AI Video Upscaler+
Tips & Tricks
Get it from Microsoft
×

Tips & Tricks

A few things worth knowing before you spend render time — none of it is required, but each one changes the result in a specific way.

Face Restoration: quality vs. speed

GFPGAN runs a dedicated face-restoration pass over every single frame. It's the difference-maker for sharpness on faces — but that extra pass takes real time, so it isn't free.

Restoration off
Faster processing
Good for quick previews or footage without close-up faces.
Restoration on
Sharper, cleaner faces
Noticeably longer processing time — best for videos where faces matter.

A practical rule: turn it on for portraits, interviews, and home videos of people; leave it off for landscapes, screen recordings, or footage without close-up faces.

How far to push a resolution target

The AI model adds real detail up to roughly 4× your source's native resolution in one natural pass. Beyond that, the app automatically switches to a tiled processing mode to reach the target safely — still good quality, just a longer job.

1–2×
Very fast, very clean
4×
Native scale — the sweet spot
>4×
Tiled — slower, still solid

Practical example: a 640×480 source comfortably reaches 2560×1920 (roughly 1440p–2K) in the fast path. Pushing the same source to 4K or 8K still works — the app just tiles the frame automatically to keep memory use safe on your GPU.

Choosing a frame rate target

Frame-rate interpolation generates new frames between the ones you already have. How much of a jump to ask for depends on the source:

Modest boost

24–30 fps → 45 fps

A clearly smoother result with the shortest processing time — a good default for most footage.

Maximum smoothness

24–30 fps → 60 fps

The smoothest possible motion, especially noticeable on panning shots — takes longer since more in-between frames are generated.

Combining a big frame-rate jump with a big resolution jump (say, 640×480 @ 24 fps → 4K @ 60 fps) works in a single pass, but is the most demanding combination — expect the longest processing times from this kind of job.

GPU selection and video codec

  • If your PC has both an integrated GPU and a dedicated one, pick the dedicated GPU in Settings for significantly faster processing.
  • Leave Video codec on Auto unless you have a specific reason not to — it picks H.264 for compatibility on typical output sizes, and switches to HEVC automatically once resolution gets large enough that H.264 playback support becomes unreliable.
  • If a finished video won't play in a particular app or older TV, try forcing Always H.264 — it's the most broadly supported codec, just less efficient at very high resolutions.

For long videos, queue them and keep working

If you have several videos to upscale, there's no need to wait around for each one:

  • Add every video to the queue up front — they process one after another automatically.
  • Processing runs entirely in the background, so you can keep using your PC for other things while it works.
  • You'll get a Windows notification the moment each individual video finishes, so you never have to keep checking progress manually.
Rule of thumb: start your biggest, most demanding job (a big resolution and fps jump) first thing, queue everything else behind it, and go back to what you were doing.