HartyBart posted at 6:32 AM Mon, 22 December 2025 - #4502467

Congratulations on learning ComfyUI. It's worth it.

Thanks @HartyBart. I've experimented with controlnet and depthmaps, with mixed results. When 2 or more characters are involved that are interacting with each other, it often messes up limbs.

The workflow I’ve now put together uses Qwen-image-edit-2509 with 2 Lora’s added. This gives good results, even on more complicated scenes consisting of more characters (often in NSFW situations ;-)), but the next stage truly makes the difference: I added a ‘Resample banding fix’ node that refers back to the original input image and prompt. This irons out a lot of artifacts and small distortions that were generated in the first stage and gives a crisp image. After that I added a VAE-decoder that upscales 2x while decoding. This speeds up the upscale process considerably!!

I will try certainly try the LifePortrait in my workflow!

This is what I've put together so far borrowing from sereval seperate workflows and adding a bit of my own.

Virtualbite.com - Creating the best vampire art you can imagine!

Follow me on Patreon @virtualbite
www.instagram.com/virtualbite_

www.virtualbite.com