---
frameworks:
- Pytorch

tasks:
- text-to-image-synthesis

#model-type:
##如 gpt、phi、llama、chatglm、baichuan 等
#- gpt

#domain:
##如 nlp、cv、audio、multi-modal
#- nlp

#language:
##语言代码列表 https://help.aliyun.com/document_detail/215387.html?spm=a2c4g.11186623.0.0.9f8d7467kni6Aa
#- cn

#metrics:
##如 CIDEr、Blue、ROUGE 等
#- CIDEr

#tags:
##各种自定义，包括 pretrained、fine-tuned、instruction-tuned、RL-tuned 等训练方法和其他
#- pretrained

#tools:
##如 vllm、fastchat、llamacpp、AdaSeq 等
#- vllm
base_model:
  - Qwen/Qwen-Image
base_model_relation: adapter
---
# Qwen-Image Image Structure Control Model

![](assets/title.png)

## Model Introduction

This model is a LoRA for image structure control, trained based on [Qwen-Image](https://www.modelscope.cn/models/Qwen/Qwen-Image), adopting the In Context technical approach. It supports multiple conditions: canny, depth, lineart, softedge, normal, and openpose. The training framework is built upon [DiffSynth-Studio](https://github.com/modelscope/DiffSynth-Studio) , and the dataset used is[Qwen-Image-Self-Generated-Dataset](https://www.modelscope.cn/datasets/DiffSynth-Studio/Qwen-Image-Self-Generated-Dataset) It is recommended to start the input Prompt with "Context_Control. ".

Please note that when using Openpose control, due to the particularity of this type of control, it cannot achieve a similar "point-to-point" control effect as other control types.

## Effect Demonstration

|Control Condition|Control Image|Generated Image 1|Generated Image 2|
|-|-|-|-|
|canny|![](./assets/1_canny.png)|![](./assets/canny_image_seed_1_blue.png)|![](./assets/canny_image_seed_3_pink.png)|
|depth|![](./assets/1_depth.png)|![](./assets/depth_image_seed_2_blue.png)|![](./assets/depth_image_seed_2_pink_1.png)|
|lineart|![](./assets/1_lineart.png)|![](./assets/lineart_image_seed_1_blue.png)|![](./assets/lineart_image_seed_2_pink_1.png)|
|softedge|![](./assets/1_softedge.png)|![](./assets/softedge_image_seed_2_blue.png)|![](./assets/softedge_image_seed_2_pink_1.png)|
|normal|![](./assets/1_normal.png)|![](./assets/normal_image_seed_2_blue.png)|![](./assets/normal_image_seed_2_pink_1.png)|
|openpose|![](./assets/1_openpose.png)|![](./assets/openpose_image_seed_1_blue.png)|![](./assets/openpose_image_seed_4_pink.png)|


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

```python
from PIL import Image
import torch
from modelscope import dataset_snapshot_download, snapshot_download
from diffsynth.pipelines.qwen_image import QwenImagePipeline, ModelConfig
from diffsynth.controlnets.processors import Annotator

allow_file_pattern = ["sk_model.pth", "sk_model2.pth", "dpt_hybrid-midas-501f0c75.pt", "ControlNetHED.pth", "body_pose_model.pth", "hand_pose_model.pth", "facenet.pth", "scannet.pt"]
snapshot_download("lllyasviel/Annotators", local_dir="models/Annotators", allow_file_pattern=allow_file_pattern)

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-In-Context-Control-Union", local_dir="models/DiffSynth-Studio/Qwen-Image-In-Context-Control-Union", allow_file_pattern="model.safetensors")
pipe.load_lora(pipe.dit, "models/DiffSynth-Studio/Qwen-Image-In-Context-Control-Union/model.safetensors")

dataset_snapshot_download(dataset_id="DiffSynth-Studio/examples_in_diffsynth", local_dir="./", allow_file_pattern=f"data/examples/qwen-image-context-control/image.jpg")
origin_image = Image.open("data/examples/qwen-image-context-control/image.jpg").resize((1024, 1024))
annotator_ids = ['openpose', 'canny', 'depth', 'lineart', 'softedge', 'normal']
for annotator_id in annotator_ids:
    annotator = Annotator(processor_id=annotator_id, device="cuda")
    control_image = annotator(origin_image)
    control_image.save(f"{annotator.processor_id}.png")

    control_prompt = "Context_Control. "
    prompt = f"{control_prompt}一A beautiful girl in light blue is dancing against a dreamy starry sky with interweaving light and shadow and exquisite details."
    negative_prompt = "Mesh, regular grid, blurry, low resolution, low quality, distorted, deformed, wrong anatomy, distorted hands, distorted body, distorted face, distorted hair, distorted eyes, distorted mouth"
    image = pipe(prompt, seed=1, negative_prompt=negative_prompt, context_image=control_image, height=1024, width=1024)
    image.save(f"image_{annotator.processor_id}.png")
```

---
license: apache-2.0
---