Comfy keeps Crashing, despite RAM, CPU, VRAM noooot maxing out.

### Custom Node Testing

- [ ] I have tried disabling custom nodes and the issue persists (see [how to disable custom nodes](https://docs.comfy.org/troubleshooting/custom-node-issues#step-1%3A-test-with-all-custom-nodes-disabled) if you need help)

### Expected Behavior

Expected behaviour is Comfy official workflow for QWEN IMAGE to work. Thourgh all nodes
it does not.
Despite way enough GPU, CPU and RAM
(rtx 3090 and 64g VRam) but in any case they never maxed out at all. Tried it many times

### Actual Behavior

When we start /run the workflow, and reach the Load Diffusion Model part, it crashes
NO GPU MAX OUT
NO CPU maxxing out
NO RAM maxxing out
None of that after 4-5 experiments.

### Steps to Reproduce

Install comfyui
Use this workflow (added file)

[ComfyUI_Error2.json](https://github.com/user-attachments/files/22869986/ComfyUI_Error2.json)

The models are on the description on that text area of the workflow. On the left side.

### Debug Logs

```powershell
After these logs it says press a button to continue and close the window (and i see on the UI "reconnecting" in red showin that comfy crashed)
Logs:
## ComfyUI-Manager: installing dependencies done.
[yy25-xx-12 xx:00:yy.269] ** ComfyUI startup time: yy25-xx-12 ...
[yy25-xx-12 xx:00:yy.269] ** Platform: Windows
[yy25-xx-12 xx:00:yy.270] ** Python version: 3.13.6 (...)]
[yy25-xx-12 xx:00:yy.270] ** Python executable: E:\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\python_embeded\python.exe
[yy25-xx-12 xx:00:yy.274] ** ComfyUI Path: E:\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI
[yy25-xx-12 xx:00:yy.277] ** ComfyUI Base Folder Path: E:\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI
[yy25-xx-12 xx:00:yy.278] ** User directory: E:\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\user
[yy25-xx-12 xx:00:yy.280] ** ComfyUI-Manager config path: E:\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\user\default\ComfyUI-Manager\config.ini
[yy25-xx-12 xx:00:yy.286] ** Log path: E:\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\user\comfyui.log

Prestartup times for custom nodes:
[yy25-xx-12 xx:00:21.188] 2.4 seconds: E:\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-Manager
[yy25-xx-12 xx:00:21.188]
[yy25-xx-12 xx:00:22.767] Checkpoint files will always be loaded safely.
[yy25-xx-12 xx:00:22.980] Total VRAM 24xxx MB, total RAM 64xxx MB
[yy25-xx-12 xx:00:22.981] pytorch version: 2.8.0+cu129
[yy25-xx-12 xx:00:22.981] Set vram state to: NORMAL_VRAM
[yy25-xx-12 xx:00:22.982] Device: cuda:0 NVIDIA GeForce RTX 3090 : cudaMallocAsync
[yy25-xx-12 xx:00:23.995] Using pytorch attention
[yy25-xx-12 xx:00:25.853] Python version: 3.13.6 ()]
[yy25-xx-12 xx:00:25.854] ComfyUI version: 0.3.64
[yy25-xx-12 xx:00:25.878] ComfyUI frontend version: 1.27.xx
[yy25-xx-12 xx:00:25.880] [Prompt Server] web root: E:\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\python_embeded\Lib\site-packages\comfyui_frontend_package\static
[yy25-xx-12 xx:00:26.841] ### Loading: ComfyUI-Manager (V3.37)
[yy25-xx-12 xx:00:26.842] [ComfyUI-Manager] network_mode: public
[yy25-xx-12 xx:00:26.994] ### ComfyUI Revision: 170 [a125cd84] *DETACHED | Released on 'yy25-xx-12'
[yy25-xx-12 xx:00:27.013]
Import times for custom nodes:
[yy25-xx-12 xx:00:27.013] 0.0 seconds: E:\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\custom_nodes\websocket_image_save.py
[yy25-xx-12 xx:00:27.013] 0.2 seconds: E:\ComfyUI_windows_portable_nvidia\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-Manager
[yy25-xx-12 xx:00:27.014]
[yy25-xx-12 xx:00:27.086] [ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/model-list.json
[yy25-xx-12 xx:00:27.131] [ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/alter-list.json
[yy25-xx-12 xx:00:27.181] [ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/github-stats.json
[yy25-xx-12 xx:00:27.242] [ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/custom-node-list.json
[yy25-xx-12 xx:00:27.293] [ComfyUI-Manager] default cache updated: https://raw.githubusercontent.com/ltdrdata/ComfyUI-Manager/main/extension-node-map.json
[yy25-xx-12 xx:00:28.122] Context impl SQLiteImpl.
[yy25-xx-12 xx:00:28.123] Will assume non-transactional DDL.
[yy25-xx-12 xx:00:28.125] No target revision found.
[yy25-xx-12 xx:00:28.161] Starting server

[yy25-xx-12 xx:00:28.162] To see the GUI go to: http://127.0.0.1:8188
[yy25-xx-12 xx:00:30.960] FETCH ComfyRegistry Data: 5/99
[yy25-xx-12 xx:00:34.519] FETCH ComfyRegistry Data: xx/99
[yy25-xx-12 xx:00:37.980] FETCH ComfyRegistry Data: 15/99
[yy25-xx-12 xx:00:41.464] FETCH ComfyRegistry Data: yy/99
[yy25-xx-12 xx:00:45.312] FETCH ComfyRegistry Data: 25/99
[yy25-xx-12 xx:00:49.156] FETCH ComfyRegistry Data: 30/99
[yy25-xx-12 xx:00:50.077] Exception in callback _ProactorBasePipeTransport._call_connection_lost()
handle:
[yy25-xx-12 xx:00:52.684] FETCH ComfyRegistry Data: 35/99
[yy25-xx-12 xx:00:56.658] FETCH ComfyRegistry Data: 40/99
[yy25-xx-12 xx:01:00.184] FETCH ComfyRegistry Data: 45/99
[yy25-xx-12 xx:01:03.786] FETCH ComfyRegistry Data: 50/99
[yy25-xx-12 xx:01:07.666] FETCH ComfyRegistry Data: 55/99
[yy25-xx-12 xx:01:11.142] FETCH ComfyRegistry Data: 60/99
[yy25-xx-12 xx:01:14.828] FETCH ComfyRegistry Data: 65/99
[yy25-xx-12 xx:01:19.315] FETCH ComfyRegistry Data: 70/99
[yy25-xx-12 xx:01:22.904] FETCH ComfyRegistry Data: 75/99
[yy25-xx-12 xx:01:26.416] FETCH ComfyRegistry Data: 80/99
[yy25-xx-12 xx:01:26.821] got prompt
[yy25-xx-12 xx:01:26.847] Using pytorch attention in VAE
[yy25-xx-12 xx:01:26.848] Using pytorch attention in VAE
[yy25-xx-12 xx:01:27.053] VAE load device: cuda:0, offload device: cpu, dtype: torch.bfloat16
[yy25-xx-12 xx:01:27.116] Using scaled fp8: fp8 matrix mult: False, scale input: False
[yy25-xx-12 xx:01:27.480] Requested to load QwenImageTEModel_
[yy25-xx-12 xx:01:27.499] loaded completely 9.5367431640625e+25 7909.737449645996 True
[yy25-xx-12 xx:01:27.506] CLIP/text encoder model load device: cuda:0, offload device: cpu, current: cuda:0, dtype: torch.float16
[yy25-xx-12 xx:01:30.023] FETCH ComfyRegistry Data: 85/99
```

### Other

_No response_