4xNomos2 HQ MoSR

Fast 4× photo upscaler for clean, good-quality images.

A MoSR upscaler trained on the Nomos-v2 photo dataset for clean inputs. It is about three and a half times faster than the ESRGAN upscaler on CPU, at a quarter of the size.

Download ONNX View repository

Super-resolution Computer Vision

Scale: 4×
Parameters: 4.3M
CPU time per 256 px tile: 2.3s
ONNX file: 17MB

Example output

Drag the handle to compare. Produced by running this model.

After: Input: a 130 px landscape, enlarged with bicubic resizing. Output: the model's 4× result. — Input: a 130 px landscape, enlarged with bicubic resizing. Output: the model's 4× result.

Before: Input: a 130 px landscape, enlarged with bicubic resizing. Output: the model's 4× result. — Input: a 130 px landscape, enlarged with bicubic resizing. Output: the model's 4× result.

Technical details

Architecture: MoSR (gated CNN blocks)
Fine-tuned from: 4xmssim_mosr_pretrain
Training data: Nomos-v2, 6,000 photos
Training: 190,000 iterations
ONNX build: FP32, opset 17, any image size
Input: RGB, NCHW, values 0–1
Benchmark: ONNX Runtime CPU, Apple M2 Pro
Original release: 25 August 2024

Good for

Good-quality photos
Prints and posters
Batch upscaling

Run it in Flutter

This model runs as-is with our open-source flutter_super_resolution package (MIT), which tiles large images and uses Core ML on Apple devices when available. It works on iOS, Android (ARM devices), macOS and Windows.

pubspec.yaml · Dartgeneral4xlow.onnx

dependencies:
  flutter_super_resolution: ^1.1.1

// Load the model, then upscale a ui.Image
final upscaler = FlutterUpscaler(tileSize: 128, overlap: 8);
await upscaler.initializeModel('assets/models/general4xlow.onnx');

final result = await upscaler.upscaleImage(
  image,
  scale: 4,
  onProgress: (progress, message) => print('$message ${(progress * 100).round()}%'),
);

upscaler.dispose();

Package on GitHub All models that work with it