pipersr 1.0.0

First super-resolution model designed for Apple Neural Engine. 2x upscale, real-time, on-device.

PiperSR

First super-resolution model designed and optimized for Apple Neural Engine.

2x upscale. Real-time. Runs entirely on the ANE in every Mac, iPhone, and iPad — no GPU, no cloud.

Metric	PiperSR
PSNR (Set5)	37.54 dB
FPS (M2 Max)	44.4
Compute	Apple Neural Engine
Precision	fp16
Scale	2x

Built by Ben Racicot. Available on ModelPiper.com.

---

Quick Start

pip install pipersr

from pipersr import upscale

output = upscale("input.png")
output.save("output_2x.png")

Or use the script directly:

python inference.py --input photo.png --output photo_2x.png

Download

The CoreML model (PiperSR_2x.mlpackage, 916 KB) is included in this repo. Clone and run — no separate download needed.

Also available on ModelPiper.com.

Benchmarks

Tested on M2 Max, macOS 15+. All models running on the same hardware.

Model	Set5 PSNR (dB)	FPS (M2 Max)	Compute	Format
PiperSR	37.54	44.4	ANE	CoreML
ESPCN (CoreML)	33.0	—	CPU/GPU	CoreML
IMDN (CoreML)	34.4	—	GPU	CoreML
Real-ESRGAN (CoreML)	35.8	—	GPU	CoreML

Run python benchmark.py on your own hardware to fill in the FPS column. We publish what we measure. Reproduce it yourself.

Samples

Input (1x)	PiperSR Output (2x)

Why ANE?

Every Apple Silicon device has a Neural Engine — a dedicated inference accelerator that most ML models ignore. The entire ML ecosystem targets CUDA GPUs. PiperSR was designed from scratch for ANE, not converted from a GPU model.

The result: real-time super-resolution with zero GPU usage. Your GPU stays free for rendering, compositing, or whatever else you're doing.

For real-time video upscaling powered by PiperSR, see ToolPiper.

Supported Hardware

Any Apple Silicon device:

• Mac: M1, M1 Pro/Max/Ultra, M2, M2 Pro/Max/Ultra, M3, M3 Pro/Max/Ultra, M4, M4 Pro/Max/Ultra, M5
• iPhone: A15+
• iPad: M1+

Performance scales with ANE generation. M2 Max benchmarks shown above.

Model Details

• Task: Single-image 2x super-resolution
• Input: RGB image, any resolution (CoreML flexible input)
• Output: 2x upscaled RGB image, fp16
• Format: CoreML .mlpackage
• Compute target: Apple Neural Engine via .cpuAndNeuralEngine

For the full model card, see MODEL_CARD.md.

For Video & Real-Time

PiperSR is an image model. For real-time video super-resolution with frame scheduling, buffering, and streaming — see ToolPiper, which integrates PiperSR into a full on-device video pipeline.

License

• Code (inference.py, benchmark.py, pipersr package): AGPL-3.0
• Model weights (.mlpackage): PiperSR Model License — free for personal and academic use, attribution required, commercial use requires a separate license

If you use PiperSR in your project, include:

Powered by PiperSR from ModelPiper — https://modelpiper.com

Links

• ModelPiper.com — Download PiperSR, browse benchmarks, explore on-device models
• ToolPiper — Local macOS AI toolkit with real-time video upscaling
• Ben Racicot — Author

Citation

@software{pipersr2026,
  author = {Racicot, Ben},
  title = {PiperSR: Real-Time Super-Resolution for Apple Neural Engine},
  year = {2026},
  url = {https://github.com/ModelPiper/PiperSR},
  note = {Available at https://modelpiper.com}
}