BlobCtrl: A Unified and Flexible Framework for Element-level Image Generation and Editing

Project Page

Official Gradio Demo for BlobCtrl: A Unified and Flexible Framework for Element-level Image Generation and Editing
🦉 BlobCtrl enables precise, user-friendly element-level visual manipulation.

Step 1: Upload an image and click to segment the object

Step 2: Input the scene prompt and 🎩 generate the blob

💡 Hint: Adjust the control strength and control timesteps range to balance appearance and flexibility
0 2.5
0 1
0 1

Click to adjust the diffusion sampling options 👇
0 2147483647
0 4
1 12
1 100

Click to expand more previews 👇

Step 3: Edit the blob, such as move/resize/remove the blob

Click to adjust the target blob size 👇
0.1 2

Click to adjust the initial blob size to ensure it surrounds the object👇
0.1 2

Click to achieve more edit types, such as single-sided resize, composition, etc. 👇

slide to achieve single-sided resize and rotation
0 2
0 2
-180 180

slide to adjust the initial blob (single-sided)
0 2
0 2

🎨 Click to set the initial blob and upload object image for compositional generation👇

Step 4: 🚀 Run Generation
Quick Example
input Scene Prompt 🎚️ Control Strength: Blobnet Control Timestep Start Blobnet Control Timestep End Seed: Example Index

If BlobCtrl is helpful, please help to ⭐ the Github Repo. Thanks!

📝 Citation
If our work is useful for your research, please consider citing:

@misc{li2025blobctrl,
  title={BlobCtrl: A Unified and Flexible Framework for Element-level Image Generation and Editing}, 
  author={Yaowei Li},
  year={2025},
  eprint={2502.09477},
  archivePrefix={arXiv},
  primaryClass={cs.CV}
}

📧 Contact
If you have any questions, please feel free to reach me out at liyaowei@gmail.com.