An AI-generated 720p clip can look softer than the finished project around it. Magnific's Video Upscaler offers a 1440p output setting and a Creative mode that can synthesize new detail. The useful result is the downloaded file you have checked against your source, including faces, text, motion and sound.

This guide is for a creator with one 1280×720 clip they own or are licensed to upload, a paid Magnific account and authority to approve credit use. You also need a local video player, PowerShell and FFprobe, the media inspection tool distributed with FFmpeg. BIG CHANGE reviewed the current documentation on 24 October 2026. We did not sign into an account, upload or process a clip, or inspect an output. Treat every result below as a check to perform on your own files.

The big change

  • What changed: Magnific documents Creative and Precision video upscaling, with 1440p labeled 2K in its output settings. Creative can reconstruct and synthesize detail in low-resolution or AI-generated footage.
  • Why it matters: A larger file can still alter a face, letter or moving object. Comparing the downloaded output with the original gives the creator a basis for accepting it.
  • What to watch: The app's upload validator, credit quote and actual output govern an attempt. The public page mentions clips up to eight seconds, but we did not see a signed-in account's limits or a billable run.

1. Prepare one source and record its properties

Finish editing and color work before upscaling. Keep an untouched copy of the 720p source. Check your permission to upload the clip and use the result, including any depicted person's likeness, voice, marks and underlying media. Uploading sends the source to Magnific, so obtain any necessary consent first.

On Windows, install FFmpeg so ffprobe is available in PowerShell. Run ffprobe -version and resolve an unrecognized-command error before continuing. The block below asks for the absolute path to your source. Leave the output prompt empty for now; run it again after downloading the result and provide both paths. PowerShell's Get-FileHash calculates a SHA-256 identifier for each file. FFprobe reports container and stream metadata; neither command changes the media.

PowerShell
$sourcePath = Read-Host 'Absolute path to the untouched 720p source file'
$outputPath = Read-Host 'Absolute path to the downloaded Magnific output file (press Enter until it exists)'
Get-FileHash -Algorithm SHA256 -LiteralPath $sourcePath | Select-Object Path, Hash
& ffprobe -version | Select-Object -First 1
& ffprobe -v error -show_entries 'stream=index,codec_type,codec_name,width,height,avg_frame_rate:format=format_name,duration' -of json $sourcePath
if ($outputPath) {
  Get-FileHash -Algorithm SHA256 -LiteralPath $outputPath | Select-Object Path, Hash
  & ffprobe -v error -show_entries 'stream=index,codec_type,codec_name,width,height,avg_frame_rate:format=format_name,duration' -of json $outputPath
}

In the JSON, find the video stream's width, height and avg_frame_rate, the format's duration in seconds, and any stream whose codec_type is audio. Record the source hash and values privately. If a probe fails or omits a value you need, hold the file inspection until you resolve the path, format or tool error. Metadata alone cannot show whether every frame plays or the audio stays synchronized.

2. Choose the upscaling mode and price one attempt

Open Video Upscaler in your signed-in Magnific dashboard and upload that exact clip or select it from your creations. Check its preview against your local source and record the accepted file or creation ID. The public product page says uploaded clips can be up to eight seconds. We could not verify that as a universal limit for current accounts. Follow the validator shown for your own account; if it rejects the file, record its exact message and hold the attempt.

For an AI-generated clip with visible defects, choose Magnific Creative. Its mode guide describes Natural for organic subjects and Vivid for stylized, digital or 3D work. If the clip already looks clean and preserving its visual identity matters more, compare a section with Precision using Preview. Inspect faces and skin, hands, lettering, object continuity and movement before choosing. Preview may use credits; read its quote first. Creative's added texture is synthetic interpretation, never evidence of detail that was in the original scene. Precision is intended to stay closer to the input, but it can carry existing artifacts into a sharper output.

Set resolution to 1440p (2K) and frame rate to Original, unless you have a separate reason and approval to change the pacing. Read the output summary above Upscale. Then check the displayed charge and current balance under Magnific's credit rules. Credit use depends on model and output; a fixed price quoted here would be unreliable. Record the settings and exact quote, and obtain the clip owner's approval for one paid run before clicking Upscale.

3. Download the result and check the whole file

Watch the same creation until Magnific shows completion or a specific failure. If the status or charge is uncertain, check that creation and credit history before any retry. Download the completed output and preserve it beside the untouched source. Run the PowerShell block again with both absolute paths. Record the output hash, dimensions, frame rate, duration, format and audio-stream presence. The target is 1440 vertical pixels with the intended aspect ratio; for an unchanged 16:9 source, that means 2560×1440. Check actual frame rate and duration against the original. Magnific says upscaling does not change video length, but the downloaded file is the evidence for your attempt.

Play both files from beginning to end with sound enabled. Compare matched start, middle and end moments, then revisit any difficult motion or visible artifact. Look for changed face or skin texture, identity, hands, lettering, moving objects, edges, flicker and invented elements. Compare audio presence and audible content throughout. If the source has audio, check lip or action sync at those matched points. If it is silent, write “silent source” and determine whether the output stayed silent or acquired unexpected sound. The docs do not establish an audio-preservation guarantee.

Use this blank record for one attempt. Keep file paths, hashes, account information and any private media evidence out of public notes.

Snippet
Owner and UTC time:
Source path / SHA-256 / rights and upload permission:
Source width x height / fps / duration / format / audio present:
Magnific upload-validator result or exact rejection:
Mode and Natural/Vivid choice; Preview comparison and quote:
Output summary / credit balance / run quote / one-run approval:
Creation ID / final status / actual credits charged:
Output path / SHA-256:
Output width x height / fps / duration / format / audio present:
Full source and output playback completed, with player and time:
Matched start: visual changes / audio presence and sync:
Matched middle: visual changes / audio presence and sync:
Matched end: visual changes / audio presence and sync:
Difficult moment: time and observed change:
Faces, identity, hands, text, motion, edges, flicker, invented detail:
Silent-source disposition, or full audio-content and sync comparison:
Output use rights and permission:
Decision on this exact output hash: ACCEPT / REVISE / HOLD
Reason and next check; separate credit approval if revising:

Accept only the specific file whose properties, playback, rights and charge you recorded. If a face, word, moving object or audio track is wrong, keep the rejected file and creation record. A revision may use a focused mode or setting change after a new credit approval; an uncertain first run needs a readback before another paid click. A 1440p label on the creation page cannot settle the quality decision.

If your starting task is generating an original clip under face and IP restrictions, see our Seedance guide. If you need to assemble owned takes into one film first, see the separate Claude Code editing guide.

Sources & further reading