This is the workflow I use to make all the images for my models and loras. It's a beast. I'm providing this AS-IS. No warranties or guarantees.

This Workflow Contains NSFW Elements

Features

• random resolution function

• big fat lora stack

• easy reach 1st stage controls to find that sweet sweet seed

• HiresFix and Ultimate upscale and an optional 3rd stage upscale for skin detail

• High quality Detailing using crop&stich node + depth anywhere control net for 1 person

• Detailing for all primary anatomy :)

• Detailers for Multi-person renders

• Civitai friendly Save (with a bit of manual adjustment necessary)

• A separate Manual Inpaint workflow for those hard to fix places.

This will take time to setup to your system. There's lots of models, bbox and seg models, controlnet models, cfgs, schedulers, samplers. This is big. It's not for the faint of heart.

Sometimes the primary 1person detailer gets weird. Just switch to the multi-person detailer for that iteration. It works 99% of the time and the output is higher quality than the standard detailers most of the time.

The random resolution function is not connected. It's just a toy to play with and easy to connect.

I'm using a group muter to control the detailers. You won't break anything to keep everything turned on, but this thing isn't fast, so the less running, the better.

I'm using this on a 12gb card. It runs fine. No idea if it'll run on less VRAM. You will need to find upscale models that work well for your use-case. That can make or break your VRAM and it makes a big performance difference. The perturbed attention and self-attention guidance are bypassed by default. They slow things down. There are empty nodes to connect a refiner. I don't use it much. Keep the manual inpaint off or it'll fire off to nothing. I'm sure there's more. Have fun.

I'll help if I can. Don't ask me to troubleshoot comfy. I don't think I'm using any fancy nodes that can't be found on the manager. Let me know if I'm lying.