This should be good out of the box - everything else is ComfyUI basics which you will have to learn. No support given here. Figure it out.

Based on NRDXs BFS Lora and Workflow

QWEN Edit 2511 + BFS Lora optimized for 16GB VRAM ~30-60 sec generations.

• No more / less weird twirly zig zag pattern hair (depending on inputs)

• Good facial obstruction, works for BJ, Facial, Masks, Glasses, Blindfolds etc.

To achieve good face swaps with Facial Obstruction like penis, cum, etc. your reference images should have similar resolution and aspect.

Other than that:

Optional Upscale (recommended, changing latent size fucks up the image)

Change steps between 4-12, works good, I use 6

Just load workflow and install missing stuff.

https://civarchive.com/models/2027766?modelVersionId=2556739

For the connoisseurs:

I suggest collecting face reference images, loading them via random image from folder (there are many nodes to achieve this)

Also collect some body reference images, loading them also via random image from folder.

Leave running overnight for couple thousand jobs, in the morning spend some time cherry picking the best. Make sure any seeds used are randomized or you'll get the same image many times.

Then I2V and goon to your hearts desire.