AI Upscaling

AI upscaling by up to 4× per pass (up to 16× total with two-pass mode). Multiple speed/quality variants — the scene detection system can select the optimal model for the detected content type.

Requires: NVIDIA GPU (Ampere architecture or newer, ≥ 8 GB VRAM) for production throughput.

Example

Source resolution

4× AI upscaling

Parameters

Parameter	Section	Value	Default	Description
ResizeMode	Resize	0 – 12	5	Mode selection (see below)
ResizeFacDetMode	Resize	0 – 3	1	Size calculation mode. 0 = factor, 1 = small side, 2 = width, 3 = height
ResizeFactor	Resize	0.1 – 8.0	4.0	Resize factor (used when ResizeFacDetMode = 0)
SupResThresh	Resize	Float	2.0	SR only activates above this scale factor
SupRes2xThresh	Resize	Float	3.0	Scale-factor threshold up to which the 2× model is used; above it, the 4× model is applied

Modes

Value	Description
0 — Default	Auto — selects algorithm based on content
1 — Bicubic	Bicubic interpolation
2 — Biquadratic	Biquadratic interpolation
3 — Bilinear	Bilinear interpolation
4 — Nearest Neighbour	Nearest neighbour interpolation
5 — AI SR ×4, quality	AI-based 4× super-resolution
6 — Default — no SR	Default without super-resolution
7 — AI SR ×2 / ×4, quality	Auto-selects 2× or 4× based on ResizeFactor
8 — AI SR ×2 / ×4, fast	Fast variant of mode 7
9 — AI SR ×4, fast	Fast 4× super-resolution
10 — AI SR ×2 fast / ×4 fast	Fast 2× or 4× (alternate path)
11 — AI SR ×4 with AR	AI 4× super-resolution with integrated artifact removal
12 — AI SR ×2 / ×4 Legacy	Legacy 2× and 4× super-resolution model

For the full Resize section reference (including ResizeFacDetMode, ResizePixSize, ResizeOn), see the Parameter Reference → Resizing.

Recommended: Mode 7 for quality-focused production. Mode 11 if input images have visible JPEG artifacts.