Skip to content

Upscaling

animeojisan edited this page Sep 3, 2026 · 1 revision

Upscaling

English | 日本語

Neo can use GLSL shaders and ONNX models for image enhancement and upscaling.

GLSL vs ONNX

GLSL

  • Usually lightweight
  • Good for resize, sharpening, denoise, deband, CRT, and traditional image-processing shaders
  • Useful when low latency is more important than maximum AI reconstruction quality

ONNX

  • Used for neural-network image processing and AI upscaling
  • Usually much heavier than simple GLSL filters
  • Performance depends strongly on model architecture, input resolution, backend, and GPU

Scale factor

A model's internal scale factor is not the same thing as the final display size. Neo can process at an internal resolution and then present at another output size.

Avoid processing far more pixels than necessary. A high internal resolution followed by downscaling can sometimes improve quality, but it also increases GPU cost.

Choosing a model

There is no single best model for every source. Anime, live action, compressed streaming video, and clean high-resolution sources can favor different filters.

Start with a lighter model and increase processing only when the visual benefit is clear.

Clone this wiki locally