StabilityMatrix

Manages installation, updates, and execution of 10+ Stable Diffusion UI packages (ComfyUI, AUTOMATIC1111, InvokeAI, Fooocus, etc.) through a polymorphic BasePackage architecture with Git-based version control. Each package type (BaseGitPackage, BasePackage subclasses) implements platform-specific installation logic, dependency resolution, and launch configurations. The system handles package discovery, version tracking, and isolated execution environments per package instance.

Detects GPU hardware (NVIDIA CUDA, AMD ROCm, Intel Arc, Apple Metal) and automatically provisions Python virtual environments with matching PyTorch builds and CUDA/ROCm toolchain versions. Implements platform prerequisite detection (CUDA 11.8/12.1 availability, cuDNN versions) and selects optimal PyTorch wheel variants (CPU, CUDA 11.8, CUDA 12.1, ROCm 5.7, etc.) based on detected hardware. Uses Python subprocess isolation and venv module for environment creation.

Organizes downloaded models into package-specific folders (models/Stable-diffusion, models/Lora, models/VAE, etc.) with automatic subdirectory creation. Implements symlink strategy to share models across multiple package instances without duplication (e.g., symlink models/Stable-diffusion → shared-models/Stable-diffusion). Handles platform-specific symlink creation (Windows junction points vs Unix symlinks) and validates symlink integrity on startup.

Generates platform-specific launch scripts (batch files on Windows, shell scripts on Linux/macOS) with environment variable injection for GPU acceleration, Python paths, and package-specific settings. Implements launch configuration templates per package type (ComfyUI requires specific port configuration, AUTOMATIC1111 requires specific API flags, etc.). Executes launch scripts in isolated subprocess with real-time output streaming to UI.

Validates platform prerequisites (Python version, CUDA/ROCm availability, Git installation) before package installation and provides remediation guidance. Implements prerequisite detection via system API calls (registry on Windows, environment variables on Linux, system frameworks on macOS). Generates installation guides for missing prerequisites (e.g., 'Download CUDA 12.1 from nvidia.com'). Supports multiple Python versions and validates compatibility with package requirements.

Integrates CivitAI API for browsing, searching, and filtering 100k+ community-trained Stable Diffusion models (checkpoints, LoRAs, VAEs, embeddings) with metadata caching and local model import. Implements paginated API queries with filtering by model type, base model version, and rating. Downloaded models are automatically organized into local model folders (models/Stable-diffusion, models/Lora, etc.) with metadata JSON for UI display. Supports direct model download from CivitAI URLs with progress tracking.

Orchestrates end-to-end text-to-image generation workflows by translating UI parameter cards (prompt, negative prompt, sampler, steps, CFG scale, seed) into package-specific API calls (AUTOMATIC1111 txt2img endpoint, ComfyUI node graph execution). Implements parameter validation, preset management, and result caching. Supports batch generation with parameter sweeps (e.g., multiple seeds, CFG scales). Results are saved to local output folders with metadata JSON (prompt, model, parameters) for later retrieval.

Provides visual node graph builder for ComfyUI workflows with drag-and-drop node creation, connection validation, and serialization to ComfyUI JSON format. Implements node type registry with input/output type matching to prevent invalid connections. Executes workflows by sending JSON to ComfyUI API and polling for completion. Supports workflow templates, parameter overrides, and result streaming with progress callbacks.

Provides dynamic parameter card UI for text-to-image generation with input validation, preset saving/loading, and real-time parameter preview. Cards include controls for prompt, negative prompt, model selection, sampler, steps, CFG scale, seed, and resolution. Implements preset serialization to JSON files in user config directory, enabling parameter reuse across sessions. Supports parameter history and quick-access favorite presets.

Implements cross-platform desktop UI using Avalonia (XAML-based, .NET framework) with MVVM pattern for separation of concerns. ViewModels (e.g., PackageManagerViewModel, LaunchPageViewModel) handle business logic and state management, while Views (XAML) handle rendering. Uses dependency injection for service integration (IPackageManager, IModelManager, etc.). Supports design-time data for XAML preview and hot-reload during development.

Manages application startup, shutdown, and navigation between pages (PackageManager, LaunchPage, Settings, etc.) using a navigation service pattern. Implements startup sequence: detect platform prerequisites, initialize settings, load installed packages, initialize services. Uses MainWindowViewModel as root navigation hub with page-specific ViewModels. Handles graceful shutdown with cleanup (stopping running packages, saving state).

Persists application settings (installation paths, UI preferences, inference parameters) to JSON files in user config directory using a Settings model with serialization/deserialization. Implements localization system with language resource files (RESX or JSON) supporting multiple languages (English, Chinese, Japanese, etc.). Language selection is persisted in settings and applied at runtime via resource binding in XAML.

Implements CI/CD workflows (GitHub Actions) for building platform-specific installers (Windows MSIX, macOS DMG, Linux AppImage) and distributing updates via auto-update system. Builds are triggered on release tags and produce platform-specific artifacts. Auto-update system checks for new versions at startup and downloads/installs updates in background without interrupting user workflow. Uses semantic versioning for version comparison.