#7172 [WinError 126] The specified module could not be found. Error loading

This WinError 126 when loading caffe2_nvrtc.dll almost always means the underlying NVIDIA CUDA/cuDNN runtime libraries that this DLL depends on are missing or not on the system PATH — not that the DLL itself is absent.

Please verify and fix in this order:

1. Confirm you actually have an NVIDIA GPU

   • Open CMD and run nvidia-smi. If it errors or shows nothing, there is no usable N-card and CUDA cannot be enabled at all.

2. Check the CUDA Toolkit and cuDNN versions

   • Run nvcc -V in CMD.
   • pyVideoTrans v4.11 (bundled torch build) requires CUDA Toolkit 12.8 or higher, and cuDNN 9.11 or higher.
   • If your CUDA is older (e.g. 11.x, 12.1, 12.4), this dependency error is expected.

3. Install/update the NVIDIA driver

   • Download the latest driver for your GPU from NVIDIA. CUDA 12.8+ requires a recent driver.

4. Add CUDA and cuDNN to the system PATH

   • Add the CUDA bin and lib\x64 directories, and the cuDNN bin directory, to the Windows environment variable Path.
   • Restart the software (and ideally the PC) after changing PATH.

5. Path must be ASCII-only

   • Your install path D:\Setup\PY4.11 is fine, but do not place the software under any folder containing Chinese characters, spaces, or special symbols.

6. If you do not need GPU acceleration

   • The software can run on CPU. Simply uncheck "Enable CUDA acceleration" and it will avoid loading the CUDA-dependent DLLs. Note: local Whisper and local TTS models will be much slower on CPU.

7. If CUDA is installed but still fails

   • Verify the bundled torch folder is intact (the folder D:\Setup\PY4.11\_internal\torch\lib\ should exist and not be blocked/quarantined by antivirus). Add the software folder to your antivirus whitelist.

Please refer to the related documentation:

• Configure CUDA and cuDNN on Windows
• Install CUDA 12.6 and cuDNN 9.8 on Windows