README.md

metadata

base_model:
  - FireRedTeam/FireRed-Image-Edit-1.0
language:
  - en
  - zh
library_name: diffusers
license: apache-2.0
pipeline_tag: image-to-image
tags:
  - FireRed-Image-Edit-1.0
  - distillation
  - LoRA

FireRed-Image-Edit-1.0-Lightning

This repository contains the distilled "Lightning" LoRA weights for FireRed-Image-Edit-1.0, enabling fast, high-quality instruction-based image editing in just 8 steps.

For more technical details, please refer to the FireRed-Image-Edit-1.0 Technical Report.

• Project Page: FireRed-Image-Edit-1.0 Space
• Repository: FireRedTeam/FireRed-Image-Edit
• Paper: FireRed-Image-Edit-1.0 Technical Report

Use with diffusers 🧨

To use this distilled LoRA, make sure to install diffusers from main:

pip install git+https://github.com/huggingface/diffusers.git

from diffusers import QwenImageEditPlusPipeline
import torch 
from PIL import Image

pipe = QwenImageEditPlusPipeline.from_pretrained(
    "FireRedTeam/FireRed-Image-Edit-1.0", torch_dtype=torch.bfloat16,
).to("cuda")

# Load the distilled Lightning LoRA weights
pipe.load_lora_weights(
    "FireRedTeam/FireRed-Image-Edit-1.0-Lightning", 
    weight_name="FireRed-Image-Edit-1.0-Lightning-8steps-v1.0.safetensors"
)

prompt = "在书本封面Python的下方，添加一行英文文字2nd Edition"
input_image_path = "./examples/edit_example.png"
input_image_raw = Image.open(input_image_path).convert('RGB')

image = pipe(
    image = [input_image_raw],
    prompt = prompt,
    height = None,
    width = None,
    num_inference_steps = 8,
    generator=torch.manual_seed(0),
    true_cfg_scale=1.0,  # Do not use standard CFG for Lightning LoRA
).images[0]

image.save("firered_image_edit_fewsteps.png")

Citation

If you find this work useful, please consider citing:

@article{firered2026rededit,
      title={FireRed-Image-Edit-1.0 Technical Report}, 
      author={Super Intelligence Team},
      year={2026},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2602.13344}, 
}