I've been learning how to use generative AI in general and ComfyUI in particular for some time now, and I believe I finally have a workflow worth sharing.

It's based on various sources and a lot of experimentation, and works for any model which will accept a natural language prompt.

The core

Through leverage of wildcards it obtains endless diversity. The expanded wildcards get sent to Ollama. Through use of either a generic system prompt, for models like e.g. Flux Dev, or the specific Z-Image system prompt these expanded wildcards get transformed into a well-defined prompt.

Added goodies

• SeedVR2 Upscaling

• Color Grading

• Image enhancement

• Image saving with Civitai metadata

Used custom nodes

All can be installed through ComfyUI Manager

• ComfyUI-Impact-Pack

• ComfyUI Layer Style

• rgthree-comfy

• ComfyUI-Easy-Use

• ComfyUI-KJNodes

• ComfyUI-SeedVR2_VideoUpscaler

• ComfyUI_essentials

• RES4LYF

• Comfyui_TTP_Toolset

• comfyui-ollama

• ComfyUI-Inspire-Pack

• gguf

• mikey_nodes

• was-node-suite-comfyui

• CRT-Nodes

• comfyui_image_metadata_extension

The sampler and scheduler that are configured in the workflow seem to work best for me, but as always YMMV ;-)

Additional info

A good source for free LUTs (used in the Layer Style nodes) is https://freshluts.com, the specific Rec709 LUT I use is Rec.709 Film Look