ComfyUI-Krea2T-Enhancer From Capitan01R Boosts Image Text Accuracy

ComfyUI-Krea2T-Enhancer is a custom node designed to improve how closely Krea2 diffusion models follow text prompts. It works by adjusting the internal conditioning path during the sampling process without altering the original prompt text. The release offers a simple way to enhance prompt adherence directly inside existing image generation workflows.

Developer Capitan01R who made another model enhancer node created this tool as a continuation of community work to solve prompt adherence issues. They built a one-knob experimental patch that applies a controlled adjustment inside the text-fusion stage. This approach avoids the need to rewrite prompts or add extra conditioning nodes to achieve desired results.

Key project features and benefits

This tool is for people running local image generation workflows who want more precise control over their outputs. Users can benefit from a setup that requires minimal adjustments to their existing model loader and sampler configuration. It provides a straightforward method to fix minor prompt adherence issues without overcomplicating the generation process.

Developer notes and current status

Developer capitan01R notes that this is an experimental proof of concept release and advises user discretion. The node is designed specifically for Krea2 models using a 12 x 2560 text-conditioning layout and will skip the patch if the model does not match. Users should start with a strength of 1.00 and adjust accordingly, as combining the tool with LoRAs may require lower values.

"The advanced node restores every temporary runtime patch after each model call and does not store debug counters or step-local state in the model config." Source: GitHub