No sign-up · no watermark · unlimited · runs on your device
or drop it here
Checking what this device can do…
2×, 3× or 4× on your own GPU, frame by frame.
Tightens soft edges, frame by frame.
Cleans up compressed, soft footage.
More resolution and cleaner edges.
Turns blocks and stair-steps back into gradients.
Resamples the pixels, then writes a real DPI.
Takes out the grain, leaves the edges.
Tightens soft edges, keeps your dimensions.
2× or 4×, photos and illustrations.
BeforeAfter
BeforeAfter
BeforeAfterDrag one video or a dozen onto the box above, or click to browse. The page checks what your device can handle before you choose a file, so you learn about limits up front rather than half-way through a job.
Drag the before/after slider over hair, fabric, foliage or type — not over a flat wall, where by definition nothing changes. If the after side looks brittle rather than defined, drop the Deblur dial a notch and let it re-run.
No watermark, no sign-up, no cap on how many times you do it. The audio track is copied across in its original form rather than re-encoded, so it loses nothing and stays in sync.
Both run on your own graphics card, free and unlimited. Start with the default and only change it if the result looks wrong.
| Method | Best for◆ | How much it changes | Speed |
|---|---|---|---|
| Deblur (Light) — default | Footage that genuinely looks soft. Start here. | The stronger of the two | Instant |
| Balanced | A light lift when you are not sure | Gentler — barely touches footage that was already right | Instant |
Other methods were cut on evidence rather than taste: they smooth the picture before sharpening it, and on this page’s footage the smoothing won — they handed back footage softer than they were given. Both of the ones here work only with the detail your frames already hold, which is the ceiling they share. When neither does enough, the answer is not a stronger setting — it is Topaz below.
This uploads the clip and runs on our servers for credits, on Topaz Labs' own network for damaged footage. It is a trained model rather than a shader, so it reads the whole frame instead of each edge in isolation — a much higher ceiling, and still a faithful one: it improves the pixels your frame already holds rather than inventing new ones.
| Model | Type | Best for | Per second at 1080p |
|---|---|---|---|
| Topaz Artemis | Precision | Denoises and sharpens soft, damaged sources | ≈ $0.06 |
Priced by the second at the size you choose, so 4K costs four times 1080p. The figure for your own file sits on the Run button before anything is sent, and credits are spent per run — no subscription. No numbers on this page are its. Every figure further down was measured on the free methods; the evaluation set has no cloud run in it yet, so what is claimed here is what kind of model it is, not how far ahead it finishes.
Drop the file into a browser tool that processes it locally. FreeUpscaler decodes the video frame by frame, sharpens each frame on your own graphics card, then re-encodes it — all inside the tab, with no account, no watermark and no length limit.
Sharpening does not add detail. It makes the detail already in the frame easier to see, by steepening the transition across every edge — which works well on footage that is merely soft, and does nothing where there was never much detail to find. Measured on seven clips, the free default brought back about a third of what the blur had taken. If that is not enough, the step up is Topaz in the cloud rather than a stronger setting: a trained model reads the whole frame instead of each edge in isolation.
Footage that is already sharp is the case where sharpening costs more than it returns — that measurement is further down, not hidden.
As it arrived · soft
Sharpened · the default method
The original · what the blur took
Seven clips were blurred by a known amount and sharpened. The default brought back about a third of what the blur had taken — roughly half on ordinary camera footage, and up to 78% on animation and motion graphics. One clip came out worse than it went in: a dark, fine-grained scene where the browser’s own encoder loses more than any sharpening pass puts back.
Every clip was also run in a second copy that was never blurred — the case every tool in this category claims to handle and none of them measures. Sharpening footage that did not need it pushed the edges past where they really were, which on video shows up as edges that shimmer between frames. So the rule is not “sharpen everything a bit”: sharpen footage that looks soft, and leave footage that already looks right alone.
Because the usual method raises the contrast across an edge without making the edge any narrower — and past a point, all it adds is a bright fringe.
A normal sharpen pushes each pixel away from its neighbours. Turn it up and edges do read as harder, but the transition stays exactly as wide as it was; what grows instead is a bright and dark fringe on either side, which is the halo. That is why every tool in this category ships its slider turned most of the way down.
The default here works the other way round: it moves each pixel towards whichever side of the edge it is already nearer, so the transition itself gets narrower rather than louder. And no sharpened pixel is ever allowed outside the brightest and darkest values its own neighbours already had — so a halo is impossible rather than unlikely, and that is asserted in the test suite rather than observed once.
Two of these return less than you would hope and two cannot improve at all — not by sharpening, which is what runs on your own machine.
Sharpening a 480p clip makes the existing pixels harder, not the picture better. Enlarge it instead — the upscaler reconstructs at a larger size and sharpens as part of the same pass.
It will change, and not for the better — over-sharpened edges shimmer between frames in a way they never do in a still. Use the gentler method, which barely touches footage that needed nothing.
The one clip in testing where every free method made things worse: the browser's own encoder loses more on this material than any sharpening pass puts back. Topaz in the cloud does not go through that encoder, so it is the one thing left to try.
Each point was smeared across pixels while the shutter was open, so sharpening only makes the smear crisper. Nothing on this page recovers it; an AI model on the blurry-video page will draw you a sharp frame instead, one it invented.
Drop the file onto this page. It is decoded frame by frame in your browser, each frame is sharpened on your own graphics card, and the result is re-encoded locally — no account, no watermark and no length limit. Nothing has to be downloaded first, and a one-minute clip finishes in well under a minute.
No, and anything that says otherwise is selling something. Sharpening increases contrast across edges that are already in the frame, which makes the detail that was captured easier to see. Across seven test clips it brought back about a third of what the blur had taken — a real improvement, and a long way from restoring the original.
No, and that is enforced rather than tuned. Every sharpened pixel is clamped to the minimum and maximum of its immediate neighbours, so it cannot end up brighter or darker than anything that was really there — which is exactly what a halo is. The test suite asserts it over twenty thousand random neighbourhoods rather than taking it on trust.
Not by the free methods. Decoding, sharpening and re-encoding all happen inside your browser tab — load the page, disconnect from the internet, and it still works. The Topaz row is the exception and says so before it runs: it uploads the clip for that job and needs an account, because the run costs real money.
It depends on how soft the footage was to begin with. Across seven test clips the default brought back about a third of what the blur had taken — roughly half on ordinary camera footage, up to 78% on animation and motion graphics, and least of all on flat studio shots. One clip went backwards: a dark, fine-grained scene where the browser's own re-encode costs more than any filter here adds.
Start with the default — it is the stronger of the two and it won on every test clip. Switch to the gentler one if your footage was already reasonably sharp and you only want a light lift: it barely touches footage that needed nothing, where the default will push the edges past where they really were. Both are instant and neither needs a download, so comparing them costs one re-run. If neither does enough, the step up is Topaz in the cloud rather than a stronger setting.
One, in the options popover, labelled Deblur. It controls how many times the edge-narrowing pass runs, and unlike most sharpness sliders every notch does something measurable. The cost moves with it: the higher you go, the more it overshoots on footage that did not need it. If the result still looks soft, go up; if it looks brittle, go down.
Because the same filter is the default on the blurry-video page, and one implementation gets one name across the site rather than a flattering alias per page. What it does is narrow soft edges, which is what sharpening a soft video means.
On the hard cases, yes. The free methods are shader maths that look at each edge in isolation, so they do well on footage that is merely soft and little at all where there is not much of an edge to find — flat studio shots, or dark fine-grained material the browser's own encoder chews up. Topaz is a trained model running on rented hardware: it reads the whole frame, and it does not go through your browser's encoder. It is still a faithful pass rather than a generative one, so it will not invent detail that was never filmed. It also uploads the clip, needs an account and costs credits, none of which the free methods do.
The container is rarely the problem — MP4, WebM, MOV and MKV all open. The codec inside is. Decoding runs on WebCodecs, which exposes whatever decoders the browser ships: H.264 and VP9 everywhere, AV1 on most current builds, H.265 and ProRes usually nowhere. Converting to H.264 first is the workaround.