Is it possible to respect flags such as --lowvram, --novram, and --cpu?

[Nubsy]

As far as I can tell, these flags have little to no effect on any of the nodes included in this pack. I don't know how big an ask this is, but it would be very useful to me even if processing took ages. I have 11GB of VRAM, and am unable to run most of the models locally. I am not concerned about speed, because most of the trial and error involved happens during image creation, I would be fine queuing up images and letting the computer sit for a long time to convert them to meshes.

I've also tried the LayerUtility: Purge VRAM node which doesn't seem to have any effect on these nodes. I wonder if something similar could be implemented within this pack?

[MrForExample]

Hi @Nubsy I understand that those MVs/3D gen models could take a lot of vram, and it would indeed be nice to be able to run it with low vram To your question:

• Most of the nodes need GPU & CUDA to run
• I may be able to find some simple way to reduce vram between nodes but not within nodes (That would take a lot of time to optimize the model itself)
• If you are installed xformer package successfully (If you are on Windows or Linux then it should be installed automatically), then you should be able to run more than half of the example workflows
• Also, within each workflow, some nodes allow you to choose model size like InstantMesh, you may also try that

[Duodecimus]

Using a node like https://github.com/willblaschko/ComfyUI-Unload-Models will help a little bit. ComfyUI doesn't unload models while running, which means some of the example workflows can require a base of 6+gb just to load all the models in.