Due to how comfy_extra nodes are imported as modules, importing of torchaudio_ffmpeg pyd libraries fails since there are no available dll libraries. That resulted in the error:
```
Loading FFmpeg6
Failed to load FFmpeg6 extension.
Traceback (most recent call last):
File "C:\Users\ishim\Tools\ComfyUI\venv\Lib\site-packages\torio\_extension\utils.py", line 116, in _find_ffmpeg_extension
ext = _find_versionsed_ffmpeg_extension(ffmpeg_ver)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\ishim\Tools\ComfyUI\venv\Lib\site-packages\torio\_extension\utils.py", line 108, in _find_versionsed_ffmpeg_extension
_load_lib(lib)
File "C:\Users\ishim\Tools\ComfyUI\venv\Lib\site-packages\torio\_extension\utils.py", line 94, in _load_lib
torch.ops.load_library(path)
File "C:\Users\ishim\Tools\ComfyUI\venv\Lib\site-packages\torch\_ops.py", line 1357, in load_library
ctypes.CDLL(path)
File "C:\Program Files\Python312\Lib\ctypes\__init__.py", line 379, in __init__
self._handle = _dlopen(self._name, mode)
^^^^^^^^^^^^^^^^^^^^^^^^^
FileNotFoundError: Could not find module 'C:\Users\ishim\Tools\ComfyUI\venv\Lib\site-packages\torio\lib\libtorio_ffmpeg6.pyd' (or one of its dependencies). Try using the full path with constructor syntax.
```
Which it repeats for version 5 and 4 as well.
This PR fixes this by adding paths in the "PATH" environment variable as dll library directories using os.add_dll_directory() which was introduced with Python version 3.8.
There might be need for some additions for conda environments but as I myself do not use this and it is not among the official methods of running ComfyUI apart from with Intel GPU, I leave the task of adding that to someone with the experience in that area.
* add LoadImageOutput node
* add route for input/output/temp files
* update node_typing.py
* use literal type for image_folder field
* mark node as beta
I'm not not sure which arches are supported yet. If you see improvements in
memory usage while using --use-pytorch-cross-attention on your AMD GPU let
me know and I will add it to the list.
* Fix for running via DirectML
Fix DirectML empty image generation issue with Flux1. add CPU fallback for unsupported path. Verified the model works on AMD GPUs
* fix formating
* update casual mask calculation
