Soku AI
All tools

AI Video Sharpener Recover the Detail Your Re-Encode Threw Away

Most footage does not arrive soft — it becomes soft.

RestorationTopaz Precision

Three deliberate steps

01

Upload the clip that looks soft

Up to 15 seconds of source video. Pick the specific cut that is letting the ad down rather than a whole master — this is a per-shot repair, and short ad cuts are exactly the shape it suits.

02

It runs a precision restoration pass

The Proteus model reconstructs edge and texture detail and cleans up compression artefacts, doubling the resolution as it goes. A 1080p source comes back at roughly 4K, so a clip that was only ever soft because it was small stops being small.

03

Check it against the original before you ship

Play the two side by side at the size the ad will actually run. Restoration is convincing on lightly damaged footage and honest about its limits on badly damaged footage, and a fast A/B is how you tell which one you had.

Use cases

Who reaches for this, and when

Rescue a winning ad you only have a compressed copy of

The creative that carried last quarter often survives only as a platform download, two re-encodes away from the master nobody can find. Sharpening that copy is usually faster than reshooting it, and it keeps the exact cut the algorithm already learned to spend on.

Bring supplier and UGC footage up to platform standard

Clips that arrive from a manufacturer, a reseller or a creator are frequently phone-grade and heavily compressed. A restoration pass makes them sit alongside your own material in the same ad set instead of visibly dragging it down.

Make old library footage usable at today’s placements

A 720p archive clip that looked fine on a 2019 feed looks soft full-bleed on a modern phone. Doubling resolution and restoring detail is what lets an existing library be re-cut rather than retired.

Questions before you run it

Does this re-generate my video with AI?
No, and that is the important distinction. This runs Topaz Video Precision, a restoration model that works from the detail already present in your frames. Generative video models asked to "upscale" something effectively re-shoot it — the subject drifts, faces change, and product details you were legally required to keep accurate quietly become approximations. Your shot, your product, your talent, all stay put here.
What is the difference between this and the AI video upscaler?
They run the same restoration model, and the difference is which problem you came with. The upscaler page is framed around a clip that is too small for the placement. This page is framed around a clip that is the right size but looks soft, mushy or over-compressed. Because the model both resolves detail and doubles resolution, either page will do both jobs — pick whichever describes what is actually wrong with your footage.
Will it fix motion blur or a badly out-of-focus shot?
Not reliably, no. Compression artefacts, softness from re-encoding, and low-resolution mush are what this class of model is genuinely good at, because the underlying detail is still partly there to recover. A shot where the camera missed focus, or where fast motion smeared the subject across the frame, has lost information that no restoration pass can invent back. If the original was out of focus, re-shoot it.
How long a clip can I upload?
Up to 15 seconds of source video, one file per run. That is deliberately matched to the length of an actual ad cut rather than a full edit — restoring a two-minute master is a different job with a very different cost, and paying for it a shot at a time is how you keep the bill proportional to the ad you are fixing.
Is it free?
Signing up on this page gives you free credits to run it here in the browser. After that each run is credit-metered and the cost is shown before you submit, so you approve the price rather than discover it. Video restoration is one of the more compute-heavy things on the site, which is why the price is stated up front on every run.