Text to Image Workflow with WIldcard Node by bkidderz

This workflow gives you a a few preset wildcard generators and a background moodboard. Using them, you can create a character sheet and environment, which is then passed to a small (2B) LLM which will convert the sheet into a natural language prompt using the following instructions:

You are given facts of a character sheet and setting/environment. Rewrite the facts into one flowing natural-language image prompt. 
Rules:
 - Start with the character description, taking the elements and composing them into natural language sentences.
 - Describe the head, body, clothes, props, setting, pose, camera, lighting, palette in that order.
 - Do not include headings such as Face: Lips: etc, just describe.
 - Be succinct but do not discard facts.
 - Do not drop the art-style paragraph; put it first or last, intact in meaning.
 - No comma-separated tag lists. No "masterpiece, best quality".
 - No markdown, no quotes, no preamble. Output only the prompt.

Up front, this workflow might not work very well at first. If you skip the LLM, you're going to get a prompt that looks like a character sheet, and this can lead to some odd results.

If you want the LLM, you might want to use an AI to assist with your issues, assuming it doesn't go smoothly. I picked a node that I'm actually not too thrilled with and depending on feedback I get, I might try something else. But for now, I'm really happy that I got it working. I've decided to provide a zip of the custom node CONTAINING the gguf. Unzip it into your custom_nodes folder, and run this command from within the folder itself (while at a prompt with your venv activated, if applicable)

pip install -r requirements.txt

Having the GGUF in the folder saves you the trouble of hunting it down on HF.

Locations of downloads are noted in the workflow and the custom node repository has a very helpful guide. If you are having trouble though I recommend asking an LLM.

This is really a very preliminary hash-out of the concept I wanted to created, and it has some rough edges but it works, so.. I may or may not find time to fix it. It's already taken me several evenings to get this far.

I really liked bkidderz' custom node and I wanted to use it, and after taking the trouble to make a workflow that used it, he mentioned that he normally passes it through Qwen to make it natural language. That was a real forehead slap moment and I decided I had to go the extra mile and make that before I reached the publishing finish line. So here we are! And I hope you enjoy!