Can you show a wf example for anima, please?
Should we use any other node for anima other than the ones you show or use different launch parameters? I'm getting an error when I reach the ksampler, I can send the error if necessary
Sorry for the late answer.
I recommend to add --disable-dynamic-vram option during launch, and Use https://github.com/BobJohnson24/ComfyUI-INT8-Fast custom node.
Please refer README.md.
If you get error on KSampler, try
- disable torch.compile (usually, it has issues.)
- disable sage attention node. (sage attention is quite difficult to install, and sometime has issues.)
I managed to solve it, it seems sage wasn't installed properly, needed some python libraries that didn't came with the portable version of comfy. Then I had some trouble with the INT8 (W8A8) node to load the model, had to fix it manually on the code by changing this instance around line 800 and around 900 something:
return super().patch_weight_to_device(key, device_to, inplace_update, return_weight, force_cast)
to return super().patch_weight_to_device(key, device_to) because I think comfy updated the thing that node relied on to call that instance so it was broken, at least for me