image-Edit). |
|
|
|
|
|
- **Model type**: LoRA for Qwen Image Edit |
|
|
- **Use case**: Multi-reference clothing try-on |
|
|
- **Input format**: One top image (subject) + multiple bottom images (garments). For best results, use garment images that are high quality, and full body-shots for the subject. |
|
|
- **Output format**: Stylized subject wearing all garments from bottom row |
|
|
|
|
|
find us on Discord: https://discord.gg/UXN7zFuxbk |
|
|
|
|
|
# Examples: |
|
|
 |
|
|
 |
|
|
 |
|
|
|
|
|
|
|
|
## Usage |
|
|
For usage within ComfyUI using the [provided workflow](./TryOn-Alpha-Workflow.json). Recommended settings are 30 steps, with 2.5 CFG using Euler sampler and Simple scheduler. |
|
|
|
|
|
:warning: **Warning:** Both the 4-steps and 8-steps lightning lora’s did not perform well with the Try-On lora. I would recommend using it without low step lora's for now. |
|
|
|
|
|
832x1248 is the recommended resolution used. If you would like to use a different one please use one of the reccomended resolutions below provided by Qwen: |
|
|
|
|
|
|
|
|
(672, 1568), |
|
|
(688, 1504), |
|
|
(720, 1456), |
|
|
(752, 1392), |
|
|
(800, 1328), |
|
|
(832, 1248), |
|
|
(880, 1184), |
|
|
(944, 1104), |
|
|
(1024, 1024), |
|
|
(1104, 944), |
|
|
(1184, 880), |
|
|
(1248, 832), |
|
|
(1328, 800), |
|
|
(1392, 752), |
|
|
(1456, 720), |
|
|
(1504, 688), |
|
|
(1568, 672), |
|
|
|
|
|
|
|
|
--- |
|
|
|
|
|
|
|
|
### Limitations |
|
|
- Only trained/tested on multi-garment try-on tasks. |
|
|
- May not generalize to unrelated image editing tasks. |
|
|
- Requires careful prompt alignment (see below). |
|
|
- Struggles with 5 or more garments (more training required) |
|
|
|
|
|
--- |
|
|
|
|
|
## Example Prompts |
|
|
- **For Woman**: |
|
|
`"Style the woman in the top of the image, with every article of clothing on the bottom"` |
|
|
|
|
|
- **For Man**: |
|
|
`"Style the man in the top of the image, with every article of clothing on the bottom"` |
|
|
|
|
|
--- |
|
|
|
|
|
## Training Details |
|
|
- **Base Model**: Qwen Image Edit |
|
|
- **Approach**: LoRA training with [Ostris ai toolkit](https://github.com/ostris/ai-toolkit) |
|
|
|
|
|
--- |
|
|
|
|
|
|
|
|
## Donations |
|
|
We’re working toward a scaled-up V2 of this project, with expanded capabilities and larger dataset/compute requirements. Your support helps make this possible: |
|
|
- [Ko-fi](https://ko-fi.com/foxbaze) |
|
|
- **BTC Address**: `bc1qnqa7ufcw4f7cp8pzahd8jlut0ve2trq4c7t75h` |
|
|
- **ETH Address**: `0x596A4Bd554337d9D1FfbA6AA346f847CE942f945` |
