This is the workflow I'm using with Ain't Another Head Swap Lora.
https://civarchive.com/models/2305269/qie2511-aint-another-head-swap?modelVersionId=2593793
Description
VRAM utility optimization, dedicated for LoRA v4.1.
FAQ
Comments (6)
had an error
min(): Expected reduction dim to be specified for input.numel() == 0. Specify the reduction dim with the 'dim' argument.
no idea how to fix
It seems you did not draw the mask in Mask Editor in Image 1.
This workflow turned out to be quite interesting. I adopted a couple of approaches for my own processes. The author created a very complex mask, which can be done with two nodes: SolidMask + FeatherMask. This is much better. For example, if a face touches the top edge of the frame, the mask shouldn't blur at the top, and this can be easily fixed by setting the top value to 0 in FeatherMask. I also noticed that the best results are achieved with a lora influence of 0.4-0.5.
SAM3Segment nodes via ComfyUI-RMBG are red regardless of updating everything including the node. I am unsure what I can replace them with as likely the specific node is outdated to a new one by the same creator. Trying v4 of this workflow.
Guys! There a few ways of gain decent depth image. SAM3 is one of them. Just adopt the workflow to your ComfyUI capabilities. I've had problem with SAM3 nodes till i removed the folders from custom_nodes and reinstalled them by git clone ... method. I published the workflow because someone asked me. This is side project for me. The main is LoRA.
always: out of memory error ( and I run comfy on 24gb rtx 4090)
and also when I get it to work it was never accurate.
I suggest everyone to use the workflow here