Filmora video upscaling: what the export setting does, and what it cannot

Filmora's export dialog lets you choose an output frame size independently of the clip you are editing, and it is tempting to read that as an upscale: a 720p phone clip goes in, a 1080p file comes out, the file is bigger and the delivery spec is satisfied. What happened inside is resampling. The exporter computed new pixels by interpolating the ones it already had and then encoded that larger grid. The result contains more pixels and exactly the same information, and the softness that made you pick a bigger number is still there, now stretched across more of them.

A bundled enhancement pass is a different operation, which is why the menu label matters less than the arithmetic underneath it. Interpolation averages neighbours; a model predicts what plausibly belongs there from patterns it has learned. One is a rule applied to the frame, the other is a guess constrained by the frame. Both can produce a file that looks sharper than the source, and only one of them can produce detail that was not in the source, in the narrow sense that the model invented a plausible edge rather than smeared an existing one.

You can tell the two apart on your own footage in a couple of minutes. Take the worst shot in the timeline, export a five-second slice at the source resolution, then export the same slice at the larger size. Open one frame of each at one hundred percent, side by side, and look at faces, at lettering, at any repeating texture such as fabric or brickwork. A bigger export looks like the same image seen through a soft lens. A pass that genuinely rebuilds looks different at the edge, not just smoother at the contour.

If you have one clip, one delivery and no reason to keep a second copy of anything, the bundled route is genuinely the cheaper one — not because the model is better, but because the total cost includes your time. The clip stays in the project, the result lands on the same timeline at the same timecode, the audio is already aligned, and nothing has to be exported, downloaded, re-imported or re-linked. For a single social cut or a thumbnail-bound clip that will be watched on a phone, that saving is real and the quality difference often is not visible at the size it will be seen.

There is a second, quieter advantage: you have already paid for the editor. An in-app pass has no marginal price attached to it, which makes it the correct first attempt when you are not yet sure the footage is worth spending anything on. Try it, watch the result at full size, and treat that as the free sample. If it holds, you are done. If it turns skin waxy or reshapes text, you have learned what this particular shot needs without having spent a credit on the diagnosis.

The boundary is repetition and control. A bundled pass is designed to do something reasonable to every clip any subscriber throws at it, which is the right design for an editor and the wrong one when you need to argue with a specific shot. The moment you have to reproduce the same look across a series, hand a client a before-and-after, or explain why one clip got a different treatment than its neighbour, you want settings you can name, change and write down — and that is what a dedicated pass provides.

Twenty clips is the point where an in-editor pass starts to cost more in attention than in money. Twenty exports, twenty operations, twenty files to keep track of, and no single number that tells you what the batch cost until it is over. A hosted pass processes one source at a time as well, but it lets you plan: each submission has a duration, each duration converts to a credit figure before you submit, and the batch as a whole has a price you can quote to whoever is paying for it.

The second case is the client who wants to see the difference. A dedicated pass gives you a clean pair — the untouched export and the enhanced file — at the same frame size and cadence, which is what a before-and-after comparison needs to be honest. The third is a series: a course, a channel, a product line where the same kind of footage returns every week. Once you know which model and tier handled the first episode well, the settings are a recipe rather than a guess, and the per-minute figure turns into a line in a budget instead of a surprise.

None of this makes the dedicated pass the default. For two clips on a Tuesday afternoon, exporting, uploading, waiting, downloading and re-importing is more friction than pressing the button already in the window, and pretending otherwise would be bad advice. The honest trade is convenience against control: the editor wins on the round trip, the hosted pass wins on a forecastable price, a model you can change per shot, and a result you can show someone at full size.

Export the slice at the project's native frame size, from the original media, once. Exporting at the size you are aiming for defeats the purpose, because the extra pixels would be the editor's interpolation and the model would then treat them as real detail. Use a codec that spends its bitrate on texture rather than convenience — a generous H.264 export for delivery-bound work, an intermediate codec if the file will be processed again — and export straight from the source clip rather than from a file that has already been exported once.

Upload the file, or paste a public HTTPS URL if the master is larger than the upload limit. Choose the model and the output tier, read the preflight quote, and submit. The request carries the video, the model, the target resolution and the billed duration; it does not carry an audio track, so the sound stays in the editor and gets re-attached on the way back.

When the enhanced file comes down, import it into the media library instead of dropping it straight onto the timeline. Put it on the track above the original at the same timecode, disable the original, and compare the two at full size in motion before you commit. Then replace the clip on the sequence rather than re-cutting around it, so transitions and effects survive. Keep the disabled original on its spare track until the final render is signed off — that is your rollback if the client prefers the softer version.

Pick the slice that contains the problem, not the shot you like most. If the complaint is a soft face in a dim room, the test slice is five to ten seconds of that face in that room, because that is the material the model will actually have to work with. A well-lit hero shot will make almost any pass look good and will tell you nothing about the shot you are worried about.

Judge the result at full size, at one hundred percent, in motion. A small preview window hides exactly the failure modes that matter: waxy skin, lettering that changed shape, edges that shimmer as the camera moves, fine texture that has become a repeating pattern. Pause on two or three frames and flip between the source and the result rather than watching one after the other, because memory is generous and a side-by-side is not. If the improvement only reads when you are told it is there, it is not an improvement worth paying for on the whole project.

Then write down what you used. The model, the resolution tier, and the exact duration you submitted are the three things you need to reproduce the result on the other nineteen clips, and they are the three things you will have forgotten by the time the client asks. The test scene is also the cheapest possible quote: a ten-second submission tells you the real per-clip duration, so the batch estimate is arithmetic rather than optimism.

At the Standard model, 1080p bills 540 credits per minute of submitted source footage, which is 32,400 credits an hour. That is the reference figure on this site and the one to budget with, because it is the tier most editor work needs: a 1080p timeline that has to become a clean 1080p master, or a 720p clip that has to survive a 1080p delivery. Billing follows the duration you submit, rounded up from a five-second floor, so trimming a clip to the part that needs the pass is the single largest saving available.

If the deliverable genuinely needs a 4K frame, the same model bills more per second at 4K than at 1080p, and the difference buys compatibility rather than recovered detail. Convert to money at the Creator yearly rate of $0.0093 per credit: a minute of 1080p Standard work is about $5.01, an hour about $300. Creator's 106,800 credits cover about 197 minutes of 1080p Standard footage, or about 53 minutes at Standard 4K.

Run the arithmetic against a real batch. Twenty clips at twelve seconds each is four minutes of submitted footage, not the twenty minutes the project file suggests, and the per-minute figure multiplied by four is your quote. That is the whole discipline: count submitted seconds, not timeline length, and let the preflight quote confirm the count before anything is charged. If the number looks wrong, you trimmed the wrong clip or picked the wrong tier — both are cheaper to fix before submitting than after.