TACK ComfyUI Launcher

A desktop tool for managing multiple isolated ComfyUI installations.

Create, clone, snapshot, and version-control separate ComfyUI environments — each with independent plugins, Python versions, and CUDA configurations.

English | 繁體中文

---

Features

Multi-Environment Management

• Create fully isolated ComfyUI environments with dedicated Python virtual environments
• Clone existing environments to experiment without risk
• Rename and delete environments freely
• Merge packages and plugins between environments (add or replace strategy)
• Each environment has its own venv, ComfyUI installation, and plugin set

Plugin Management

• Install custom nodes from Git URLs with automatic dependency resolution
• Enable / Disable plugins by toggling directories (no data loss)
• Update individual plugins or batch-update all at once
• 6-step conflict analysis before installation:
  1. Dependency extraction (requirements.txt + AST parsing of install.py)
  2. Pip dry-run simulation
  3. Version comparison against current environment
  4. Critical package detection (torch, numpy, transformers, etc.)
  5. Risk classification (GREEN / YELLOW / HIGH / CRITICAL)
  6. User-facing recommendations

Snapshot & Rollback

• Point-in-time snapshots capturing pip freeze, ComfyUI commit, Python/CUDA versions, custom node states, and config backups
• One-click restore to any previous snapshot
• Auto-snapshots created before clone, merge, and version-switch operations
• Configurable max snapshots with automatic pruning

Version Control

• Browse and switch between ComfyUI tags and branches
• View recent commit history
• Update ComfyUI to the latest version with a single click

Launch & Runtime

• Start / stop ComfyUI instances per environment
• Configurable per-environment launch settings:
  • Cross-attention mode (pytorch, split, quad, sage, flash, etc.)
  • VRAM management (gpu_only, high, normal, low, cpu)
  • Network settings (listen IP, port, CORS, TLS)
  • Custom CLI arguments
• PyTorch engine switching — change CUDA version on the fly
• Pre-launch diagnostics — dependency checks, conflict detection, duplicate node scanning
• Log viewer with export capability
• Multiple environments can run simultaneously on different ports

GPU & Version Detection

• Automatic GPU detection via nvidia-smi
• Recommended CUDA tag selection based on driver version
• Python version management (3.10 – 3.13) with embeddable builds
• PyTorch version fetching from official wheel indexes

Self-Update

• Checks GitHub for new versions
• One-click update with progress tracking
• Automatic restart after update

Internationalization

• Full UI support for English and Traditional Chinese (zh-TW)

---

Screenshots

The GUI is a modern HTML5 SPA rendered inside a PySide6 desktop window.

The application includes pages for:

• Home — Quick actions and folder shortcuts
• Environments — Create, clone, edit, delete environments
• Launch — Start/stop ComfyUI with advanced settings and diagnostics
• Plugins — Install, manage, and analyze custom nodes
• Versions — Browse and switch ComfyUI versions
• Snapshots — Create, restore, and manage snapshots

---

Installation

Prerequisites

• Windows 10/11
• Python 3.10+ (or use the bundled embedded Python)
• Git (or use the bundled embedded Git)
• NVIDIA GPU recommended (CPU mode available)

Quick Start

1. Download the latest release from GitHub Releases or clone the repository:

   git clone https://github.com/tackcrypto1031/tk_comfyui_start_tool.git
   cd tk_comfyui_start_tool

2. Install dependencies (skip if using the bundled Python):

   pip install -r requirements.txt

3. Launch the application:

   Double-click start.bat, or run manually:

   pythonw launcher.py

4. Create your first environment — click "Create" in the Environments page, choose a ComfyUI version, and the tool handles venv creation, cloning, PyTorch installation, and dependency setup automatically.

Bundled Tools

The tool can include embedded executables for zero-dependency setup:

tools/
├── git/cmd/git.exe        # Embedded Git
└── python/pythonw.exe     # Embedded Python

If these are present, the launcher uses them automatically. Otherwise, it falls back to system-installed Git and Python.

---

Usage

GUI Mode (Default)

Double-click start.bat or run:

pythonw launcher.py

CLI Mode

The tool also provides a full command-line interface:

# Environment management
python launcher.py env list
python launcher.py env create my_env --branch master
python launcher.py env create my_env --tag v0.3.0
python launcher.py env clone main my_experiment
python launcher.py env delete old_env
python launcher.py env merge source_env target_env --strategy add
python launcher.py env info my_env
python launcher.py env analyze my_env path/to/custom_node

# Snapshot management
python launcher.py snapshot create my_env --reason "before major change"
python launcher.py snapshot list my_env
python launcher.py snapshot restore my_env snap-20250101-120000-000000
python launcher.py snapshot delete my_env snap-20250101-120000-000000

# Version management
python launcher.py version list my_env
python launcher.py version switch my_env v0.3.0
python launcher.py version update my_env

# Launch ComfyUI
python launcher.py launch start my_env --port 8188
python launcher.py launch stop my_env
python launcher.py launch status

---

Configuration

Settings are stored in config.json at the project root:

Key	Default	Description
default_env	"main"	Default environment name
environments_dir	"./environments"	Where environments are stored
models_dir	"./models"	Shared models directory
snapshots_dir	"./snapshots"	Snapshot storage
max_snapshots	20	Max snapshots per environment (auto-prunes oldest)
auto_snapshot	true	Auto-snapshot before clone/merge
auto_open_browser	true	Open browser after launching ComfyUI
default_port	8188	Default ComfyUI port
theme	"dark"	UI theme
language	"zh-TW"	UI language (en or zh-TW)

Shared Models

All environments share a single models/ directory with subdirectories:

models/
├── checkpoints/
├── loras/
├── vae/
├── controlnet/
├── clip/
├── embeddings/
└── upscale_models/

The tool auto-generates extra_model_paths.yaml for each environment to point to this shared location.

---

Project Structure

tk_comfyui_start_tool/
├── launcher.py              # Entry point (GUI or CLI)
├── cli.py                   # CLI commands (Click)
├── start.bat                # Windows GUI launcher
├── config.json              # Configuration
├── VERSION.json             # Version & update info
├── requirements.txt         # Python dependencies
├── src/
│   ├── core/                # Business logic
│   │   ├── env_manager.py       # Environment lifecycle
│   │   ├── snapshot_manager.py  # Snapshot backup & restore
│   │   ├── version_manager.py   # GPU detection, Python/CUDA mgmt
│   │   ├── version_controller.py# ComfyUI version switching
│   │   ├── conflict_analyzer.py # 6-step plugin conflict analysis
│   │   ├── comfyui_launcher.py  # Process start/stop
│   │   ├── diagnostics.py       # Dependency & conflict checks
│   │   └── updater.py           # Self-update system
│   ├── gui/
│   │   ├── bridge.py            # QWebChannel Python-JS bridge
│   │   └── web/                 # HTML5 SPA frontend
│   │       ├── index.html
│   │       ├── js/
│   │       │   ├── app.js       # SPA routing
│   │       │   ├── bridge.js    # JS API client
│   │       │   ├── i18n.js      # Translations
│   │       │   └── pages/       # Page modules
│   │       └── css/
│   ├── models/              # Data models
│   └── utils/               # Utilities (pip, git, process, fs)
├── environments/            # Runtime environments
├── models/                  # Shared model files
├── snapshots/               # Environment snapshots
└── tools/                   # Bundled Git & Python

---

Architecture

The application is a PySide6 + QWebEngineView hybrid desktop app. The frontend is an HTML5 SPA that communicates with the Python backend through QWebChannel.

┌─────────────────────────────────────────────┐
│                  PySide6 Window              │
│  ┌───────────────────────────────────────┐  │
│  │         QWebEngineView (SPA)          │  │
│  │  ┌─────────┐  ┌───────────────────┐   │  │
│  │  │ Sidebar  │  │   Page Content    │   │  │
│  │  │  Nav     │  │                   │   │  │
│  │  └─────────┘  └───────────────────┘   │  │
│  └──────────────┬────────────────────────┘  │
│                 │ QWebChannel                │
│  ┌──────────────┴────────────────────────┐  │
│  │          Bridge (Python)              │  │
│  │  EnvManager | Launcher | Snapshots    │  │
│  │  Versions   | Plugins  | Updater      │  │
│  └───────────────────────────────────────┘  │
└─────────────────────────────────────────────┘

Long-running operations use an async worker pattern: the bridge returns a request_id immediately, and the frontend polls for progress and results.

---

Development

Running Tests

# All tests
pytest

# With coverage
pytest --cov

# Specific test file
pytest tests/test_core/test_env_manager.py

# Pattern matching
pytest -k test_create

Dependencies

Package	Purpose
PySide6	Desktop window & QWebChannel
click	CLI framework
rich	CLI formatting
gitpython	Git operations
psutil	Process management
pyyaml	YAML config files
packaging	Version comparison
requests	HTTP requests

---

License

This project is licensed under the Apache License 2.0 — see the LICENSE file for details.

---

Links

• GitHub Repository
• ComfyUI