🏷️ What is TagForge?
TagForge is a powerful, dedicated desktop application designed specifically for creating perfect Stable Diffusion prompts with the help of modern Large Language Models (LLMs). Whether you're generating SFW or NSFW content, TagForge transforms your simple ideas into detailed, comma-separated tag lists optimized for image generation.
Available on GitHub
✨ Features at a Glance
🧠 Intelligence
🎨 Specialized Generator: Turn vague ideas into optimized, comma-separated tag lists for Stable Diffusion or rich prose for FLUX/Midjourney.
👁️ Vision Analysis: Drag-and-drop images to extract tags or generate detailed captions using multimodal models.
💬 Contextual Chat: A coding-capable assistant with persistent session history and markdown rendering.
🛡️ Architecture
🔒 Zero-Trust Security: API keys are encrypted at rest using Windows DPAPI. Keys never leave your device except to hit the API endpoint.
⚡ Native Performance: Built on .NET 9 with heavy threading. The UI stays buttery smooth, even while generating complex batches.
🔌 Supported Providers
Google Gemini: High-speed, high-quality multimodal reasoning.
Groq: Lightning-fast inference for Llama & Mixtral.
OpenRouter: Unified gateway to OpenAI, Anthropic, and premium models.
HuggingFace: Access the massive open-source library via Inference API.
LM Studio: Connect to your own private inference server.
Ollama: Run Llama 3, Mistral, and more directly on your CPU/GPU.
🎭 Persona System
Customize the AI's identity to fit your workflow:
🖊️ System Roles: Define the AI's expertise (e.g., "You are a Danbooru tagging expert").
🔄 Dynamic Injection: Use the {input} placeholder to create reusable templates.
⚙️ Built-in Presets:
Tag Prompts: The classic tagging engine for Stable Diffusion.
Natural Prompts: For generating rich, descriptive prose.
Tag Vision: Inspects images and generates tag lists.
Natural Vision: Inspects images and generates detailed captions.
Description
🚀 New Features
👁️ Vision Intelligence:
VLM Support: Full support for Vision-Language Models (e.g., Llama 3.2 Vision, Qwen-VL). Attach images into the prompt generator for analysis.
Smart Detection: Automatically detects vision capabilities based on model metadata. Includes a Manual Override toggle for custom or unrecognized models.
Auto-Prompts: Context-aware injection of system prompts (e.g., "Describe every aspect...") when images are attached or vision personas are active.
🎭 Persona System 2.0:
Retired legacy defaults in favor of 4 high-performance presets: Natural Prompts, Tag Prompts, Natural Vision, and Tag Vision.
📜 Advanced Logging:
Complete overhaul of the console with collapsible details, log level coloring, and "One-Click Copy" functionality for easier debugging.