Generators

Generators Guide

TagForge includes two powerful generation engines: one for Prompts (text-to-image) and one for Vision (image-to-text).

🎨 Prompt Generator

This tool helps you create optimized prompts for Stable Diffusion, Midjourney, or FLUX.

Operation Modes

The behavior depends on your selected Persona:

1. Tag Prompts (Danbooru Style)

• Best For: Stable Diffusion 1.5, SDXL, Pony Diffusion.
• Output: Comma-separated tags.
• Example: 1girl, blue_eyes, solo, masterpiece, best quality, cinematic lighting

2. Natural Prompts (Prose Style)

• Best For: FLUX.1, Midjourney v6, DALL-E 3.
• Output: Rich, descriptive sentences.
• Example: "A cinematic shot of a warrior standing on a cliff edge, rain pouring down, neon lights reflecting off the wet pavement..."

[!TIP]
Mixing Styles: You can ask the "Natural Prompts" persona to include tags by adding instructions like "Include a tag list at the end" in your input.

---

👁️ Vision Analysis

TagForge supports Multimodal Vision, allowing the AI to "see" and analyze images.

Requirements

• You must use a model that supports vision (e.g., Gemini 1.5 Flash, Llama 3.2 Vision, Claude 3.5 Sonnet).

[!IMPORTANT]
Model Compatibility: If you select a text-only model (like Llama 3 8B), the "Attach Image" button will be visually disabled to prevent errors.

How to Use

1. Attach Image: Click the Paperclip icon to attach an image for visual analysis.
   • Supported Images: Select .jpg, .png, or .webp.
2. Select Persona:
   • Tag Vision: Extracts tags from the image (Reverse Tagging).
   • Natural Vision: Writes a detailed caption describing the image.
   • Standard Personas: Standard text-generation personas are available for manual prompting alongside an image.
3. Generate: The AI will process the attachments and provide the result based on your instructions.

Vision Override

[!WARNING]
For Advanced Users: If you are using a custom model that TagForge doesn't recognize as "Vision-Capable" (e.g. a new local model in LM Studio), you can force-enable the image button:
1. Go to Agent Manager.
2. Check the Vision Override box.

---

🎭 Personas Guide

Personas define how the agent behaves, speaks, and formats its output during generation.

Built-in Presets

TagForge comes with verified presets:

• Tag Prompts: Strictly outputs comma-separated tags. Rejects conversational filler.
• Natural Prompts: Writes descriptive, artistic paragraphs.
• Tag Vision: Inspects images and lists visible elements as tags.
• Natural Vision: Inspects images and describes them in natural language.

🛠️ Creating Custom Personas

You can create your own agents for any purpose in Settings -> Personas Tab.

Dynamic Injection: {input}

Use the {input} placeholder to insert your prompt into a specific part of the system instructions.

• With {input}: Your message replaces the tag (useful for templates).
• Without {input}: Your message is appended to the end of the conversation.

Example Template:

You are a character design expert. Take the user's description and expand it.
Concept: {input}