Kun1009/qwen-image-edit-2511-lora-relight

Qwen-Image-Edit-2511 LoRA - Portrait Relight

This is a LoRA model trained on Qwen-Image-Edit-2511 for portrait relighting.

Usage

With fal.ai API

import fal_client

result = fal_client.subscribe(
    "fal-ai/qwen-image-edit-2511/lora",
    arguments={
        "prompt": "人像relight",
        "image_url": "YOUR_IMAGE_URL",
        "loras": [
            {
                "path": "https://huggingface.co/Kun1009/qwen-image-edit-2511-lora-relight/resolve/main/epoch-2.safetensors",
                "scale": 1.0
            }
        ]
    }
)

With DiffSynth-Studio

from diffsynth.pipelines.qwen_image import QwenImagePipeline, ModelConfig
import torch

pipe = QwenImagePipeline.from_pretrained(
    torch_dtype=torch.bfloat16,
    device="cuda",
    model_configs=[
        ModelConfig(model_id="Qwen/Qwen-Image-Edit-2511", origin_file_pattern="transformer/diffusion_pytorch_model*.safetensors"),
        ModelConfig(model_id="Qwen/Qwen-Image", origin_file_pattern="text_encoder/model*.safetensors"),
        ModelConfig(model_id="Qwen/Qwen-Image", origin_file_pattern="vae/diffusion_pytorch_model.safetensors"),
    ],
    processor_config=ModelConfig(model_id="Qwen/Qwen-Image-Edit", origin_file_pattern="processor/"),
)

# Load LoRA
pipe.load_lora(pipe.dit, "path/to/epoch-2.safetensors")

# Run inference
result = pipe(
    prompt="人像relight",
    input_image=[image],
    edit_image=[image],
    denoising_strength=1.0,
    num_inference_steps=30,
)

Training Details

• Base Model: Qwen/Qwen-Image-Edit-2511
• Training Task: Portrait Relighting (Flash Effect)
• Framework: DiffSynth-Studio

License

Please refer to the Qwen-Image-Edit-2511 model license.