ComfyUI Extension: ComfyUI_PerformanceLab

Authored by laboratoiresonore

Created

Updated

1 stars

Make any ComfyUI workflow faster, use less VRAM, or produce better quality - with AI assistance

Custom Nodes (0)

    README

    ComfyUI Performance Lab

    Make any ComfyUI workflow faster, use less VRAM, or produce better quality - with help from AI.

    Performance Lab Python 3.7+ No Dependencies Version

    ╔══════════════════════════════════════════════════════════════════════════════╗
║                   ⚡ COMFYUI PERFORMANCE LAB v0.7.0 ⚡                        ║
║           Load → Test → Get AI Suggestions → Accept/Reject → Repeat          ║
╚══════════════════════════════════════════════════════════════════════════════╝

    What Does It Do?

    Performance Lab helps you optimize ANY ComfyUI workflow - from a simple SD generator to a complex multi-machine network of AI services. It creates a simple loop:

    ┌─────────────────────────────────────────────────────────────────────────────┐
│  1. LOAD      Load any ComfyUI workflow                                     │
│  2. TEST      Run it and collect performance metrics (time, VRAM)           │
│  3. ASK AI    Generate a prompt and paste it to Claude/GPT/Gemini           │
│  4. GET MOD   Copy the AI's suggested improvement                           │
│  5. PASTE     Paste the mod into Performance Lab                            │
│  6. TEST      Run the modified workflow                                     │
│  7. DECIDE    Better? Accept. Worse? Reject.                                │
│  8. REPEAT    Keep optimizing until you're happy!                           │
└─────────────────────────────────────────────────────────────────────────────┘

    That's it. Performance Lab handles all the complexity - you just copy-paste between it and your favorite LLM.

    Installation

    Method 1: ComfyUI Manager (Easiest)

    If you have ComfyUI Manager installed:

    1. Open ComfyUI in your browser
    2. Click Manager → Install via Git URL
    3. Paste: https://github.com/laboratoiresonore/ComfyUI_PerformanceLab
    4. Click OK and restart ComfyUI

    After restart, you'll see:

    • 27 new nodes in the "⚡ Performance Lab" category
    • A startup message in the console confirming installation

    To use the full CLI:

    cd ComfyUI/custom_nodes/ComfyUI_PerformanceLab
python performance_lab.py

    Method 2: Git Clone (Recommended)

    Open a terminal in your ComfyUI folder:

    # Navigate to your ComfyUI installation
cd /path/to/ComfyUI

# Clone Performance Lab
git clone https://github.com/laboratoiresonore/ComfyUI_PerformanceLab.git custom_nodes/ComfyUI_PerformanceLab

# Run the installer (sets everything up)
cd custom_nodes/ComfyUI_PerformanceLab
python install.py

# Start Performance Lab
python performance_lab.py

    Method 3: Download ZIP

    1. Download from GitHub: Code → Download ZIP
    2. Extract to ComfyUI/custom_nodes/ComfyUI_PerformanceLab/
    3. Open terminal in that folder and run: python performance_lab.py

    No additional dependencies required! Works with Python's standard library.

    Quick Start (5 Minutes)

    1. Start ComfyUI (must be running for metrics collection)

    2. Run Performance Lab:

      python performance_lab.py

    3. Load your workflow when prompted (enter the path to your .json file)

    4. Try Quick Actions [2] - Instant one-click optimizations:

      • Cap resolution to 768px (~60% faster testing)
      • Bypass upscalers (saves 2-4GB VRAM)
      • Speed Test preset (all optimizations at once)

    5. Get AI Help [3] - Generate a prompt, paste it to Claude/ChatGPT/Gemini, copy the response back

    6. Accept or Reject - If it's better, keep it. If not, discard it.

    Main Menu

    ⚡ MAIN MENU
Target: my_workflow.json
ComfyUI: ● Connected

  1  Apply a Mod (from mods folder)
  2  ⚡ Quick Actions (instant one-click optimizations)
  3  Generate LLM Prompt (for Claude/GPT/Gemini)
  4  Paste New Mod (from AI response)
  5  🧠 Smart Suggestions (AI-free analysis)
  6  📊 Benchmark Mode (run 3x for reliable metrics)
  7  📈 View Dashboard (see all your optimizations)
  8  ⚙️  Presets (8GB VRAM, Speed Test, etc.)
  9  Set Goal (tell the AI what you want)
  L  🤖 LLM Enhancer (advanced AI context)
  M  🎛️  Model Tuner (auto-detect SD/SDXL/Flux)
  D  🌐 Distributed (multi-machine optimization)
  Q  Quit

    What Makes It Different?

    Simple Iterative Loop

    Most optimization tools give you one suggestion. Performance Lab lets you iterate - make a change, test it, keep or discard, repeat. This is how real optimization works.

    Works with Any LLM

    Use Claude, ChatGPT, GPT-4, Gemini, Llama, Mistral - whatever you have access to. Performance Lab generates optimized prompts for each.

    Minimal File Clutter

    • Accept = overwrites your original workflow
    • Reject = discards the changes completely
    • No _v1, _v2, _final_FINAL files everywhere

    Smart Protection

    Performance Lab knows when changes are risky:

    • Won't overwrite an SD 1.5 workflow with SDXL settings
    • Warns you about significant changes
    • Tracks what model family you're using

    Quick Actions (No AI Needed)

    Press [2] for instant optimizations:

    | Action | What It Does | Expected Improvement | |--------|-------------|---------------------| | Cap 768px | Reduce all resolutions to 768 | ~60% faster | | Cap 1024px | Reduce all resolutions to 1024 | ~40% faster | | Bypass Upscalers | Skip all upscaler nodes | 2-4GB VRAM saved | | Reduce Steps | Set sampling steps to 20 | Faster iteration | | 🚀 Speed Test | All optimizations combined | Maximum speed | | 💾 8GB VRAM | Optimized for 8GB GPUs | Fits on 8GB cards |

    AI-Assisted Optimization

    Press [3] to generate a prompt for your favorite LLM:

    1. Performance Lab creates a detailed prompt with:

      • Your workflow structure
      • Current performance metrics
      • Your optimization goal
      • System specs (GPU, VRAM, etc.)

    2. Copy the prompt to Claude/ChatGPT/Gemini

    3. The AI responds with optimized workflow code

    4. Press [4] and paste the response

    5. Test it, then Accept or Reject

    Supported Goals

    Tell the AI what you want with [9] Set Goal:

    • Speed - "Make this run in under 10 seconds"
    • VRAM - "Make this work on 8GB VRAM"
    • Quality - "Improve image quality without changing speed much"
    • Debug - "Fix why this keeps giving black images"

    Multi-Machine / Network Workflows (v0.4)

    Press [D] for distributed workflow optimization.

    If your workflow uses remote services (Kobold LLM, remote ComfyUI, TTS, STT), Performance Lab can:

    • Health check all your endpoints
    • Measure latency to each service
    • Find bottlenecks (which machine is slow?)
    • Suggest parallelization (what can run simultaneously?)
    • Generate smart prompts with your machine specs so the AI can suggest things like "move STT to machine 2" or "lower context size on the LLM server"

    Included Network Nodes

    The ComfyUI_NetworkServices custom node pack includes:

    | Node | What It Does | |------|-------------| | KoboldLLM | Connect to Kobold instances for text generation | | RemoteComfyUI | Run workflows on remote ComfyUI servers | | LocalGenerator | Universal REST API node (70+ presets) | | EndpointHealthCheck | Monitor service availability |

    Supported Services (70+ presets)

    Image/Video: ComfyUI, Automatic1111, Forge, InvokeAI, Fooocus, AnimateDiff, SVD, Mochi, CogVideo...

    LLM: KoboldCpp, Ollama, llama.cpp, Text-Gen-WebUI, vLLM, LMDeploy, TGI, LocalAI...

    STT: Whisper, Faster-Whisper, whisper.cpp, WhisperX, NeMo ASR, Vosk...

    TTS: Coqui TTS, XTTS, AllTalk, Silero, Piper, Bark, Tortoise, StyleTTS2...

    And more: Embeddings, Vision models, Audio generation, Upscaling...

    ComfyUI Nodes (v0.7.0)

    After installation, find 27 nodes in the "⚡ Performance Lab" category. Every node has a ? tooltip explaining exactly how to use it.

    ⭐ Start Here (NEW in v0.7.0)

    The easiest way to get started! These nodes are designed for new users:

    | Node | What It Does | |------|-------------| | ⚡ One-Click Optimize | Everything in one node! Toggle test/production mode | | 📚 Quick Start Guide | Interactive guide with topics to learn | | 🔘 Test Mode Toggle | Simple on/off for your whole workflow | | 🔍 Auto Detect GPU | Automatically detects your GPU and suggests settings | | 🔍 Model Detector | Detects model type from checkpoint name |

    Monitoring Nodes

    | Node | What It Does | |------|-------------| | ⏱️ Start Timer | Place at START of workflow to begin timing | | 📊 Performance Report | Place at END to see duration, VRAM, and report | | 💾 VRAM Monitor | Check GPU memory at any point (passthrough) |

    Quick Optimize Nodes

    | Node | What It Does | |------|-------------| | 📐 Cap Resolution | Limit dimensions for faster testing | | 🔢 Reduce Steps | Lower sampling steps (15-20 for testing) | | 📦 Reduce Batch | Force batch size to 1 for VRAM savings | | 🎯 Optimize CFG | Auto-adjust CFG for your model type | | 🚀 Speed Test Preset | All optimizations in one node | | 💾 Low VRAM Preset | Optimized for 6GB/8GB/12GB GPUs |

    Analysis Nodes

    | Node | What It Does | |------|-------------| | 🔍 Workflow Analyzer | Analyze workflow and get suggestions | | 🔧 Black Image Fix | Diagnose why you're getting dark images | | 📊 Compare Results | See before/after % improvement |

    LLM Integration

    | Node | What It Does | |------|-------------| | 🤖 Generate LLM Prompt | Create prompts for Claude/GPT/Gemini |

    Utility Nodes

    | Node | What It Does | |------|-------------| | 📝 Show Text | Display any text output in node | | 🔀 A/B Switch | Toggle between two any-type inputs | | 🔢 Int A/B Switch | Toggle between two integers | | 🔢 Float A/B Switch | Toggle between two floats |

    Meta-Workflow Nodes (NEW in v0.6.0)

    Use these to test and analyze other workflows without leaving ComfyUI:

    | Node | What It Does | |------|-------------| | 📂 Load Workflow | Load a workflow JSON file for analysis | | ▶️ Queue Workflow | Send a workflow to ComfyUI for execution | | 🏁 Benchmark Runner | Run multiple times and average results |

    Network Nodes (NEW in v0.6.0)

    Discover and monitor AI services on your network:

    | Node | What It Does | |------|-------------| | 🏥 Endpoint Health | Check if a network service is running | | 🔍 Network Scanner | Find ComfyUI, Ollama, A1111, etc. on your network |

    Example: Track Execution Time

    [⏱️ Start Timer] → [Your Workflow...] → [📊 Performance Report]
        ↓                                          ↓
     timer ─────────────────────────────────→  (shows duration)

    Example: Test Another Workflow

    [📂 Load Workflow] ─→ [🔍 Workflow Analyzer] ─→ [📝 Show Text]
        │
        └─→ [▶️ Queue Workflow] ─→ [📝 Show Text (status)]

    Master Workflow (Recommended)

    A single unified workflow with all features is included:

    • examples/performance_lab_master.json

    This workflow includes toggles for all modes - just load it and customize!

    Model Tuner

    Press [M] to auto-detect your model and apply optimal settings.

    | Model | Resolution | Steps | CFG | Best Samplers | |-------|------------|-------|-----|---------------| | SD 1.5 | 512x512 | 25 | 7.5 | dpmpp_2m, euler_ancestral | | SDXL | 1024x1024 | 30 | 7.0 | dpmpp_2m_sde, euler_ancestral | | SD3 | 1024x1024 | 28 | 4.5 | euler, dpmpp_2m | | Flux Dev | 1024x1024 | 28 | 3.5 | euler, ipndm | | Flux Schnell | 1024x1024 | 4 | 1.0 | euler |

    LLM Enhancer

    Press [L] for advanced AI context generation:

    • Node Catalog - Shows all your installed ComfyUI nodes
    • System Specs - Includes your GPU, VRAM, CPU info
    • Knowledge Base - Common issues and solutions
    • Mod Validation - Validates AI responses before applying
    • Conversation Memory - Remembers context across sessions

    Writing Custom Mods

    Create a Python file in the mods/ folder:

    # mods/my_optimization.py

description = "What this mod does (shown in menu)"

def apply(content):
    """
    content: The workflow as a Python dict
    Returns: Modified dict, or None if no changes
    """
    for node in content.get("nodes", []):
        # Your optimization logic here
        pass
    return content

    Troubleshooting

    ComfyUI not detected?

    • Make sure ComfyUI is running before starting Performance Lab
    • Check connection with [C] Test Connection
    • Default URL is http://127.0.0.1:8188

    Clipboard not working on Linux?

    • Install xclip: sudo apt install xclip
    • Or manually copy/paste the prompts

    Mod not working?

    • Test syntax: python -m py_compile mods/yourmod.py
    • Make sure apply() returns the modified content

    File Structure

    ComfyUI_PerformanceLab/
├── __init__.py              # 22 ComfyUI nodes (main integration)
├── performance_lab.py       # CLI application
├── llm_enhancer.py          # AI context generation
├── model_tuner.py           # Model detection & tuning
├── workflow_utils.py        # Workflow analysis
├── distributed_optimizer.py # Multi-machine support
├── services_config.py       # Network services config
├── logging_config.py        # Logging setup
├── install.py               # One-step installer
├── mods/                    # Your mod collection
│   ├── vram_optimizer.py
│   ├── bypass_upscalers.py
│   └── ...
├── examples/                # Example workflows
│   └── test_workflow_runner.json
├── custom_nodes/            # Additional node packs
│   └── ComfyUI_NetworkServices/
└── tests/                   # Test suite

    Version History

    v0.7.0 - User Experience & ComfyUI Manager Integration

    • 5 NEW beginner-friendly nodes in "⭐ Start Here" category:
      • One-Click Optimize: Single toggle controls everything
      • Quick Start Guide: Interactive 5-topic guide for new users
      • Auto Detect GPU: Automatically detects GPU and suggests settings
      • Model Detector: Detects SD1.5/SDXL/Flux from checkpoint name
      • Test Mode Toggle: Simple on/off for your whole workflow
    • Added tooltips to ALL input parameters (hover for help)
    • New pyproject.toml for proper ComfyUI Manager integration
    • Improved requirements.txt for better dependency management
    • Added __all__ exports for cleaner module interface
    • Interactive "Getting Started" example workflow
    • Total: 27 nodes now available

    v0.6.0 - Full Node Integration & Meta-Workflows

    • 22 ComfyUI nodes with comprehensive DESCRIPTION tooltips (? button)
    • Meta-Workflow nodes: Load, Queue, and Benchmark other workflows
    • Network nodes: Endpoint Health Check, Network Scanner
    • All capabilities now accessible directly in ComfyUI (no CLI needed)
    • Quick Optimize nodes: Cap Resolution, Reduce Steps/Batch, CFG, Presets
    • A/B Switch nodes for easy testing vs production toggle
    • Example meta-workflow included

    v0.4.2 - ComfyUI Native Integration

    • Native ComfyUI nodes in "Performance Lab" category
    • Performance Timer & Monitor nodes for tracking execution time
    • Workflow Analyzer node for instant analysis
    • Show Metrics node for VRAM/GPU info
    • Launch Performance Lab node to open CLI
    • Before/After comparison table with % change
    • VRAM sparkline visualization
    • Black Image Diagnostic [X] menu option
    • Dynamic impact estimates in Quick Actions
    • Markdown report export
    • "Already tried" context in LLM prompts
    • Auto-queue after pasting mods
    • Improved GitHub install feedback

    v0.4.0 - Multi-Machine Distributed Optimization

    • New [D] Distributed menu for multi-machine pipelines
    • ComfyUI_NetworkServices node pack (70+ service presets)
    • KoboldLLM, RemoteComfyUI, LocalGenerator, HealthCheck nodes
    • Machine profiling and bottleneck detection
    • Services configuration file support
    • Improved error handling and logging
    • Test suite with pytest

    v0.3.0 - LLM Enhancer

    • Advanced AI context generation
    • Node Catalog, System Specs, Knowledge Base
    • Mod Validation, Conversation Memory

    v0.2.0 - Model Tuner

    • Auto-detection for SD1.5, SDXL, Flux, SD3
    • Model-specific presets and LoRA recommendations

    v0.1.0 - Initial Release

    • Quick Actions, Benchmark Mode, Smart Suggestions
    • Presets, Dashboard, Multi-platform clipboard

    License

    MIT - Use freely, modify as needed.

    Made with ⚡ for the ComfyUI community