Video Analyzer

One upload → the full report. Runs 100% in your browser — your video never leaves your device.

New here? See a sample report measured from a real clip before you run yours.

Drop a file here or click to choose — either way, the video never leaves your device.

Optional: also check thumbnail, title, tags + subtitles

Drop a thumbnail file here or click to choose

Optional subtitle file (.srt or .vtt) — timing, reading speed, line shape and a profanity scan get audited

Big file? Analyze one segment only

Files over 1 GB or over 60 minutes are only analyzed in segment mode — pick the stretch that matters (the hook lives at the start). In segment mode the loudness passes are skipped; run the LUFS tool on a short export for the audio number. On phones the file caps are lower (300 MB on Android, 200 MB on iPhone/iPad) and files past them need trimming first, or a computer.

Desktop sweet spot: up to ~200 MB / 15 min for a full pass. Larger files are sampled honestly (hook + audio in full). Phones run tighter gates (Android: ~150 MB full pass, 300 MB max; iPhone: ~100 MB full pass, 200 MB max) — mobile browsers kill oversized tabs. Keep this tab open during analysis; you can cancel any time.

The checks, in the order they run

  1. File facts — real type (by bytes, not name), size, duration, resolution, bitrate sanity.
  2. Thumbnail & metadata — if added: thumbnail format/spec, brightness/contrast, faces and embedded location data; title length plus a mechanical form score; the description's opening snippet, chapter rules and hashtag tiers; the ~500-character tag budget; and a subtitle-file audit (timing, reading speed, profanity flags).
  3. The hook (0–5s) — motion, black/frozen opening, face presence, whether anything happens before viewers leave.
  4. Audio — silence gaps with timestamps, opening silence, peak level, and a true loudnorm LUFS measurement when the file size allows.
  5. Visual timeline — black intervals, frozen stretches, cut rate and pacing, longest static stretch, and the loop point on short videos.
  6. Safe zones — for vertical video: whether the centre of frame survives Shorts/TikTok/Reels overlays.

How the Analyzer reads your file

Nothing is uploaded, and nothing is approximated in secret — but not everything is measured the same way, and the report tells you which is which. The run starts with the bytes themselves: the file's real type is read from its first bytes, not its name, and the browser's own video decoder reports duration, dimensions and average bitrate. A container probe then reads codec, frame rate, rotation and the audio/video track start times from the file's structure, and a byte scan looks for provenance declarations (C2PA, IPTC) and the colour space the file declares.

The hook and timeline stages work on sampled frames: the analyzer seeks through the video and draws small 160-pixel-wide frames to a canvas, measuring brightness, motion between frames, black and frozen stretches and the cut rate. These are real measurements — of samples, not of every frame. A freeze shorter than the sampling step can slip past, which is why findings carry a "≈" and name the stretch instead of pretending to frame accuracy. Audio, by contrast, is decoded in full: silence gaps, opening silence and levels all come from the decoded waveform.

Loudness is the one stage with two grades. At 500 MB and under, the broadcast-standard loudnorm measurement (ITU-R BS.1770) runs on the whole file and the number is exact — the same figure the dedicated LUFS tool reports. Above that cap, decoding and measuring an entire file inside a browser tab risks the tab's memory, so the analyzer falls back to a waveform-based estimate and labels it as one. When a figure has to be exact — a final master, a client delivery — measure the export in the LUFS tool.

Three more stages deserve the same honesty. When a file claims 1440p or 4K, the analyzer runs the Real Quality Checker's round-trip test: frames grabbed at native resolution (capped at 2560 pixels on the long edge), shrunk to 1080p and 720p and blown back up, with the surviving fine detail measured — an estimate with evidence, never a verdict. The face scan samples the whole video (a frame every few seconds, at most 40) with the same on-device detector used for the opening, so the timestamps it lists are a starting point for review, not a complete census. And the provenance scan reads the first 8 MB and last 4 MB of the file's bytes — where C2PA and IPTC declarations live; anything it extracts from a manifest is labelled extracted, not verified, because reading a claim is not the same as checking its signature.

What this tool cannot tell you

  • It cannot query Content ID or guarantee anything about claims, strikes or monetization review — those live inside YouTube.
  • Loudness: when the file is small enough (≤500 MB), this analyzer also runs the broadcast-standard loudnorm measurement in your browser. On larger files it falls back to a labelled estimate — the dedicated LUFS tool is the reference.
  • Face detection (the opening check and the whole-video sample) runs Google's MediaPipe model in your browser; it reports presence only, never identity, and a miss doesn't prove no face is there. If the model can't load, the report says the face checks did not run instead of guessing.
  • It measures the file, not the idea. A technically perfect video can still fail; a rough one can still fly. This report removes the fixable, mechanical failures.

Frequently asked questions

What does the Video Analyzer check first?

In order: (1) the file itself — type, size, duration, resolution, bitrate; (2) your thumbnail and title/tags if you add them; (3) the first 5 seconds — the hook — frame by frame; (4) the audio — silence gaps and levels; (5) the full visual timeline — black frames, frozen stretches, pacing; (6) safe zones for vertical video. The report leads with the fixes that matter most, not with a score.

Is my video uploaded for analysis?

No. Analysis runs entirely in your browser using your own device. You can verify it in your browser's network tab: the video file is never sent anywhere.

How big a video can I analyze?

Full analysis up to ~200 MB / 15 minutes on desktop; beyond that the timeline is sampled, up to 1 GB / 60 minutes. Past that gate, use segment mode (the 'Analyze one segment only' fields) — the browser checks just the stretch you pick, which is how very large masters get checked at all in a browser. Over 2 GB is a hard browser memory limit. Phones have tighter gates, because mobile browsers kill oversized tabs: on Android a full pass runs up to ~150 MB / 10 minutes, sampled up to 300 MB; on iPhone/iPad a full pass is ~100 MB / 5 minutes, sampled up to 200 MB — past those caps, trim the file or run it on a computer.

Can it tell me if my video will get a copyright claim?

No — nobody outside YouTube can query Content ID. What it can do is flag the risk factors you control (for example, long stretches of someone else's footage pattern is beyond any honest checker). See the Song Copyright guide for what can and can't be known before upload.

Why does the first-5-seconds check matter so much?

Because retention data across platforms shows the opening seconds decide whether viewers stay. A black, frozen or silent opening is the single most common fixable cause of early drop-off — so the analyzer checks it first and in the most detail.

What can't this analyzer judge?

Whether your content is good, original or policy-safe in YouTube's eyes; whether the music will be claimed; and how the algorithm will treat the video. It measures the file. Judgment about the content stays yours — see the honesty box on this page.

Related tools

Related guides

CreatorChecks is not affiliated with YouTube, Google, TikTok or Meta.