ComfyUI Extension: Arctenoxs-Essentials_ComfyUI

Authored by Arctenox

Created 2 days ago

Updated about 9 hours ago

0 stars

A collection of efficient workflow nodes for ComfyUI with advanced sampling, seed management, temporal prompt processing, and workflow optimization tools

Custom Nodes (0)

README

Arctenox's Essentials

A collection of efficient workflow nodes for ComfyUI, designed to streamline and optimize your generation process with combined functionality and improved performance.

📧 Contact & Support

Discord: https://discord.gg/UVXPdkgedh
GitHub: @Arctenox
Issues: GitHub Issues

📦 Installation

Via ComfyUI Manager (Recommended)

Open ComfyUI Manager
Search for "Arctenoxs-Essentials_ComfyUI" or "Arctenox's Essentials"
Click Install
Restart ComfyUI

Manual Installation

Navigate to your ComfyUI custom nodes directory:
```
cd ComfyUI/custom_nodes/
```

Clone this repository:

git clone https://github.com/Arctenox/Arctenoxs-Essentials_ComfyUI

Restart ComfyUI

Requirements

ComfyUI (latest version recommended)
PyTorch >= 1.12.0
Python >= 3.8
Optional: psutil for CPU memory monitoring

If you encounter missing dependencies, install them with:

pip install torch numpy psutil

🎯 Overview

Arctenox's Essentials provides a suite of powerful workflow nodes that enhance ComfyUI with advanced sampling techniques, intelligent seed management, temporal prompt processing, and workflow optimization tools. Each node is designed to integrate seamlessly into your existing workflows while adding sophisticated functionality. Also if you see "Sampling failed: . Check that all inputs are valid tensors." upon canceling it while running the KSampler in this, don't panic, that will happen and you can completely ignore it, I couldn't get that error pop up to go away, so it's just there.

Key Features

🎲 Advanced KSampler - Combined latent generation and sampling with golden ratio sonar transformation
🗺️ Seed Topology Mapping - Generate mathematically related seed families
⏱️ Temporal Prompt Splitting - Progressive refinement through composition → detail → texture phases
⚠️ Artifact Risk Prediction - Detect potential issues before VAE decoding
💰 Cost Estimation - Real-time VRAM, time, and efficiency metrics
💾 Metadata Preservation - Save images with complete generation parameters
🎨 Prompt Styling - 40+ presets for different artistic styles
📦 Workflow Utilities - Organization boxes, checkpoint passthrough, and more
🚀 Performance Optimized - Memory-efficient processing across CUDA/MPS/CPU

📚 Node Documentation

🎲 KSampler (Arctenox's Essentials)

An enhanced sampler that combines latent generation with advanced sampling features and golden ratio sonar transformation.

Location: Arctenox Essentials/Sampling

Inputs

| Input | Type | Description | |-------|------|-------------| | model | MODEL | The model to use for sampling | | positive | CONDITIONING | Positive conditioning (what you want) | | negative | CONDITIONING | Negative conditioning (what to avoid) | | width | INT | Image width (64-8192, step 8) | | height | INT | Image height (64-8192, step 8) | | batch_size | INT | Number of images (1-64) | | seed | INT | Base random seed | | sonar | INT | Golden ratio transformation offset | | steps | INT | Sampling steps (1-10000+) | | cfg | FLOAT | Classifier-free guidance scale (0.0-100.0+) | | sampler_name | STRING | Sampling algorithm | | scheduler | STRING | Noise schedule | | denoise | FLOAT | Denoising strength (0.0-1.0) | | vae_decode | BOOLEAN | Enable/disable automatic decoding | | latent_image | LATENT (optional) | Input latent for img2img | | optional_vae | VAE (optional) | VAE for decoding |

Outputs

| Output | Type | Description | |--------|------|-------------| | MODEL | MODEL | Passthrough of input model | | CONDITIONING+ | CONDITIONING | Passthrough of positive conditioning | | CONDITIONING- | CONDITIONING | Passthrough of negative conditioning | | LATENT | LATENT | Generated latent samples | | VAE | VAE | Passthrough of input VAE | | IMAGE | IMAGE | Decoded image (if vae_decode enabled) |

Features

Golden Ratio Sonar Transformation

The sonar parameter applies a sophisticated seed transformation based on the golden ratio (φ ≈ 1.618):

φ⁻¹ ≈ 0.618033988749895

How It Works:

Base hash transformation for deterministic offset
Golden ratio influence for values > 10
Conservative 10% influence factor
Sign preservation for negative sonar values
32-bit seed space normalization

Benefits:

Aesthetically pleasing variation patterns
Deterministic and reproducible
Smooth transitions between related seeds
Natural-feeling exploration of latent space

Example Usage:

Base Seed: 42
Sonar: 0 → Standard generation
Sonar: 100 → Subtle golden ratio influenced variation
Sonar: -100 → Opposite direction variation

Intelligent CFG Handling

Handles extreme CFG values gracefully:

Normal range (≤100): Direct passthrough
Large values (>100): Logarithmic scaling
Prevents sampling instability
Maintains quality across extreme settings

Large Step Support

Intelligently handles astronomical step counts:

Normal range (≤10000): Direct passthrough
Large values: Capped at reasonable maximum
Prevents memory issues
Maintains sampling quality

Negative Seed Support

Deterministic hashing for negative seed values:

Consistent results for negative seeds
Reproducible transformations
Expands exploration space

Use Cases

Exploring seed variations systematically with sonar
Batch generation with consistent quality
Streamlined workflows without separate latent nodes
Experimenting with extreme CFG values
Quick iterations without manual latent creation

🗺️ Seed Topology Mapper

Generate families of mathematically related seeds for structured exploration of the latent space.

Location: Arctenox Essentials/Sampling

Inputs

| Input | Type | Description | |-------|------|-------------| | base_seed | INT | Starting seed for topology generation | | topology_type | STRING | Mathematical spacing method | | num_seeds | INT | Number of seeds to generate (1-16) | | spread | FLOAT | Spacing multiplier (0.1-10.0) |

Outputs

Returns 16 seed outputs (seed_1 through seed_16), each of type INT.

Topology Types

| Type | Description | Best For | Recommended Spread | |------|-------------|----------|-------------------| | Golden Angle | Seeds spaced by 137.5° golden angle | Natural, organic variations | 0.5-2.0 | | Harmonic | Musical harmony ratios (1:2:3:4...) | Rhythmic, balanced sets | 1.0-5.0 | | Fibonacci | Fibonacci sequence spacing | Natural progression patterns | 0.1-1.0 | | Chaos Neighbors | Controlled chaotic spacing | Unpredictable but related results | 0.5-1.5 | | Prime Spiral | Prime number spiral distribution | Mathematical exploration | 1.0-3.0 | | Phi Spacing | Golden ratio (φ) based intervals | Aesthetically pleasing variations | 0.5-2.0 |

Mathematical Details

Each topology uses distinct mathematical spacing:

Golden Angle: 137.5077° rotation spacing
Harmonic: Multiply by integer ratios
Fibonacci: Add Fibonacci sequence values
Chaos: Logistic map with r=3.9
Prime Spiral: Prime number intervals
Phi Spacing: Exponential golden ratio growth

Use Cases

Create seed families with related characteristics
Systematic exploration of variations
Reproducible batch processing workflows
Finding "neighboring" results in latent space
Generating consistent variation sets

Example Workflow

Base Seed: 42
Topology: golden_angle
Num Seeds: 8
Spread: 1.0

Result: 8 deterministically related seeds, each exploring
a different "direction" in the latent space while maintaining
a golden angle relationship

⏱️ Prompt Phase Splitter

Split your generation into temporal phases for progressive refinement and enhanced control.

Location: Arctenox Essentials/Sampling

Phase System

┌─────────────┐     ┌─────────────┐     ┌─────────────┐
│ COMPOSITION │ ──> │   DETAIL    │ ──> │   TEXTURE   │
│   Phase     │     │   Phase     │     │   Phase     │
│  (Early)    │     │  (Middle)   │     │   (Late)    │
└─────────────┘     └─────────────┘     └─────────────┘

Inputs

| Input | Type | Default | Description | |-------|------|---------|-------------| | clip | CLIP | - | CLIP model for text encoding | | composition_prompt | STRING | "" | Early phase - overall composition and layout | | detail_prompt | STRING | "" | Middle phase - subject details and refinement | | texture_prompt | STRING | "" | Late phase - surface qualities and fine details | | composition_weight | FLOAT | 1.0 | Strength of composition phase (0.0-2.0) | | detail_weight | FLOAT | 1.0 | Strength of detail phase (0.0-2.0) | | texture_weight | FLOAT | 1.0 | Strength of texture phase (0.0-2.0) | | phase_1_steps | INT | 10 | Steps for composition phase | | phase_2_steps | INT | 10 | Steps for detail phase | | phase_3_steps | INT | 5 | Steps for texture phase | | enable_character_focus | BOOLEAN | false | Optional character/subject emphasis | | character_focus_prompt | STRING | "" | Subject-specific refinement | | character_weight | FLOAT | 1.0 | Character focus strength (0.0-2.0) |

Outputs

| Output | Type | Description | |--------|------|-------------| | combined_conditioning | CONDITIONING | Weighted combination of all phases | | total_steps | INT | Sum of all phase steps |

How It Works

The temporal phase system divides the sampling process into three cognitive stages:

Total Steps = Phase1 + Phase2 + Phase3

Phase 1 (Composition): Steps 0 → N₁
  └─ Global structure, layout, composition

Phase 2 (Detail): Steps N₁ → N₂  
  └─ Subject refinement, feature details

Phase 3 (Texture): Steps N₂ → Total
  └─ Surface qualities, fine textures

Each phase is weighted and combined to create a temporally-aware conditioning that guides the sampler through progressive refinement stages.

Use Cases

Separate composition from detail refinement
Better control over generation progression
Prevent early detail interference with composition
Character-focused image generation
Architectural and landscape rendering with distinct phases
Portrait generation with facial detail emphasis

Example: Portrait Generation

Phase 1 (10 steps, weight 1.0):
"portrait of a person, professional photography"

Phase 2 (10 steps, weight 1.0):
"detailed facial features, expressive eyes, natural skin"

Phase 3 (5 steps, weight 1.0):
"fine skin texture, hair detail, fabric texture"

Character Focus (enabled, weight 1.2):
"the subject's face and expression"

This ensures the composition establishes the overall portrait first, then refines facial features, and finally adds texture details, with extra emphasis on the face throughout.

⚠️ Artifact Risk Predictor

Analyze latent samples before VAE decoding to predict potential artifacts and quality issues.

Location: Arctenox Essentials/Utilities

Inputs

| Input | Type | Description | |-------|------|-------------| | latent | LATENT | Latent samples to analyze |

Outputs

| Output | Type | Description | |--------|------|-------------| | risk_score | FLOAT | Numerical risk assessment (0.0-1.0) | | risk_level | STRING | Categorized risk level | | latent_passthrough | LATENT | Original latent for downstream processing |

Detection Capabilities

The predictor analyzes latent samples for:

High-frequency noise patterns - Excessive noise in high frequencies
Extreme value spikes - Abnormal peaks in latent values
Abnormal distribution patterns - Statistical anomalies in value distribution
Channel inconsistencies - Imbalances between latent channels
Statistical anomalies - Deviation from expected latent statistics
Overbakedness/Artifacts - Signs of overprocessing

Risk Levels

| Level | Range | Color | Description | |-------|-------|-------|-------------| | Low | 0.0-0.3 | 🟢 | Clean generation expected, no issues detected | | Medium | 0.3-0.6 | 🟡 | Minor artifacts possible, generally acceptable | | High | 0.6-1.0 | 🔴 | Significant issues likely, consider regenerating |

Use Cases

Quality control before VAE decoding
Pre-publication image validation
Automatic bad sample filtering
Debugging problematic generations
Batch processing optimization
Saving time by catching issues early

Example Workflow

[KSampler] → [Artifact Risk Predictor] → [Conditional VAE Decode]
                      ↓
              [Filter High Risk]

Only decode latents with acceptable risk scores, saving computation time and storage.

💰 Execution Cost Estimator

Real-time performance metrics and resource usage estimation for your workflows.

Location: Arctenox Essentials/Utilities

Inputs

| Input | Type | Description | |-------|------|-------------| | model | MODEL | Model to analyze | | width | INT | Target image width | | height | INT | Target image height | | batch_size | INT | Number of samples | | steps | INT | Sampling steps | | Additional analysis toggles | BOOLEAN | Enable various analysis features |

Outputs

| Output | Type | Description | |--------|------|-------------| | cost_info | STRING | Detailed cost breakdown | | vram_estimate | FLOAT | Estimated VRAM usage (GB) | | time_estimate | FLOAT | Expected generation duration (seconds) | | efficiency_score | FLOAT | Performance optimization rating |

Metrics Provided

VRAM Usage: GPU memory consumption estimate
Execution Time: Expected generation duration
Efficiency Score: Performance optimization rating
Bottleneck Analysis: Identify performance limiters
Resource Recommendations: Suggestions for optimization

Use Cases

Workflow optimization
Hardware upgrade planning
Batch size optimization
Performance debugging
Resource allocation planning
Preventing OOM errors

Note: Estimates are not 100% accurate but provide useful guidance for workflow planning.

💾 Save Image With Metadata (Arctenox's Essentials)

Save images with complete generation parameters embedded in metadata for full reproducibility.

Location: Arctenox Essentials/Output

Inputs

| Input | Type | Description | |-------|------|-------------| | images | IMAGE | Images to save | | filename_prefix | STRING | Prefix for saved filenames | | Various metadata fields | STRING/INT/FLOAT | Generation parameters to embed |

Features

Embeds complete generation parameters in PNG metadata
Supports Civitai resource detection via model hashes
Compatible with A1111/Automatic1111 metadata format
Preserves seed, prompt, model, sampler, and all settings
Enables perfect reproducibility of generations

Use Cases

Creating reproducible workflows
Sharing generation settings with others
Building dataset with complete provenance
Documenting successful generations
Civitai automatic resource linking

🎨 Prompt Styler (Arctenox's Essentials)

Apply professional styling presets to your prompts with 40+ built-in styles.

Location: Arctenox Essentials/Prompting

Available Styles

Photography Styles:

Cinematic
Portrait
Landscape
Macro
Street Photography
Fashion
Product Photography
Documentary

Artistic Styles:

Oil Painting
Watercolor
Digital Art
Anime
Sketch
Comic Book
Concept Art
Abstract

Technical Styles:

Photorealistic
Hyperrealistic
Studio Lighting
Natural Light
Dramatic Lighting
Soft Focus
High Contrast
Vintage

And many more...

Use Cases

Quick style application without remembering complex prompts
Consistent styling across batches
Learning effective prompt structures
Experimenting with different artistic styles
Professional-quality prompt construction

📝 CLIP Text Encode + String (Arctenox's Essentials)

Standard CLIP text encoding with additional string output for metadata capture.

Location: Arctenox Essentials/Conditioning

Inputs

| Input | Type | Description | |-------|------|-------------| | text | STRING | Text to encode with CLIP | | clip | CLIP | CLIP model |

Outputs

| Output | Type | Description | |--------|------|-------------| | conditioning | CONDITIONING | CLIP-encoded conditioning | | text | STRING | Original text for metadata |

Use Cases

Capturing prompts for metadata embedding
Documenting exact prompts used
Connecting to SaveImageWithMetadata
Maintaining prompt history

🔧 Load Checkpoint + String (Arctenox's Essentials)

Load checkpoints with model name and hash output for Civitai detection.

Location: Arctenox Essentials/Loaders

Inputs

| Input | Type | Description | |-------|------|-------------| | ckpt_name | STRING | Name of checkpoint file |

Outputs

| Output | Type | Description | |--------|------|-------------| | model | MODEL | Loaded model | | clip | CLIP | Loaded CLIP | | vae | VAE | Loaded VAE | | model_name | STRING | Checkpoint filename | | model_hash | STRING | SHA256 hash (first 10 chars) |

Features

Efficient hash caching to avoid recalculation
SHA256 AutoHash format (compatible with A1111/Civitai)
Automatic Civitai resource detection when embedded in metadata

Use Cases

Metadata capture for reproducibility
Civitai automatic model linking
Documenting exact model versions used
Building model usage databases

📦 Checkpoint Passthrough + Notes (Arctenox's Essentials)

Pass through MODEL, CLIP, and VAE together with optional documentation.

Location: Arctenox Essentials/Utilities

Inputs

| Input | Type | Description | |-------|------|-------------| | model | MODEL | Model to pass through | | clip | CLIP | CLIP to pass through | | vae | VAE | VAE to pass through | | notes | STRING (optional) | Documentation notes |

Outputs

| Output | Type | Description | |--------|------|-------------| | model | MODEL | Passthrough of input model | | clip | CLIP | Passthrough of input CLIP | | vae | VAE | Passthrough of input VAE |

Use Cases

Organizing checkpoint flows in complex workflows
Adding documentation to checkpoint sections
Creating clean connection points for checkpoints
Improving workflow readability
Zero processing overhead - direct passthrough

📦 Box (Empty Node)

Simple organizational box for workflow documentation and structure.

Location: Arctenox Essentials/Utilities

Features

No inputs or outputs
Pure visual organization
Workflow sectioning
Documentation anchor points

Use Cases

Visually separate workflow sections
Add structure to complex graphs
Create workflow "chapters"
Improve readability
Label different workflow stages

🛠️ Advanced Configuration

Custom Sonar Curves

For advanced users, the sonar transformation can be tuned by modifying the influence factor in ArctenoxEssentials_KSampler.py:

golden_influence = (abs_sonar * PHI_INVERSE * 0.1) % 1.0  # 0.1 = 10% influence
golden_adjustment = int(base_noise_seed * golden_influence * 0.05)  # 5% max change

Adjust these percentages to control how strongly the golden ratio influences seed transformations.

Topology Spread Optimization

Different topology types respond differently to spread values:

| Topology | Optimal Spread Range | Notes | |----------|---------------------|-------| | Golden Angle | 0.5-2.0 | Subtle to extreme variations | | Harmonic | 1.0-5.0 | Musical spacing | | Fibonacci | 0.1-1.0 | Natural progressions | | Chaos | 0.5-1.5 | Controlled chaos | | Prime Spiral | 1.0-3.0 | Mathematical spacing | | Phi Spacing | 0.5-2.0 | Aesthetic distributions |

🛟 Troubleshooting

Node Not Appearing

Verify installation in ComfyUI/custom_nodes/
Check console for import errors
Ensure PyTorch is properly installed
Restart ComfyUI completely
Clear browser cache if using web interface

Import Errors

Missing dependencies? Install with:

pip install torch numpy psutil

Performance Issues

Reduce batch size for VRAM constraints
Use Cost Estimator to identify bottlenecks
Enable CPU offloading in ComfyUI settings
Consider lower resolution for testing
Close other GPU-intensive applications

Unexpected Results

Verify seed values are within valid range
Check sonar values aren't too extreme (±1000000)
Ensure prompt weights are reasonable (0.5-2.0)
Review phase step distributions
Check that total steps match your expectations

Known Issues

Cancelling KSampler may show an error message - this is expected and safe to ignore
Very large sonar values (>1000000) may produce unpredictable results
Artifact Risk Predictor is a heuristic - not 100% accurate

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request. For major changes, please open an issue first to discuss what you would like to change.

Development Setup

git clone https://github.com/Arctenox/Arctenoxs-Essentials_ComfyUI
cd Arctenoxs-Essentials_ComfyUI
# Make your changes
# Test in ComfyUI
# Submit PR

Guidelines

Follow existing code style
Add documentation for new nodes
Test thoroughly before submitting
Update README with new features

📄 License

This project is licensed under the GNU General Public License v3.0 - see the LICENSE file for details.

🗺️ Roadmap

[ ] More Nodes that I find suitable for this node pack that I can think up
[ ] Workflow templates

Made with ❤️ by Claude Sonnet 4.5/3.5 and Arctenox

Version 1.2.0 - Last Updated: December 21st 2025