#8752 Video file selection required

This message means the software's input list is empty: when you click Start Execution, it detects that no valid video/audio file has been loaded, so the task is not started. It is not a crash, and it is usually caused by the file not being actually added to the list, or by the path/name not being recognized.

Follow these steps:

1. Click the Select Video/Audio button in the upper-left corner, and pick a file. After selection, the file name must appear in the list box at the top of the main window. If the list stays empty, the selection was not accepted.
2. If you used the Folder checkbox, you must also click the selection button to choose an existing folder that contains video files; checking Folder alone with no folder chosen produces the same prompt.
3. Rename the file and move it to a shallow path: only English letters, digits and underscores, for example D:/videotrans/video1.mp4. Files downloaded from YouTube/Bilibili often have extremely long names plus emoji, and the software will silently fail to add them.
4. Make sure the software has been fully extracted from the archive, and that sp.exe is run from the extracted folder on disk (e.g. D:/Downloads/win-pyvideotrans-v4.18 is fine), not from inside the compressed package.
5. Confirm the file is actually playable and has an audio track — a video with no audio track will also fail to be processed, although it would normally be added to the list first.
6. Test once with a short, simply named mp4 to confirm the interface is working; then retry the original file.

If a file is clearly displayed in the list and it still reports "Video file selection required", the cause cannot be determined. Please provide the last 30 lines of the newest log file in the logs/ folder under the software installation directory (D:/Downloads/win-pyvideotrans-v4.18/logs/).

Please refer to the related documentation:

• Quick start
• FAQ

Turned off "Enable Cuda" and reslected file. Seems to be working at the moment. No other changes were made to the default settings.

Turning off CUDA only changes the compute device; the software does not use CUDA to determine whether a file can be loaded into the list. Therefore, the actual reason this run started successfully is that the file was actually added to the list this time. Turning off CUDA simply stopped the software from using the GPU for speech recognition, so if there was a CUDA-related failure earlier, switching to CPU avoided it.

Current state: running on CPU, which is stable but noticeably slower in the speech recognition stage.

If you want to re-enable CUDA acceleration later, first confirm all of the following:

1. You have an NVIDIA GPU (AMD/Intel or integrated graphics cannot use CUDA).
2. The NVIDIA driver is up to date, CUDA 12.8 and cuDNN 9.x are installed, and the cuDNN bin/lib directories are in the system Path.
3. In Menu → Tools/Options → Advanced Options → Speech Recognition (ASR), set Compute Type to float16.
4. VRAM is sufficient: large-v3 needs at least 8GB free VRAM; large-v3-turbo needs about 6GB. If you hit CUDA out of memory / Unable to allocate, lower the model (large-v3 → medium → small) and set Beam Size and Best Of to 1.

Keep the following points in mind regardless:

• Keep the video file name short and use only English letters, digits, and underscores. Avoid special characters and emoji in the path.
• Do not repeatedly translate the same file without checking Clear Output, or the old srt/wav cache from the previous run will be reused.
• Switch to CPU mode if an error reappears. Report the exact error text from the log.

Please refer to the related documentation:

• https://pyvideotrans.com/gpu
• https://pyvideotrans.com/faster

PC is a HP 840Z workstation with 2x Xeon Processors and 128gb of ram. NVIDIA video card w/8gb ram. However, the user/Human didn't see that the first attempt to select the file didn't work. Had I done that, all might be well. We will see how this works. I have 5 more files to test so will note how long it takes to translate. Thank you.

The 8GB of VRAM is the main constraint, and it directly determines which models you can use. The dual-Xeon CPU and 128GB of RAM make CPU mode practical for you; it is slower, but it considerably reduces the risk of exhausting VRAM. When you process the remaining 5 files, follow these settings:Keep Enable CUDA off for now. With 2 Xeon CPUs and 128GB of RAM, CPU-mode faster-whisper is stable, and you avoid CUDA/cuDNN configuration problems and out-of-memory crashes.If you later turn CUDA back on, use the large-v3-turbo model (large-v3 needs at least 8GB free VRAM and will almost certainly run out of memory on your card), and set Compute Type to float16, Beam Size 1, Best Of 1.In Menu → Tools/Options → Advanced Options → General, set Max Concurrent GPU Tasks to 1. For a batch of 6 files, set Batch Size for Video Translation to 1 or 2; running too many tasks at the same time will compete for VRAM and RAM.When you process several files at once, the program runs them interleaved concurrently. The total elapsed time is therefore not equal to the sum of the per-file processing times, so use a single video for timing when estimating speed.This Video file selection required prompt appears only when the file has not been added to the list. After you select a file, confirm that its name appears in the top list box before you click Start Execution. If the prompt appears again, provide the last 30 lines of the newest log file in logs/.Please refer to the related documentation:https://pyvideotrans.com/besterhttps://pyvideotrans.com/adv

How well with a 8gb vram NVIDA card work

Workable, but 8GB of VRAM is the minimum tier for speech recognition. If you size the models correctly, it is stable rather than fast, and it only accelerates one stage of the pipeline.

What fits and what does not on 8GB:

Practical points when using an 8GB card:

1. Select CUDA Acceleration, and in Advanced Options → Speech Recognition (ASR) choose the large-v3-turbo model with Compute Type float16, Beam Size 1, Best Of 1. If you insist on large-v3, set Compute Type to int8 to reduce the memory footprint; if it still crashes, step down to medium.
2. Set a limit of one GPU task at a time: Advanced Options → General → Max Concurrent GPU Tasks = 1; set Batch Size for Video Translation to 1. Running several videos at the same time will use up the VRAM.
3. Do not use the VRAM-heavy built-in channels: Higgs-audio-v3 (needs more than 10GB of VRAM) and VibeVoice-ASR (15GB+) are not suitable; VibeVoice is also unstable on an 8GB card. For dubbing, use Qwen3-TTS, F5-TTS, OmniVoice, or Confucius, which fit within this range, or use Edge-TTS, which requires no local VRAM.
4. Denoise and vocal/BGM separation are CPU-only by default; the 128GB of RAM and the dual Xeon CPU handle them. If you have a GPU and want GPU processing, you can switch to the zipenhancer denoise model and the mel_band_roformer separation model in Advanced Options (these models use audio.cpp); however, this will compete with ASR for VRAM, so do not run them simultaneously.
5. Expect the GPU to be idle most of the time. Only the speech recognition stage uses CUDA; translation is network-bound, TTS is either network-bound or light, and final muxing uses the CPU. Therefore, a faster card mainly shortens the first stage of each file, not the whole job.

Overall judgment: for 5–6 files with clear audio in common languages, an 8GB card with large-v3-turbo plus CUDA is a reasonable configuration. If you hit repeated CUDA out-of-memory errors, either switch to the medium model or run on the CPU. With your hardware, the CPU fallback is not slow enough to be painful.

Please refer to the related documentation:

• Configuring CUDA and cuDNN on Windows
• faster-whisper (Local, Built-in)
• Recommended best configuration
• Advanced Options