Modelpublic

chestnutlzj/Edit-R1-Qwen-Image-Edit-2509

sourceHugging Faceapache-2.0updated 11mo agoView on Hugging Face

47likes103downloads

Model Card

Edit-R1-Qwen-Image-Edit-2509: A model from UniWorld-V2

This model is a checkpoint (Edit-R1-Qwen-Image-Edit-2509) developed using the Edit-R1 framework, as presented in the paper Uniworld-V2: Reinforce Image Editing with Diffusion Negative-aware Finetuning and MLLM Implicit Feedback. The Edit-R1 framework focuses on reinforcing instruction-based image editing through Diffusion Negative-aware Finetuning (DiffusionNFT) and MLLM Implicit Feedback.

<a href="https://huggingface.co/papers/2510.16888">Paper</a> | <a href="https://github.com/PKU-YuanGroup/UniWorld-V2">Code</a> | <a href="https://github.com/PKU-YuanGroup/Edit-R1">Dataset</a>

Performance

Benchmark	Qwen-Image-Edit-2509	Edit-R1-Qwen-Image-Edit-2509
GEdit-Bench	7.54	7.76
ImgEdit	4.35	4.48

Usage

python

import os
import torch
from PIL import Image
from diffusers import QwenImageEditPlusPipeline

pipeline = QwenImageEditPlusPipeline.from_pretrained("Qwen/Qwen-Image-Edit-2509", torch_dtype=torch.bfloat16)
print("pipeline loaded")

pipeline.load_lora_weights(
    "chestnutlzj/Edit-R1-Qwen-Image-Edit-2509",
    adapter_name="lora",
)
pipeline.set_adapters(["lora"], adapter_weights=[1])

pipeline.to('cuda')
pipeline.set_progress_bar_config(disable=None)
image1 = Image.open("input1.png")
image2 = Image.open("input2.png")
prompt = "The magician bear is on the left, the alchemist bear is on the right, facing each other in the central park square."
inputs = {
    "image": [image1, image2],
    "prompt": prompt,
    "generator": torch.manual_seed(0),
    "true_cfg_scale": 4.0,
    "negative_prompt": " ",
    "num_inference_steps": 40,
    "guidance_scale": 1.0,
    "num_images_per_prompt": 1,
}
with torch.inference_mode():
    output = pipeline(**inputs)
    output_image = output.images[0]
    output_image.save("output_image_edit_plus.png")
    print("image saved at", os.path.abspath("output_image_edit_plus.png"))