ntsCuda

Physically accurate NTSC & VHS analog video artifact emulator

GPU acceleration · Compare slider · VHS audio simulation · Drag & drop · 30+ formats · Modern dark UI

Based on zhuker/ntsc · Base Version JargeZ/ntscQT · MIT License

---

✨ Features

⚡ GPU / CUDA Acceleration Zero-config. Install CuPy — done. ntsCuda auto-detects your GPU and activates hardware acceleration. Falls back to CPU silently if no GPU is found. A live CUDA badge appears in the UI when GPU is active. FFT ringing → 5–20× faster Color space → 2–5× faster	🚀 Vectorized Processing All per-scanline Python loops replaced with batched NumPy operations across the entire signal pipeline. Composite lowpass / VHS luma → 40–60× Phase noise, chroma decode → 15–50×	🎨 30+ Formats Open 19 video and 11 image formats. Save processed images in 5 output formats with quality control. All / Video / Image filter categories in the picker.
🔊 VHS Audio Simulation Optional audio effect that simulates the lo-fi, degraded sound of VHS tapes — saturation boost, lowpass filter, and tape noise. Toggle and fine-tune directly from the UI.	🖱️ Drag & Drop Drop any supported video or image directly onto the window — no dialog needed. Full drag & drop support for instant loading.	🌗 Dark Modern UI Clean, dark-themed interface with consistent styling, CUDA status indicator, and optimized slider layout. Appearance theme follows your OS automatically.

---

🖥️ UI Preview

Modern dark UI with CUDA badge, VHS audio controls, and parameter sliders

---

🔀 Compare Mode

Vertical split-screen with draggable slider — see original vs. processed side by side

---

🎬 How to Use

Load a file, adjust parameters, and render — it's that simple

---

Supported Formats

	Input	Output
Video	mp4 mkv avi webm gif mov flv wmv mpg mpeg ts m4v 3gp ogv mxf vob m2ts mts f4v rm rmvb	mp4 · mkv (lossless FFV1)
Image	png jpg jpeg webp bmp tiff tif tga ppm pgm pbm	png · jpg (95%) · bmp · tiff · webp (90%)

---

Simulated Artifacts

Artifact	Description
Dot Crawl	Animated crawling dots along high-saturation horizontal edges from luma/chroma multiplex
Ringing	Echo fringes around sharp edges — oscillation artifact common in early VTR hardware
Color Bleeding	Y/C timing mismatch shifts chroma relative to luma, creating blurry color halos
Rainbow Effects	Cross-color from imperfect luma/chroma separation on fine patterns and subtitles
Chrominance Noise	Colored specks in dark, saturated areas from low-light or multi-gen dubs
Head Switching Noise	Horizontal noise band at frame bottom from VHS head switching transients
LP/EP Degradation	Reduced tape speed artifacts — banding, smearing, increased noise sensitivity
Luminance Noise	White noise in brightness channel from worn tape, long cables, or dirty heads
Oversaturation	Chroma amplitude beyond 120 IRE causes color to bleed past object boundaries
VHS Audio	Lo-fi sound simulation — saturation, lowpass filter, tape noise

---

Installation

Windows

# Install FFmpeg (if not already installed)
choco install ffmpeg

# Clone and set up
git clone https://github.com/AllastorV/ntsCuda
cd ntsCuda
python -m venv venv
.\venv\Scripts\activate
pip install -r requirements.txt

# Run
python ntsCuda.py

Next time:

cd ntsCuda && .\venv\Scripts\activate && python ntsCuda.py

Or use the provided batch files:

install.bat    # First-time setup
start.bat      # Launch the app

macOS & Linux

# macOS
brew install ffmpeg

# Ubuntu / Debian
sudo apt install ffmpeg libxcb-xinerama0

# Clone and set up
git clone https://github.com/AllastorV/ntsCuda ~/ntsCuda
cd ~/ntsCuda
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

# Run
python ntsCuda.py

Next time:

cd ~/ntsCuda && source venv/bin/activate && python ntsCuda.py

Apple Silicon (M1 / M2)

brew install pyqt@5
git clone https://github.com/AllastorV/ntsCuda ~/ntsCuda
cd ~/ntsCuda
python3 -m venv venv
cp -R /opt/homebrew/Cellar/pyqt@5/5.15.6/lib/python3.9/site-packages/* \
      ./venv/lib/python3.9/site-packages/
source venv/bin/activate
pip install --pre -i https://pypi.anaconda.org/scipy-wheels-nightly/simple scipy
pip install --ignore-installed -r requirements.m1-temp.txt
venv/bin/python ntsCuda.py

GPU / CUDA Acceleration (optional)

Install CuPy matching your CUDA version. ntsCuda auto-detects it — no other setup required.

nvcc --version          # check your CUDA version

pip install cupy-cuda12x   # CUDA 12.x
pip install cupy-cuda11x   # CUDA 11.x
pip install cupy-cuda102   # CUDA 10.2

If no GPU is detected, ntsCuda runs in CPU-only mode automatically. A green CUDA badge appears in the UI when GPU acceleration is active. See requirements-cuda.txt for details.

---

Usage

Control	Description
Open File	Browse for a video or image — or just drag & drop it onto the window
URL	Load an image directly from a web URL
Seed	Deterministic randomness — same number always produces the same artifacts
Height	Processing resolution. Lower = faster preview
Effect ON	Toggle the NTSC effect on/off without stopping the render
Compare	Vertical split-screen with draggable slider — original left vs. processed right
x2 Upscale	Output at double resolution (nearest-neighbor)
Lossless	Encode with FFV1 lossless codec (outputs .mkv)
Pro Mode	Reveals advanced low-level signal parameters
Live Preview	Show every rendered frame (default: every 10th)
VHS Audio	Enable lo-fi VHS audio simulation with adjustable parameters
Refresh	Re-render current frame with current slider values
Pause	Pause mid-render, tweak sliders, resume — creates time-varying effects
Stop	Abort the current render
Export / Import Config	Share your exact slider configuration as JSON

Tips

Cut before you process. Run only the clip you need — not the whole file.

Sliders work live. Drag them during render (no pause) to create smoothly evolving effects.

Seed is king. Same seed = identical artifact pattern every time. Great for reproducible results.

Preview fast, render full. Use a low Height for preview, raise it to source resolution for final output.

Drag & drop. Drop any supported file directly onto the window for instant loading.

---

Performance

All per-scanline Python loops replaced with batched NumPy operations. GPU via CuPy is optional and zero-config.

Function	Optimization	CPU Speedup	GPU Speedup
composite_lowpass	lfilter(axis=1) batch	~40–60×	—
composite_lowpass_tv	lfilter(axis=1) batch	~40–60×	—
composite_preemphasis	single-stage lfilter	~40×	—
vhs_luma_lowpass	3-stage + highpass batch	~50×	—
vhs_chroma_lowpass	U+V channels batched	~60×	—
vhs_sharpen	in-place broadcast	~30×	—
video_chroma_phase_noise	sin/cos vectorized	~15×	—
chroma_from_luma	cumsum + batch decode	~50×	—
ringing / ringing2	FFT (NumPy → CuPy)	2–5×	5–20×
bgr2yiq / yiq2bgr	NumPy broadcast	2–3×	2–5×

---

Bug Fixes

#	Fix
1	np.float deprecation → np.float32 / np.float64
2	Audio getattr fallbacks — no crash when audio stream is absent
3	composite_layer field parameter corrected (field=0, fieldno=0)
4	Null-guard on preview renderer — no crash before first file is opened
5	Video track slider signal leak fixed on file re-open
6	Template loading moved to background thread — UI no longer freezes
7	status_string NameError in Renderer.run() when paused before first frame
8	PNG export now saves the current processed frame instead of the cover image
9	float division by zero crash on startup fixed
10	2x upscale rendering corrected

---

Alternatives

Project	Notes
NTSC-RS	Rust — significantly faster · OpenFX & After Effects plugins
NtscQT+	Fork with additional parameters
valadaptive/ntscQT	Fork with various bug fixes
The Signal (paid)	AE plugin, same algorithm — free alternative: ntsc-rs

---

Community

RGM's ntscQT Presets — community preset collection

Complete guide by Jonah Longoria