I decided to post my workflow because i figured out a few things that irritated me for a long time.

You certainly won't use it a i set it up, but it can give you a few ideas too.

major ideas here include:

• one button hi-res fix subgraph. it will create downscaled image, upscale and put through second pass. it's much cleaner/smaller than other solutions i saw

• one button prompt enhance with clip model doing the job. it has multiple presets, clean enhance, nsfw, nsfw++ etc. (you need uncensored clip, because normal will often refuse to write a risqué prompt)

• model and lora selector that keeps hashes and feeds them to the resulting image. It isn't elegant and requires a setup, but if you publish your images to civitai and want to have all models used automatically recognized, that's the only way i know.

• conditional loras. a subgraph that loads a lora when prompt uses a specific phrase.

• image saver that actually saves .webp with full workflow and metadata. For the longest time i was sure .webp doesn't work because two other nodes were broken ;p

• one button face detailer. for those pictures with small subject. it's rarely needed, but if you never used inpaint with sam3, it shows how it works.

• prompt node that allows dynamic prompts and wildcards, because i'm addicted to randomness.

this workflow won't make much sense to a novice or someone who doesn't want to open/edit a subgraph, but for an amateur tinkerer it might give you some ideas.