DiffSynth-Studio/Qwen-Image-EliGen-V2

Qwen-Image Precise Region Control Model

Model Introduction

This model is the V2 version of a precise region control model trained based on Qwen-Image. The model architecture uses LoRA, enabling control over the position and shape of each entity by providing textual descriptions and regional conditions (mask maps) for each entity. The training framework is built on DiffSynth-Studio, and the dataset used is the Qwen-Image-Self-Generated-Dataset.

Compared to the V1 version, this model is trained on a self-generated dataset from Qwen-Image, resulting in generated images whose styles are more consistent with the base model.

Result Demonstration

Entity Control Condition	Generated Image 1	Generated Image 2	Generated Image 3

Inference Code

git clone https://github.com/modelscope/DiffSynth-Studio.git  
cd DiffSynth-Studio
pip install -e .

from diffsynth.pipelines.qwen_image import QwenImagePipeline, ModelConfig
from modelscope import dataset_snapshot_download, snapshot_download
import torch
from PIL import Image

pipe = QwenImagePipeline.from_pretrained(
    torch_dtype=torch.bfloat16,
    device="cuda",
    model_configs=[
        ModelConfig(model_id="Qwen/Qwen-Image", 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"),
    ],
    tokenizer_config=ModelConfig(model_id="Qwen/Qwen-Image", origin_file_pattern="tokenizer/"),
)

snapshot_download("DiffSynth-Studio/Qwen-Image-EliGen-V2", local_dir="models/DiffSynth-Studio/Qwen-Image-EliGen-V2", allow_file_pattern="model.safetensors")
pipe.load_lora(pipe.dit, "models/DiffSynth-Studio/Qwen-Image-EliGen-V2/model.safetensors")

global_prompt = "Poster for the Qwen-Image-EliGen Magic Café, featuring two magical coffees—one emitting flames and the other emitting ice spikes—against a light blue misty background. The poster includes text: 'Qwen-Image-EliGen Magic Café' and 'New Product Launch'"
entity_prompts = ["A red magic coffee with flames rising from the cup", 
                  "A red magic coffee surrounded by ice spikes", 
                  "Text: 'New Product Launch'", 
                  "Text: 'Qwen-Image-EliGen Magic Café'"]

dataset_snapshot_download(dataset_id="DiffSynth-Studio/examples_in_diffsynth", local_dir="./", allow_file_pattern=f"data/examples/eligen/qwen-image/example_6/*.png")
masks = [Image.open(f"./data/examples/eligen/qwen-image/example_6/{i}.png").convert('RGB').resize((1328, 1328)) for i in range(len(entity_prompts))]

image = pipe(
    prompt=global_prompt,
    seed=0,
    eligen_entity_prompts=entity_prompts,
    eligen_entity_masks=masks,
)
image.save("image.jpg")

Citation

If you find our work helpful, please consider citing our research:

@article{zhang2025eligen,
  title={Eligen: Entity-level Controlled Image Generation with Regional Attention},
  author={Zhang, Hong and Duan, Zhongjie and Wang, Xingjun and Chen, Yingda and Zhang, Yu},
  journal={arXiv preprint arXiv:2501.01097},
  year={2025}
}