workflow tutorial

ComfyUI is a powerful and modular diffusion model GUI, API, and backend with a graph/nodes interface for AI image generation. Unlike traditional AI art tools that use simple prompts, ComfyUI employs a visual programming approach where you connect different nodes together to create custom workflows, giving you granular control over every step of the image generation process. You can build complex pipelines for tasks like character consistency, architectural visualisation, batch processing, and advanced image manipulation using Stable Diffusion and other AI models. This comprehensive guide will walk you through every aspect of ComfyUI setup, from initial installation to advanced workflow creation, ensuring you can harness the full potential of this remarkable tool for your AI image generation projects.

AI Image Generator ComfyUI Setup and Configuration Guide

The platform is open-source under GPL-3.0 license with over 69,000 stars on GitHub, demonstrating strong community support and active development. ComfyUI supports various AI models including Stable Diffusion 1.5, SDXL, LoRA models, and ControlNet, while offering features like workflow sharing, custom node extensions, and professional-grade batch processing capabilities. You can download it from the official GitHub repository or explore workflow examples at the ComfyUI Examples site to see what’s possible with this versatile platform.

Advanced Model Support:

• Stable Diffusion 1.5, SDXL, and other checkpoint models
• LoRA (Low-Rank Adaptation) models for fine-tuning
• ControlNet for precise image control
• VAE models for image encoding/decoding
• Upscaling and enhancement models

System Requirements and Prerequisites

Before beginning your ComfyUI installation, ensuring your system meets the necessary requirements is crucial for optimal performance. The software demands substantial computational resources, particularly when processing high-resolution images or complex workflows.

Minimum Hardware Requirements:

• Graphics Card: NVIDIA GTX 1060 6GB or AMD RX 580 8GB
• RAM: 16GB system memory (32GB recommended)
• Storage: 50GB+ free space for models and outputs
• CPU: Intel i5-8400 or AMD Ryzen 5 2600 equivalent

Recommended Specifications:

• Graphics Card: NVIDIA RTX 3070 or better with 12GB+ VRAM
• RAM: 32GB+ for handling large models simultaneously
• Storage: NVMe SSD with 100GB+ free space
• CPU: Intel i7-10700K or AMD Ryzen 7 3700X or higher

Software Prerequisites:

• Windows 10/11, macOS 10.15+, or Ubuntu 18.04+
• Python 3.8-3.11 (for manual installation)
• Git (for cloning repositories and updates)
• CUDA 11.8+ or 12.1+ for NVIDIA GPUs

Your graphics card’s VRAM capacity significantly impacts model compatibility and generation speed. While 8GB VRAM can handle most standard models, 12GB+ enables working with larger models like SDXL without performance degradation.

Installation Methods: Portable vs Manual Setup

ComfyUI offers two primary installation approaches, each catering to different user preferences and technical expertise levels. Understanding these options helps you choose the method that best suits your needs and system configuration.

Portable Installation Method

The portable installation provides the simplest setup experience, ideal for beginners or users who prefer minimal configuration. This method bundles all necessary dependencies in a single package, eliminating potential conflicts with existing Python installations.

Portable Installation Steps:

1. Visit the official ComfyUI GitHub releases page
2. Download the latest portable release for your operating system
3. Extract the archive to your preferred directory (avoid paths with spaces)
4. Navigate to the extracted folder and run comfy installfrom Command Line
5.  Wait for initial setup to complete and dependencies to install
6. Run the run_nvidia_gpu.bat to run ComfyUI
7. You can access WorkFlow GUI on http://127.0.0.1:8188/

The portable version automatically handles Python environment management and package installations, making it perfect for users who want immediate functionality without technical complexity.

Manual Installation Process

Manual setup provides greater flexibility and control over the installation environment, making it suitable for advanced users or those with specific configuration requirements.

1. Clone the ComfyUI repository: git clone https://github.com/comfyanonymous/ComfyUI.git
2. Create a virtual environment: python -m venv comfyui_env
3. Activate the environment and navigate to the ComfyUI directory
4. Install dependencies: pip install -r requirements.txt
5. Install PyTorch with appropriate CUDA support for your system
6. Launch ComfyUI: python main.py

Manual installation allows customisation of Python versions, package versions, and integration with existing development environments, providing maximum flexibility for power users.

Downloading and Organising Models

Effective model management forms the foundation of successful ComfyUI setup, as the quality and variety of your AI-generated images depend heavily on the models you choose and how you organise them.

Essential Model Categories

Model Download Sources

Recommended Platforms:

• Hugging Face: Official and community models
• CivitAI: Extensive community-created content
• GitHub repositories: Developer-maintained collections
• Official Stability AI releases
• OpenArt Workflows has some interesting resources

Always verify model compatibility with your ComfyUI version and ensure you understand licensing terms before downloading. Some models require attribution or have commercial use restrictions.

Understanding the Node-Based Interface

ComfyUI’s node-based interface revolutionises AI image generation by providing unprecedented control over every aspect of the creation process. Unlike traditional prompt-based systems, nodes allow you to visualise and modify each step of the generation pipeline.

Core Interface Components

Node Types:

• Input Nodes: Text prompts, images, parameters
• Model Nodes: Checkpoints, VAE, LoRA loaders
• Processing Nodes: Samplers, schedulers, processors
• Output Nodes: Image savers, previewers, converters

Connection System: Nodes connect through input and output ports, with colour-coded cables indicating data types:

• Purple: Model connections
• Yellow: Conditioning (prompts)
• Pink: Images
• Blue: Masks
• Green: Numbers/parameters

Workflow Canvas: The main workspace where you arrange and connect nodes. Right-click to add new nodes, drag to reposition, and double-click nodes to access detailed settings.

Essential Node Operations

Adding Nodes: Right-click on empty canvas space to open the node menu. Browse categories or use the search function to find specific nodes quickly.

Connecting Nodes: Click and drag from output ports to compatible input ports. ComfyUI prevents incompatible connections, reducing setup errors.

Node Configuration: Click on nodes to reveal parameter settings. Many nodes offer advanced options accessible through right-click menus.

Workflow Navigation: Use mouse wheel to zoom, middle-click to pan, and Ctrl+scroll for precise navigation. The minimap helps navigate complex workflows efficiently.

ComfyUI node interface showing connected workflow elements

Creating Your First Workflow

Building your initial ComfyUI workflow establishes foundational understanding of the platform’s capabilities. This step-by-step approach ensures you grasp essential concepts before advancing to complex configurations.

Basic Text-to-Image Workflow

Required Nodes:

1. CheckpointLoaderSimple: Loads your base model
2. CLIPTextEncode: Processes positive and negative prompts
3. EmptyLatentImage: Defines output dimensions
4. KSampler: Handles the generation process
5. VAEDecode: Converts latent space to viewable images
6. SaveImage: Outputs final results

Connection Sequence:

• Connect checkpoint MODEL output to KSampler model input
• Link CLIP output to both CLIPTextEncode nodes
• Connect positive conditioning to KSampler positive input
• Connect negative conditioning to KSampler negative input
• Link EmptyLatentImage to KSampler latent_image input
• Connect KSampler output to VAEDecode samples input
• Link VAE from checkpoint to VAEDecode vae input
• Connect VAEDecode output to SaveImage image input

Configuration Parameters

Checkpoint Selection: Choose an appropriate base model based on your desired output style. Realistic Vision excels at photorealistic imagery, while DreamShaper produces more artistic results.

Prompt Engineering: Craft detailed positive prompts describing desired imagery and comprehensive negative prompts excluding unwanted elements.

Generation Settings:

• Steps: 20-30 for most models
• CFG Scale: 7-12 for balanced adherence
• Sampler: DPM++ 2M Karras for quality results
• Scheduler: Karras for smooth progression

Image Dimensions: Standard dimensions include 512×512, 768×768, or 1024×1024. Higher resolutions require more VRAM and processing time.

Testing Your Workflow

Execute your first generation by clicking “Queue Prompt” and monitor the progress through the console window. Successful execution validates your  workflow and confirms proper node connections.

Importing Community Workflows

The ComfyUI community creates sophisticated workflows addressing specific use cases, from portrait enhancement to architectural visualisation. Learning to import and modify these workflows accelerates your proficiency with the platform.

Workflow Sources

Popular Repositories:

• ComfyUI-Manager: Centralised workflow collection
• GitHub repositories: Developer-shared workflows
• Discord communities: Real-time workflow sharing
• Reddit forums: User-contributed examples

Workflow Categories:

• Character generation and consistency
• Background removal and replacement
• Style transfer and artistic effects
• Upscaling and enhancement
• Animation and video processing

Import Process

JSON Workflow Files:

1. Download the .json workflow file
2. Open ComfyUI in your browser
3. Drag the JSON file onto the workflow canvas
4. ComfyUI automatically recreates the node structure

PNG Embedded Workflows: Many community members embed workflow data within PNG images:

1. Drag PNG files with embedded workflows onto the canvas
2. ComfyUI extracts and loads the workflow automatically
3. Examine the recreated nodes and connections

Workflow Adaptation

Missing Nodes: If imported workflows reference unavailable custom nodes, ComfyUI displays red error nodes. Install required custom nodes through ComfyUI-Manager or manually clone repositories.

Model Compatibility: Replace model references with your available checkpoints, ensuring compatibility between model types and workflow requirements.

Parameter Adjustment: Modify generation parameters to suit your preferences and hardware capabilities. Higher-end workflows may require parameter reduction for systems with limited VRAM.

Custom Node Installation

Custom nodes extend ComfyUI’s functionality beyond core capabilities, enabling specialised features for advanced image generation techniques. Understanding custom node management is essential for maximising your ComfyUI potential.

Model Management and Organisation

Effective model management becomes increasingly important as your ComfyUI grows more sophisticated. Proper organisation, version control, and storage strategies ensure efficient workflow development and consistent results.

Storage Strategies

Directory Structure Best Practices: Maintain separate folders for different model categories, with subdirectories based on style, quality, or use case. This organisation accelerates model selection during workflow development.

Model Naming Conventions: Adopt consistent naming schemes including version numbers, training details, and style indicators. Clear names prevent confusion and simplify model selection in complex workflows.

Backup and Versioning: Regularly backup model collections and maintain version records for models you modify or fine-tune. This practice prevents data loss and enables workflow recreation.

Model Testing and Evaluation

Quality Assessment: Test new models with standardised prompts to evaluate output quality, style consistency, and generation reliability before integrating them into production workflows.

Compatibility Verification: Ensure model compatibility with your preferred samplers, schedulers, and generation parameters. Some models perform better with specific configuration combinations.

Performance Monitoring: Track generation times and memory usage for different models to optimise workflow efficiency and prevent system resource exhaustion.

Batch Processing Configuration

Batch processing capabilities transform ComfyUI from single-image generation to production-scale content creation. Understanding batch processing configuration enables efficient handling of large projects and automated generation tasks.

Queue System Configuration

Batch Generation Setup: Configure workflows to accept multiple inputs simultaneously, using array inputs for prompts, seeds, or parameters. This approach enables variation generation without manual intervention.

Queue Management: ComfyUI’s queue system processes multiple generation requests sequentially. Monitor queue status through the web interface and adjust priorities as needed.

Resource Allocation: Configure memory management settings to prevent system overload during batch processing. Balance generation speed against system stability based on your hardware capabilities.

Automated Workflow Triggers

Script Integration: Develop Python scripts that automatically queue workflows with varying parameters, enabling systematic exploration of generation possibilities.

Parameter Variation: Create workflows that automatically vary specific parameters across batch generations, useful for style experimentation or parameter optimisation.

Output Management: Configure automated file naming and organisation systems to handle large volumes of generated images efficiently.

ComfyUI batch processing queue showing multiple generation tasks

Output Configuration and Management

Professional ComfyUI setup, installation and usage requires sophisticated output management to handle diverse project requirements and maintain organised asset libraries.

File Format Options

Image Formats:

• PNG: Lossless compression, metadata preservation
• JPEG: Smaller file sizes, suitable for web use
• TIFF: Professional printing, maximum quality
• WebP: Modern format balancing quality and size

Metadata Embedding: Configure ComfyUI to embed generation parameters within image metadata, enabling workflow recreation and parameter analysis for successful generations.

Quality Settings: Adjust compression levels and quality parameters based on intended use. Archive copies warrant maximum quality, while preview versions can use higher compression.

Folder Organisation Systems

Project-Based Structure: Organise outputs by project, client, or campaign, with subdirectories for different generation phases or variations.

Date-Based Archives: Implement date-based folder structures for chronological organisation, particularly useful for ongoing projects or iterative development.

Automatic Sorting: Configure workflows to automatically sort outputs based on prompts, models, or generation parameters, reducing manual organisation overhead.

Performance Optimisation Techniques

Maximising ComfyUI performance and setting up ComfyUI correctly ensures efficient resource utilisation and faster generation times, particularly important for complex workflows or batch processing operations.

Memory Management

VRAM Optimisation:

• Unload unused models from VRAM between generations
• Configure model sharing between workflow components
• Implement memory cleanup routines for long-running sessions
• Monitor memory usage through system tools

System RAM Configuration: Allocate sufficient system RAM for model loading and workflow processing. Inadequate RAM forces excessive disk access, significantly impacting performance.

Cache Management: Configure ComfyUI’s caching behaviour to balance generation speed against storage requirements. Aggressive caching accelerates repeated operations but consumes disk space.

Generation Speed Optimisation

Sampler Selection: Different samplers offer varying speed-quality trade-offs. DPM++ 2M provides excellent results with moderate step counts, while Euler ancestral offers faster generation with slightly reduced quality.

Step Count Optimisation: Experiment with step counts to find the minimum required for acceptable quality. Many models produce excellent results with 20-25 steps, significantly faster than default 50-step configurations.

Resolution Strategies: Generate at lower resolutions for initial composition, then upscale using dedicated upscaling models. This approach reduces initial generation time while maintaining final image quality.

ComfyUI vs Automatic1111 Comparison

Understanding the differences between ComfyUI setup and Automatic1111 setup helps you choose the most appropriate platform for your specific needs and workflow requirements.

Interface Philosophy

Interface Philosophy Comparison between ConfyUI and Automatic1111

Performance Considerations

Performance Considerations between ConfyUI and Automatic1111

Use Case Scenarios

What to Choose between ConfyUI and Automatic1111

Workflow Examples and Templates

Practical workflow examples demonstrate ComfyUI capabilities across various use cases, providing templates for common generation scenarios.

Portrait Enhancement Workflow

Workflow Components:

1. Face detection and cropping
2. Detail enhancement processing
3. Lighting and colour correction
4. Background blur application
5. Final composition assembly

Key Nodes:

• FaceDetailer for facial feature enhancement
• ControlNet for pose consistency
• Upscaling nodes for resolution improvement
• Mask generation for selective processing

Architectural Visualisation Pipeline

Process Flow:

1. Sketch or wireframe input processing
2. ControlNet line art interpretation
3. Style application and material definition
4. Lighting and atmosphere enhancement
5. Final rendering and post-processing

Specialised Requirements:

• Architectural ControlNet models
• Professional rendering styles
• Precise geometric interpretation
• Realistic material application

Character Consistency System

Workflow Elements:

1. Reference character embedding
2. Pose and expression variation
3. Clothing and accessory modification
4. Scene integration processing
5. Quality assurance validation

Technical Considerations:

• LoRA training for character consistency
• Face embedding techniques
• Prompt engineering strategies
• Variation control mechanisms

Troubleshooting Common Issues

Effective troubleshooting skills ensure your ComfyUI setup remains functional and productive, minimising downtime and maximising creative output.

Installation Problems

Dependencies Conflicts:

• Verify Python version compatibility
• Use virtual environments for isolation
• Check CUDA version alignment
• Review package version conflicts

Model Loading Failures:

• Confirm file integrity and format
• Verify adequate storage space
• Check file permissions and access
• Validate model compatibility

Runtime Errors

Memory-Related Issues:

• Monitor VRAM usage during generation
• Implement model unloading strategies
• Reduce batch sizes for complex workflows
• Configure system virtual memory appropriately

Node Connection Problems:

• Verify data type compatibility
• Check for missing custom nodes
• Validate input parameter ranges
• Review workflow logic flow

Performance Degradation

Generation Speed Issues:

• Monitor system resource utilisation
• Check for background process interference
• Verify cooling system effectiveness
• Review power management settings

Quality Inconsistencies:

• Standardise generation parameters
• Verify model versions and sources
• Check for corrupted cache files
• Review prompt engineering techniques

Advanced Configuration Options

Advanced ComfyUI setup and configuration options unlock professional-level capabilities and customisation possibilities for demanding creative workflows.

Command Line Parameters

Performance Optimisation:

• --preview-method: Configure preview generation methods
• --use-split-cross-attention: Memory optimisation for limited VRAM
• --use-pytorch-cross-attention: Performance enhancement option
• --disable-safe-unpickle: Advanced model loading (use cautiously)

Development Options:

• --enable-cors-header: Cross-origin resource sharing
• --extra-model-paths-config: Custom model directory configuration
• --output-directory: Specify custom output locations

Configuration File Customisation

Model Paths Configuration: Create custom configuration files specifying model directories, enabling organised collections across multiple storage devices.

UI Customisation: Modify interface elements, themes, and layout options through configuration files, tailoring the interface to your workflow preferences.

Security Settings: Configure access controls, API permissions, and network security options for production deployment scenarios.

Conclusion

Mastering ComfyUI and ComfyUI setup opens unprecedented possibilities for AI image generation, transforming creative workflows through flexible node-based architecture and advanced customisation options. This comprehensive guide provides the foundation for building sophisticated generation pipelines tailored to your specific creative needs. The journey from initial installation to advanced workflow development requires patience and experimentation. Unlike Stable Diffusion on WSL or Standalone Stable Diffusion that is beginner friendly, you can start with ComfyUI basic configurations and then gradually incorporating custom nodes and complex processing chains as your understanding deepens. The modular nature of ComfyUI rewards incremental learning, allowing you to build expertise progressively.

Success with ComfyUI setup and using it depends on understanding its core philosophy: every aspect of image generation can be visualised, modified, and optimised through node connections. This transparency enables precise control over creative output while maintaining the flexibility to adapt workflows as requirements evolve. Whether you’re creating professional artwork, exploring AI capabilities, or developing commercial applications, ComfyUI provides the tools and flexibility needed for success. Embrace the learning curve, engage with the community, and explore the endless possibilities this remarkable platform offers for AI-powered creativity.