This workflow is basically all I'm looking for in a workflow :

My configuration :

Intel I7 / 32 GB RAM / RTX 3070 8GB VRAM

SSD 2TB / PYTHON 3.12.6

NO XFORMERS. I'm not against xformers. I installed it on a previous configuration with Python 3.10. I recently changed Python version by adding a new venv to my existing Comfyui. It took me long enough on personal time so I just didn't feel like going further.

Prompt formatting :

ai formatting using your own selfhosted ollama host

Requires Ollama, recommended with a uncensored llm like dolphin-mistral :  https://ollama.com/

In my set up, I set the OLLAMA_HOST system environment variable, but this is not mandatory. If you serve your ollama on your own PC, 127.0.0.1 will do just fine.

Multiple upscalers :

choose your own upscale : LDSR or not

just before making your choice, to use LDSR or not, there is a pause in progression, where you choose to keep going or cancel

There is also a node to free some RAM : at this stage, there is a raw export already saved

4x Nomos Ultimate, then 4x Ultrasharp, then downscale

By the way, I download my upscalers from :  https://openmodeldb.info/

Other :

separated clips : please be careful to load the right model

separated VAE : depending on you using the model's vae, please bypass the node

Destined to SD3.5, but ok with fp16 compatible models or more, if you know what you're doing :)

Last but no least, if you can't achieve a download via "Install Missing Custom Nodes", please post a comment. but you should be alright with the ComfyUi Manager node.

Enjoy !!!