Model description
license: apache-2.0
tags:
- image-to-image
- style-transfer
- anime
- qwen
- lora
- image-restoration
library_name: diffusers # 如果适用
Qwen-Anime-Transformer-LoRA
This is a LoRA trained on the Qwen-image-Edit 2511 base model. It is designed for seamless anime style transfer and image restoration.
Model Description
The Qwen-Anime-Transformer allows users to convert real-world imagery or diverse art styles into a refined anime aesthetic with a single instruction. Beyond transformation, it serves as a powerful "Restoration Layer" for existing anime images, cleaning up noise, sharpening line art, and enhancing overall visual fidelity.
- Developed by: [Your Name/Organization]
- Model type: Image-to-Image / Style Transfer LoRA
- Base Model: Qwen-image-Edit 2511
- Trigger Word: `transform into anime style`
Key Capabilities
- One-Click Stylization: Transform photos into high-quality anime scenes.
- Semantic Consistency: Leverages Qwen's vision-language capabilities to ensure objects, poses, and backgrounds remain contextually accurate after transformation.
- Artifact Correction: Acts as a refiner for low-quality or AI-generated anime images, fixing structural inconsistencies.
Usage
Recommended Settings
- Prompt: `transform into anime style`
- LoRA Weight: `0.6 - 0.9`
- Inference Steps: 20-30 (depending on the base model configuration)
Example Instruction
> Input Image: [Upload your image]
> Prompt: "transform into anime style"
Technical Specifications
This LoRA fine-tunes the cross-attention layers of the Qwen-image-Edit architecture. It balances the high-dimensional features of the Qwen2.5-VL encoder with the stylistic flexibility required for stylized output.
Training Details
- Training Framework: [e.g., Swift / PEFT]
- Dataset: Curated high-resolution anime frames and paired real-world/anime datasets.
- Precision: BF16/FP16
Limitations and Bias
While this model excels at general anime styles, it may inherit biases from the training data regarding specific character archetypes. Users are encouraged to adjust the prompt to fine-tune specific character features.
Citation
If you find this LoRA useful in your research or projects, please cite:
```bibtex
@misc{qwen-anime-lora-2024,
author = {Your Name},
title = {Qwen-Anime-Transformer-LoRA},
year = {2024},
publisher = {Hugging Face},
journal = {Hugging Face Model Hub}
}
Download model
Download them in the Files & versions tab.
