🎨 Object Fusion Pipeline

🌟 Motivation

Have you ever fantasized about crafting a single masterpiece from elements of different photos? 😲 What if I told you that manually prompting each object's shape, color, and features for diffusion could be a thing of the past? And let’s be real, using GPT to describe each object step by step can feel like chasing an AI in slow motion. 🐢💤

Here, I introduce a complete pipeline that takes two images as input, allowing you to choose which objects from the images you want to fuse seamlessly.

🚀 Features

Combine objects from two different images into a single scene.
Easy selection and customization of objects to be fused.
Optimized integration with ComfyUI.

🖼️ Some Examples

Click to expand/collapse

🛠️ Installation

Clone this repo into the custom_nodes directory in ComfyUI:

git clone https://github.com/ducido/ObjectFusion_ComfyUI_nodes

Clone these amazing repositories and follow their instructions:
- ComfyUI-SD3-nodes
  Note: Place the 3 clips model into models/clip.
- ComfyUI-Custom-Scripts

Install the required packages:

conda create -n objectfusion python=3.10 -y
conda activate objectfusion
pip install -r requirements.txt
wget https://huggingface.co/camenduru/YoloWorld-EfficientSAM/resolve/main/efficient_sam_s_gpu.jit -P custom_nodes/ObjectFusion_ComfyUI_nodes/Custom_ComfyUI_YoloWorld_EfficientSAM

📌 Note

All the folders, except CROP_OBJECT, are from other repositories. However, I have made some minor modifications to fit this project. Here are the details:

Custom_ComfyUI-YoloWorld-EfficientSAM
- Added 2 output fields: BBOX, categories.
- Displayed ID also in the IMAGE output (e.g., {ID} - {class} - {confidence}).
comfyui-llm-assistant
- Removed input field: prompt.
- Added 4 input fields: object1, desc_obj1, object2, desc_obj2.
- Change all default value to "" because the newest frontend of ComfyUI consider None value as a bug
img2txt-comfyui-nodes
- Add file tensor_img_utils.py in src
- Fix bug import part in src/img2txt_node.py

🤝 Contributing

Contributions are welcome! Please open an issue or submit a pull request.

📄 License

This project is licensed under the MIT License.

3.1 KiB Raw Blame History Unescape Escape