Both versions are unrunnable

### Your question

I have latest portable version and desktop versions. Neither will work now. The desktop won't accept the directory I choose- on 5th attempt at install and the dependency errors for portable version have rendered my version useless. I have spent 100x the amount of time I've spent creating, on fixing this app. Days and sometimes weeks to get up and running. There has to be some answer. I refuse to reinstall ever model I have and every custom node one by one. I say I don't have the time but I've probably spent over 2 months just putzing around with python, dependency errors, node won't install. etc. Now I am stuck. How can I possibly get these up and running and still use my existing models and nodes

### Logs

```powershell
one of the many.. I've been working with GPT4o and Gemini for last 4 days trying to get a working version. ## ComfyUI-Manager: installing dependencies done.
[2025-01-12 14:07:00.575] ** ComfyUI startup time: 2025-01-12 14:07:00.575685
[2025-01-12 14:07:00.576] ** Platform: Windows
[2025-01-12 14:07:00.576] ** Python version: 3.12.7 (tags/v3.12.7:0b05ead, Oct 1 2024, 03:06:41) [MSC v.1941 64 bit (AMD64)]
[2025-01-12 14:07:00.576] ** Python executable: K:\ComfyUI_windows_portable\python_embeded\python.exe
[2025-01-12 14:07:00.577] ** ComfyUI Path: K:\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-Manager
[2025-01-12 14:07:00.577] ** User directory: K:\ComfyUI_windows_portable\ComfyUI\user
[2025-01-12 14:07:00.578] ** ComfyUI-Manager config path: K:\ComfyUI_windows_portable\ComfyUI\user\default\ComfyUI-Manager\config.ini
[2025-01-12 14:07:00.579] ** Log path: K:\ComfyUI_windows_portable\ComfyUI\user\comfyui.log
[2025-01-12 14:07:02.460] [MaraScott] Initialization

Prestartup times for custom nodes:
[2025-01-12 14:07:02.466] 0.0 seconds: K:\ComfyUI_windows_portable\ComfyUI\custom_nodes\rgthree-comfy
[2025-01-12 14:07:02.466] 0.0 seconds: K:\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI_MaraScott_Nodes
[2025-01-12 14:07:02.466] 0.0 seconds: K:\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-Easy-Use
[2025-01-12 14:07:02.466] 5.0 seconds: K:\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-Manager
[2025-01-12 14:07:02.466]
[2025-01-12 14:07:04.648] Total VRAM 8192 MB, total RAM 32433 MB
[2025-01-12 14:07:04.648] pytorch version: 2.4.1+cu124
[2025-01-12 14:07:04.649] Set vram state to: LOW_VRAM
[2025-01-12 14:07:04.649] Device: cuda:0 NVIDIA GeForce RTX 3070 Ti Laptop GPU : cudaMallocAsync
[2025-01-12 14:07:11.116] Traceback (most recent call last):
[2025-01-12 14:07:11.117] File "K:\ComfyUI_windows_portable\ComfyUI\main.py", line 136, in
[2025-01-12 14:07:11.118] import execution
[2025-01-12 14:07:11.119] File "K:\ComfyUI_windows_portable\ComfyUI\execution.py", line 13, in
[2025-01-12 14:07:11.120] import nodes
[2025-01-12 14:07:11.120] File "K:\ComfyUI_windows_portable\ComfyUI\nodes.py", line 22, in
[2025-01-12 14:07:11.121] import comfy.diffusers_load
[2025-01-12 14:07:11.122] File "K:\ComfyUI_windows_portable\ComfyUI\comfy\diffusers_load.py", line 3, in
[2025-01-12 14:07:11.122] import comfy.sd
[2025-01-12 14:07:11.122] File "K:\ComfyUI_windows_portable\ComfyUI\comfy\sd.py", line 10, in
[2025-01-12 14:07:11.123] from .ldm.cascade.stage_c_coder import StageC_coder
[2025-01-12 14:07:11.124] File "K:\ComfyUI_windows_portable\ComfyUI\comfy\ldm\cascade\stage_c_coder.py", line 19, in
[2025-01-12 14:07:11.124] import torchvision
[2025-01-12 14:07:11.125] File "K:\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torchvision\__init__.py", line 10, in
[2025-01-12 14:07:11.126] from torchvision import _meta_registrations, datasets, io, models, ops, transforms, utils # usort:skip
[2025-01-12 14:07:11.126] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[2025-01-12 14:07:11.130] File "K:\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torchvision\_meta_registrations.py", line 163, in
[2025-01-12 14:07:11.133] @torch.library.register_fake("torchvision::nms")
[2025-01-12 14:07:11.133] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[2025-01-12 14:07:11.138] File "K:\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\library.py", line 654, in register
[2025-01-12 14:07:11.139] use_lib._register_fake(op_name, func, _stacklevel=stacklevel + 1)
[2025-01-12 14:07:11.140] File "K:\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\library.py", line 154, in _register_fake
[2025-01-12 14:07:11.141] handle = entry.abstract_impl.register(func_to_register, source)
[2025-01-12 14:07:11.142] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[2025-01-12 14:07:11.146] File "K:\ComfyUI_windows_portable\python_embeded\Lib\site-packages\torch\_library\abstract_impl.py", line 31, in register
[2025-01-12 14:07:11.148] if torch._C._dispatch_has_kernel_for_dispatch_key(self.qualname, "Meta"):
[2025-01-12 14:07:11.149] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[2025-01-12 14:07:11.152] RuntimeError: operator torchvision::nms does not exist
```

### Other

_No response_

Reproduce the Windows portable startup failure from main.py through execution.py, nodes.py, comfy/sd.py, and comfy/ldm/cascade/stage_c_coder.py, using the logged Python, PyTorch, and CUDA versions. Inspect the dependency environment and installation path; done would require a confirmed startup path that preserves existing models and custom nodes.

Written by the indexing model from the issue text.