ComfyUI is a node-based user interface (UI) primarily designed for building, visualizing, and customizing stable diffusion AI workflows. It allows users to create complex pipelines for AI-based image generation intuitively and modularly. ComfyUI simplifies the process of managing AI models, connecting nodes, and experimenting with different configurations to produce high-quality results.

• GPU Support:
  • NVIDIA GPUs (compatible with CUDA).
  • AMD GPUs (compatible with ROCm).
• Python 3.10 or higher.
• Operating Systems: Windows, macOS, Linux (Linux recommended for maximum performance).

Download the ComfyUI project using Git:

git clone https://github.com/comfyanonymous/ComfyUI.git

Navigate to the project directory:

cd ComfyUI

Create and activate a virtual environment to isolate dependencies:

python -m venv venv  
source venv/bin/activate     (Linux/MacOS)  
venv\Scripts\activate        (Windows)

Install required Python dependencies:

pip install -r requirements.txt

Run the server:

python main.py

Open your browser and go to: http://127.0.0.1:8188

• Design workflows visually by dragging and dropping nodes (components of the Stable Diffusion pipeline).
• Connect nodes to define the relationships between tasks.

1. Model Management:
   • Load and switch between Stable Diffusion models (checkpoints, .ckpt, or .safetensors files).
   • Manage VAE (Variational Autoencoder) for enhancing outputs.
2. Image Generation:
   • Create images using Text-to-Image (T2I) or Image-to-Image (I2I) workflows.
   • Configure parameters such as resolution, sampling method, CFG scale, and steps.
3. Customizable Pipelines:
   • Add nodes for specialized tasks (e.g., prompt preprocessing, model loading, sampler, output).
4. Batch Processing:
   • Process multiple prompts, seeds, or images simultaneously.
5. Advanced Imaging:
   • Use additional nodes like ControlNet for advanced image editing capabilities (inpainting, sketch-to-image, etc.).

1. Drag Text Prompt Node to the canvas.
   • Input your desired prompt, such as "a futuristic cityscape".
2. Add Model Loader Node and select your preferred AI model (e.g., Stable Diffusion checkpoint).
3. Attach Sampler Node to specify the sampling method (e.g., DDIM, Euler).
4. Connect an Image Output Node at the end to display/save the generated image.
5. Run the pipeline after connecting the nodes!

Place your .ckpt or .safetensors files under the appropriate models directory.

Place your VAE files into the vae directory.

• Download pre-trained ControlNet checkpoint files and add them to the models directory.
• Configure ControlNet nodes in the ComfyUI interface to enable advanced workflows.

Customize the config.json file for better performance:

{
    "device": "cuda",
    "always_use_CPU": false,
    "workflow_directory": "./workflows",
    "model_directory": "./models"
}

1. Python Version Requirements: Ensure you're using Python 3.10 or higher.

2. Missing Dependencies: Run pip install -r requirements.txt again to install dependencies properly.

3. GPU Compatibility Issues: Install the correct CUDA Toolkit for NVIDIA GPUs, or use ROCm for AMD GPUs.

4. Performance Tuning: Enable optimizations like XFormers to improve memory efficiency (if supported).

To process multiple image prompts:

1. Add a Batch Prompt Node.
2. Specify multiple text prompts (e.g., "a sunset over the ocean" and "a forest in winter").
3. Configure the Sampler Node and enable batch settings as needed.
4. Generate all images concurrently.

Run the main server script:

python main.py

Access http://127.0.0.1:8188 in your browser.

1. Install dependencies.
2. Open and configure ComfyUI.
3. Build workflows visually by connecting nodes.
4. Experiment with parameters and generate your desired output.
5. Save results and refine workflows iteratively.

This Markdown file outlines the features, installation instructions, workflow usage, and troubleshooting tips for getting started with ComfyUI. Save the content into a file named ComfyUI.md for documentation purposes.