Qwen Image Edit silently crashes with PyTorch CUDA 13 but works with CUDA 12 on RTX 3090

### 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

The Qwen Image Edit workflow should run normally on PyTorch CUDA 13 without closing ComfyUI, as it does in my CUDA 12 environment on the same computer.

### Actual Behavior

Qwen Image Edit works correctly in my PyTorch CUDA 12 environment on the same computer.

However, when using ComfyUI Windows Portable with Python 3.13 and PyTorch 2.9.1+cu130, ComfyUI closes silently shortly after loading the Qwen text encoder.

The final visible log lines are:

Requested to load QwenImageTEModel_
loaded completely; 7910.28 MB loaded, full load: True

Shortly afterward, python.exe closes without a Python traceback, and the batch window shows:

Press any key to continue...

### Steps to Reproduce

1. Start ComfyUI Windows Portable with Python 3.13 and PyTorch 2.9.1+cu130.
2. Open the Qwen Image Edit workflow.
3. Load an input image.
4. Press Queue.
5. Wait for the Qwen text encoder to finish loading.
6. ComfyUI closes silently shortly afterward.

I also tried starting ComfyUI with:

--disable-pinned-memory --disable-async-offload --disable-cuda-malloc

The crash still occurred.

### Debug Logs

```powershell
The complete terminal log is attached in the “Other” section as comfyui_qwen_full_log.txt.

The final visible log lines before the silent crash are:

Requested to load QwenImageTEModel_
loaded completely; 7910.28 MB loaded, full load: True

Immediately afterward, python.exe exits silently. No Python traceback, CUDA out-of-memory error, or other error message is displayed.
```

### Other

System information:

- Operating system: Windows
- GPU: NVIDIA GeForce RTX 3090 24GB
- CPU: AMD Ryzen 5 5500
- System RAM: 32GB
- Python: 3.13
- PyTorch: 2.9.1+cu130
- ComfyUI installation: Windows Portable
- NVIDIA driver: Studio Driver

The same Qwen Image Edit workflow works correctly in my other environment using Python 3.12.3 and PyTorch CUDA 12.x on the same computer.

No CUDA out-of-memory error or Python traceback is displayed.

This behavior appears similar to issue #11803:
https://github.com/Comfy-Org/ComfyUI/issues/11803

I can provide the workflow JSON, Windows Event Viewer report, or a crash [dump]([url](

[comfyui_qwen_full_log.txt](https://github.com/user-attachments/files/30371849/comfyui_qwen_full_log.txt)

Start by reproducing the Qwen Image Edit workflow in the Windows Portable environment with Python 3.13 and PyTorch 2.9.1+cu130, then inspect comfyui_qwen_full_log.txt and compare the behavior with issue #11803. Done means the workflow completes without python.exe exiting silently after QwenImageTEModel_ loads.

Written by the indexing model from the issue text.