🚀 Quick Start Guide
Follow these steps to customize your images with IC-Custom
IC-Custom offers two customization modes:
Position-aware
(precise placement in masked areas) and
Position-free
(subject-driven generation).
1
Initial Setup (Both Modes)
- Choose your customization mode
- Upload a reference image 📸
2A
🎯 Position-aware Mode Steps
- Select input mask mode (precise mask or user-drawn mask)
- Upload target image and create mask (click for SAM or brush directly)
- Add text prompt (optional) - use VLM buttons for auto-generation
- Review and refine your mask using mask tools if needed
- Click Run ✨
2B
🎨 Position-free Mode Steps
- Write your text prompt (required) - describe the target scene
- Use VLM buttons for prompt auto-generation or polishing (if enabled)
- Click Run ✨
💡 Quick Tips:
Use Alt + "-" or ⌘ + "-" to zoom out for better operation •
Adjust settings in Advanced Options • Use mask operations (dilate/erode/bbox) for better results •
Try different seeds for varied outputs
🎉 Ready to start? Collapse this guide and begin customizing!
1. Select a Customization Mode
Tip: Please click the Clear button first to reset all states before starting a new task.
2. Input reference image
3. Select input mask mode
4. Input target image & mask (Iterate clicking or brushing until the target is covered)
5. Input text prompt (optional)
Output aspect ratio
0 1
-1 2147483647
1 4
10 65
1 60
1 10
6. View or modify the target mask
7. Submit and view the output
Examples
| Reference Image | Target Image (precise mask) | Target Image (user-drawn mask) | Customization Mode | Input Mask Mode | Segmentation mode | Text Prompt | Seed (-1 for random): | True GS | eg_idx | Number of inference steps | Guidance scale |
|---|
If IC-Custom is helpful, please help to ⭐ the Github Repo. Thanks!
📝 Citation
If our work is useful for your research, please consider citing:
@article{li2025ic,
title={IC-Custom: Diverse Image Customization via In-Context Learning},
author={Li, Yaowei and Li, Xiaoyu and Zhang, Zhaoyang and Bian, Yuxuan and Liu, Gan and Li, Xinyuan and Xu, Jiale and Hu, Wenbo and Liu, Yating and Li, Lingen and others},
journal={arXiv preprint arXiv:2507.01926},
year={2025}
}
📧 Contact
If you have any questions, please feel free to reach me out at liyaowei@gmail.com.