CUDA/NVENC Enhanced Fork

The CUDA-supported version of the VHS video emulation program NTSCQT.I'm not very good at coding, so it's an AI-supported project. You can access the original project via this: JargeZ/ntscqt.

This fork brings optional GPU acceleration (CUDA/CuPy) and a faster FFmpeg streaming pipeline (NVDEC/NVENC) to the original ntscQT. You can still use the classic CPU path — all GPU features are opt‑in and have safe fallbacks.

New In This Fork

• CUDA acceleration (optional, toggle in UI)
  • GPU implementations for heavy stages: Ringing/Ringing2, BGR↔YIQ conversion, composite chroma lowpass (lite), VHS luma/chroma lowpass, VHS chroma vertical blend.
  • Falls back to CPU when CUDA is unavailable.
• NVENC/NVDEC streaming pipeline (optional)
  • Streams raw frames directly to FFmpeg and encodes with h264_nvenc (no temp files), while copying/encoding audio in the same pass.
  • Uses NVDEC (GPU decode) + scale_cuda when available; automatically falls back to the legacy reader if not.
• Drag & drop file opening.
• Dark mode toggle that persists independently from system theme.
• Audio handling improvements
  • For MP4 outputs, unsupported input codecs are automatically transcoded to AAC — otherwise audio is copied.
  • If FFmpeg is missing, the app writes video via OpenCV and attempts a post‑mux when FFmpeg later becomes available.

Requirements

• Python with the repo requirements installed.
• FFmpeg with NVENC/NVDEC on PATH (for best performance).
  • Windows (PowerShell): choco install ffmpeg -y or winget install ffmpeg.ffmpeg
  • Verify: ffmpeg -hide_banner -encoders | findstr nvenc should list h264_nvenc.
• CUDA + CuPy (optional, for GPU compute): pip install cupy-cuda12x (CUDA 12.x).

Quick Start

1. Install dependencies (and optional CuPy):
   • pip install -r requirements.txt
   • pip install cupy-cuda12x (optional, for CUDA compute)
2. Ensure FFmpeg is installed and on PATH (see above).
3. Run the app: python ntscQT.py
4. In the UI, open a video (or drag & drop), then enable:
   • Use CUDA (experimental)
   • Use NVENC (NVIDIA)
5. For maximum speed:
   • Disable LivePreview
   • Keep Render Height reasonable; optionally enable x2 nearest upscaling

Performance Tips

• Both toggles (CUDA + NVENC) should be ON for best results.
• Higher render resolutions increase GPU utilization (within VRAM limits).
• NVENC is tuned for speed (preset p1, constQP, qp≈22). You can increase qp to 24–26 for even faster encodes.

Troubleshooting

• FFmpeg not found
  • The app falls back to a CPU‑only writer; video renders without audio. Install FFmpeg to regain audio + NVENC speed.
• No audio in MP4 output
  • With FFmpeg present, the app will transcode unsupported input codecs to AAC automatically. Verify FFmpeg is on PATH.
• Low GPU usage
  • Ensure both toggles are ON, CuPy is installed, and the render resolution is not too small. Confirm h264_nvenc is available.