AI Ultra HD — recover lost detail, not just bigger pixels

What this pass is for

Detail hallucination that stays honest

The network fills in high-frequency information that a stretch cannot invent. It will not turn a 240p webcam into a native 4K master — the honest limit is one or two stops of resolution, not a miracle.

Lettering and hard edges first

Titles, lower-thirds, and UI screenshots benefit most. Organic texture (skin, foliage) improves too, but the model is tuned so that type stays type rather than turning into mush.

A reconstruction pass, not a denoise

If the source is crushed by compression, run AI Video Quality Repair first. Super-resolution on a blocky file will reconstruct the blocks. Sequence the two passes when both problems are present.

When to pick this over AI Video Enhancer

AI Video Enhancer is the right default when the source is already sharp and you mainly need more pixels for a larger canvas. Super-resolution is the right default when the source is soft — webcam, screen recording, older phone, heavily scaled web video — and you need the model to reconstruct what was never in the file.

If you are unsure, start with a 10-second preview of both. The difference is obvious on type and on hair; it is subtler on skies.

What it will not do

Sub-480p sources will not become true native 4K. Extreme digital zoom, already-upscaled YouTube rips, and stills that were never video will look better but will not look captured. Use restore-old-video for film grain and colour, quality-repair for blocking, and this pass for resolution — not as a substitute for any of them.

How to run it

1. 01

   Drop the clip

   Open Super-resolution from Enhance. Drop a 480p–1080p file. The preview uses a short window so you can compare against a bilinear stretch on the same frames.

2. 02

   Pick the target

   Choose 2× or 4× depending on the source. One stop (480p → 1080p, 1080p → 4K) is the honest sweet spot. Two stops is possible; three is usually a waste of time.

3. 03

   Export locally

   Write H.264 or H.265 to disk. The media is not uploaded. Chain a quality-repair pass afterwards if residual blocking remains.

System requirements

Operating system

Windows 10 (64-bit, 1809+) or Windows 11

macOS build coming — planned minimum macOS 13 or newer (Apple Silicon and Intel).

GPU

DirectX 12 compatible discrete GPU (NVIDIA / AMD / Intel Arc)

NVIDIA RTX recommended — TensorRT acceleration is available on RTX. macOS GPU requirements will ship with the Mac build.

Memory

16 GB RAM or above

Keeps previews and export staging responsive under load.

Storage

Fast SSD workspace

Better for caching, temp artifacts, and larger preview sessions.

Windows 10 (64-bit, 1809+) or Windows 11 today; macOS coming — planned minimum macOS 13+.

Frequently Asked Questions

What is AI video super-resolution?

It uses neural models to estimate missing detail when enlarging a frame, instead of only stretching pixels. Results still depend on source quality — very soft or tiny sources cannot become true native 4K.

How is this different from the AI Video Enhancer page?

Same family of local upscaling tools. This page targets the “super resolution” search intent; the AI Video Enhancer page covers the broader upscaling workflow and related settings.

Does KwaFlux upload my video?

No. Processing runs on your computer. The media is not uploaded to a render queue for these enhancement jobs.

Which platforms are supported?

Windows 10 (64-bit, 1809+) and Windows 11 are available now. macOS support is in development.

Can I try it before buying?

Yes. Download KwaFlux and render 1-, 3-, and 5-second effect previews across all AI modules without creating an account. You only sign in and pick a paid plan when you want to export or convert, and there is a $199.99 Perpetual License if you prefer to skip subscriptions.