APZmedia Fast Image Save

Saves images faster than ComfyUI's default Save Image node. That's the pitch. It uses OpenCV directly, skips the workflow embedding, and doesn't apologize for either of those choices.

Important: workflow data is not saved in the images. If you need that, use the standard node. If you need speed, use this one.

---

Overview

The built-in Save Image node embeds the workflow JSON into every PNG it produces. That's useful for reproducibility and largely irrelevant if you're saving a 10,000-frame image sequence where disk throughput needs to keep up with generation. This node skips that step.

---

Features

• PNG, JPEG, and WebP output
• Multi-backend support — auto-selects fastest library per format
• Parallel saving — multi-threaded I/O for maximum throughput
• Batch filename generation — eliminates O(n²) collision checks
• Optimized tensor operations — minimal CPU-GPU transfer overhead
• Compression control — turn it off for maximum speed
• Memory efficient — no image duplication in output
• Suitable for high-volume batch and sequence output

---

Performance Optimizations

Multi-Backend Support

The node automatically selects the fastest available library for each format:

Format	Fastest	Quality	Notes
PNG	OpenCV	Good	Fastest general use
JPEG	TurboJPEG 🚀	Excellent	2-3x faster than OpenCV
WebP	PIL / webp-native	Best	Best quality control

Optional Python packages (PyTurboJPEG, webp) are auto-installed on first load — no manual steps needed there. Note that PyTurboJPEG is just a wrapper: it also needs the native libturbojpeg library installed on your system (see TurboJPEG setup below). If that native library isn't found, the node quietly falls back to OpenCV/PIL for JPEG — it won't crash or block other formats.

Parallel Saving

Enable parallel_save to save multiple images concurrently. Uses a thread pool optimized for I/O-bound operations. Tune max_workers (1-16) based on your storage:

• SSD/NVMe: 4-8 workers
• HDD: 2-4 workers
• Network storage: 1-2 workers

Batch Filename Collision Detection

Instead of checking each file individually (O(n²)), the node batches directory listings and uses set-based lookups.

Tensor Optimization

• Minimal CPU-GPU transfers
• Contiguous memory layouts for faster conversion
• Batched processing where possible

---

Node Inputs

Input	Description
images	Image batch to save
output_path	Directory path (supports [time(%Y-%m-%d)] formatting)
filename_prefix	Base name for files
filename_delimiter	Separator between prefix and number
filename_number_padding	Zero-padding for numbers (1-10)
filename_number_start	Start numbering at 0 or 1
extension	png / jpeg / webp
dpi	DPI metadata (1-600)
quality	JPEG/WebP quality 1-100
optimize_image	Compression optimization (png/jpeg/webp)
lossless_webp	Lossless WebP mode
overwrite_mode	Overwrite existing files
show_previews	Print save location to console
parallel_save	Enable multi-threaded saving
max_workers	Thread pool size (1-16)
backend	auto / opencv / pil / turbojpeg / webp-native

Backend Options

• auto: Automatically selects fastest available (recommended)
• opencv: Always available, good general performance
• pil: Best quality, more format options
• turbojpeg: Fastest JPEG (requires the native library — see TurboJPEG setup)
• webp-native: Native WebP library (auto-installed if available)

---

Installation

1. Clone this repo into your ComfyUI custom_nodes directory:

   cd ComfyUI/custom_nodes
   git clone https://github.com/APZmedia/APZmedia-comfyui-fast-image-save.git
   

2. Restart ComfyUI — optional performance libraries will auto-install on first load
3. Find the node under the APZmedia Fast image save category

That's it! PyTurboJPEG and webp libraries are automatically installed if not present. For the JPEG speed boost, also install the native TurboJPEG library below.

---

TurboJPEG Setup

PyTurboJPEG (the Python package) is only a wrapper around the native libturbojpeg library, which pip cannot install for you. Without it, PyTurboJPEG raises "Unable to locate turbojpeg library automatically" — this node catches that and silently falls back to OpenCV, so it's safe to ignore if you don't need the extra JPEG speed. To enable it:

• Windows: Install the libjpeg-turbo SDK (the -vc64.exe installer). PyTurboJPEG looks for it at the default install path C:\libjpeg-turbo64\bin\turbojpeg.dll.
• macOS: brew install jpeg-turbo
• Linux (Debian/Ubuntu): sudo apt install libturbojpeg0
• Linux (Fedora/RHEL): sudo dnf install libjpeg-turbo

After installing, restart ComfyUI. PyTurboJPEG looks for the library via the system's standard shared-library search (and LD_LIBRARY_PATH on Linux), plus the default install paths above — installing to the default location is the most reliable option.

---

License

MIT

Author

Pablo Apiolazza — APZmedia

☕ Support

If this saved you time on a long batch run, consider buying me a coffee.