This extension provides a method for "text-presampling" or prompt upsampling using LLM models like TIPO and DanTagGen, aimed at Stable Diffusion WebUI, ComfyUI, and Forge users. It enhances prompts by generating detailed character tags, core tags, and additional elements, improving the quality and specificity of AI-generated images.

How It Works

The extension integrates with the KGen library to leverage TIPO (Text to Image with text presampling for Prompt Optimization) and DanTagGen. TIPO is an LLM system designed to generate detailed prompts from tags or natural language, offering flexibility in tag design. DanTagGen is specifically trained for generating Danbooru-style tags, beneficial for models trained on Danbooru datasets. The extension allows users to control tag and natural language output length, specify banned tags, and define prompt formats using placeholders like <|characters|> and <|generated|>.

Quick Start & Requirements

• SD-WebUI/Forge: Install via the extension tab. A new NL prompt text box appears under the positive prompt, with a TIPO accordion nearby.
• ComfyUI: Clone the repository into the custom_node folder. A TIPO node will be available under utils/promptgen.
• Dependencies: For ComfyUI, requires ComfyUI-Custom-Scripts to be installed in custom_node.

Highlighted Details

• Supports TIPO and DanTagGen for prompt upsampling.
• Allows customization of tag/NL length, banned tags, and prompt format.
• Offers control over upsampling timing (before/after other prompt processes).
• Includes an example workflow for ComfyUI.

Maintenance & Community

• Developed by KohakuBlueleaf.
• Links to KGen library and DanTagGen HuggingFace model card are provided.

Licensing & Compatibility

• The README does not explicitly state a license.

Limitations & Caveats

• The license is not specified, which may impact commercial use or integration into closed-source projects.