This repository provides custom nodes for ComfyUI, primarily focused on enhancing text conditioning for Stable Diffusion image generation. It targets users aiming to replicate results from AUTOMATIC1111's Stable Diffusion WebUI within ComfyUI, offering improved prompt parsing and weighting mechanisms for greater consistency.

How It Works

The core innovation is the CLIP Text Encode++ node, which implements prompt parsing and tokenization pipelines derived from AUTOMATIC1111's WebUI. This approach, particularly when using the A1111 or comfy++ parsers, aims to produce identical embeddings to the WebUI, facilitating cross-platform result reproduction. It supports advanced prompt features like alternating words, explicit breaks, and weight normalization, mirroring the WebUI's behavior for complex conditioning.

Quick Start & Requirements

• Installation: Via ComfyUI Manager, cloning the repository (git clone https://github.com/shiimizu/ComfyUI_smZNodes.git into ComfyUI/custom_nodes), or manual download.
• Prerequisites: ComfyUI.
• Links: ComfyUI Manager, ComfyUI Examples FAQ, ComfyUI Docs.

Highlighted Details

• Enables reproduction of AUTOMATIC1111 WebUI results in ComfyUI via identical prompt parsing and encoding.
• Offers multiple parser options (comfy, comfy++, A1111, full, compel, fixed attention) for fine-grained control.
• Includes a Settings node to dynamically adjust parameters like mean normalization and emphasis implementation.
• Supports advanced prompt syntax like AND for conditioning combination and weighted prompts.

Maintenance & Community

• Developed by shiimizu.
• Credits AUTOMATIC1111, comfyanonymous, vladmandic, and lllyasviel for foundational work.

Licensing & Compatibility

• The repository does not explicitly state a license in the provided README.

Limitations & Caveats

• LoRA syntax (<lora:name:1.0>) is not supported.
• Reproducing results with ancestral or SDE samplers may require specific scheduler configurations (e.g., ddim_uniform).
• The Settings node may require recreation after updates to prevent issues.