File size: 2,422 Bytes
15e5b0a 4c2726e 15e5b0a 4c2726e 15e5b0a e390522 15e5b0a e390522 15e5b0a |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 |
---
license: apache-2.0
datasets:
- peteromallet/high-quality-midjouney-srefs
base_model:
- Qwen/Qwen-Image-Edit
tags:
- image
- editing
- lora
- style-transfer
- qwen
pipeline_tag: image-to-image
library_name: diffusers
---
# QwenEdit InStyle LoRA
## Model Description
**QwenEdit InStyle** is a LoRA fine-tune for QwenEdit that significantly improves its ability to generate images based on a style reference. While the base model has style transfer capabilities, it often misses the nuances of styles and can transplant unwanted details from the input image. This LoRA addresses these limitations to provide more accurate style-based image generation.
## How to Use
To get the best results, start your prompt with the phrase:
`Make an image in this style of `
And then describe what you want to generate.
For example:
`Make an image in this style of a serene mountain landscape at sunset.`
### use with diffusers
```
import torch
from diffusers import QwenImageEditPipeline
pipe = QwenImageEditPipeline.from_pretrained("Qwen/Qwen-Image-Edit", torch_dtype=torch.bfloat16)
pipe.to("cuda")
pipe.load_lora_weights("peteromallet/Qwen-Image-Edit-InStyle", weight_name="InStyle-0.5.safetensors")
```
### Strengths & Weaknesses
The model excels at:
- Capturing subtle style nuances from reference images
- Avoiding unwanted detail transplantation from source images
- Generating coherent images that match the intended style
- Strong prompt adherence
The model may struggle with:
- Extremely abstract or unconventional artistic styles
- Very specific technical details that conflict with the reference style
- Occasional anatomy issues
## Training Data
The QwenEdit InStyle LoRA was trained on a curated dataset of high-quality Midjourney style references. This dataset focuses on diverse artistic styles and provides clean style-content separation for better training.
You can find the public dataset used for training here:
[https://huggingface.co/datasets/peteromallet/high-quality-midjouney-srefs](https://huggingface.co/datasets/peteromallet/high-quality-midjouney-srefs)
## Links
- Model: [https://huggingface.co/peteromallet/Qwen-Image-Edit-InStyle](https://huggingface.co/peteromallet/Qwen-Image-Edit-InStyle)
- Dataset: [https://huggingface.co/datasets/peteromallet/high-quality-midjouney-srefs](https://huggingface.co/datasets/peteromallet/high-quality-midjouney-srefs) |