By releasing this, I strive to make the best possible starting point for Krea 2 workflows whether you're a beginner or advanced user. Settings are made easy to find, and it includes an optional prompt enhancer using Krea's own text encoder. It also includes the Fantastic Lora Loader which is the best custom node I've ever used to stack loras. I'm including a version with and without Face Fix, a nodepack which I discovered today as I'm writing this. It's super useful but it required me to manually upgrade my huggingface-hub python package to from version 0.24.0 to version 0.34.0 in order to get Comfy to start. Not sure if this is a common issue or not.
Description
FAQ
Comments (3)
What's the deal with all the "replace text" nodes and the last one that says "cough"?
Even though I try to prompt the LLM not to do it, it sometimes does it anyway, which is saying stuff like "Here's the enhanced prompt:" instead of just going into the prompt. Those several "replace text" nodes are all variations of things it's replied with, so each node ensures that it'll never say that. The only reason I included the one that says "cough" is because it also looks for two line breaks to go with it, which means it'll be far less likely to remove that word from an actual prompt.
To add on to my last comment, I want to clarify that the LLM/text encoder wrote a perfectly good enhanced prompt for me but it began to write new paragraph but instead just randomly said "cough" at the end. I figured "oh well, that's the problem with 4b LLMs". I hit "run" again, but then it began to spit out gibberish after that. I restarted Comfy and things went back to normal. I think that was a symptom of a memory issue, and that "Replace text" node might as well be deleted.