Flash-Test Generated Cuts Before Broadcast

A generated montage can feel fine in the room and still fail a Harding/PSE test. Rapid cuts, lightning VFX, and strobing practicals trip the same three-flashes-per-second rule live-action always had. The test is not a taste check. Recut or dim the offending frames. Do not regenerate the shot unless the picture itself is the problem.

Editors miss this because the failure does not look like a mistake. It looks like energy. Burst-cut sequences are a legitimate device. They are also one of the fastest ways to put more than three luminance flashes into a single second on more than a quarter of the screen.

Fast cuts and lightning that pass a creative review

Generated footage produces flash failures in a short list of ways, and most of them were requested:

Lightning, sparks, muzzle flashes, camera flashes, disco and concert strobes.
Glitch transitions, hard white frames, and "impact" flashes used as cuts.
Jump-cut montages and high cut-rate social edits that alternate bright and dark frames.
Fire, explosions, and electrical arcs that pop on and off inside one generation.
Saturated red emergency lighting, sirens, and "alert" graphics.

A creative review asks whether the scene reads. A PSE review asks whether a pair of opposing luminance changes, large enough and fast enough, occupies enough of the screen. Those questions do not share a monitor preset. You can love the cut and still fail.

This is a brand-safety problem as well as a broadcast one. A spot that ships on linear, on FAST, or on an owned web player can injure a photosensitive viewer. The brand-safety checklist covers disclosure and policy. This pass covers the picture.

The three-flashes-per-second rule, named

Three documents, not a forum rule.

ITU-R BT.1702-3 (November 2023), Guidance for the reduction of photosensitive epileptic seizures caused by television, is the international recommendation. Annex 1, Guideline 1, is the part you test against. A potentially harmful flash is a pair of opposing luminance changes. Isolated single, double, or triple flashes are acceptable. A sequence of flashes is not permitted when both of the following occur:

the combined area of flashes occurring concurrently occupies more than 25 percent of the displayed screen area; and
there are more than three flashes (six changes in luminance) within any one-second period.

Successive flashes whose leading edges are separated by 360 ms or more are acceptable in a 50 Hz environment, or 334 ms or more in a 60 Hz environment, irrespective of brightness or area. Versely timelines default to 25fps, which is a 50 Hz cadence: 360 ms is nine frames. Fast-cut sequences that land a bright frame every four or five frames will fail even when each individual cut "looks fine."

A luminance difference of 20 cd/m² or more is the SDR threshold when the darker image is below 160 cd/m². A transition to or from saturated red is potentially harmful irrespective of luminance. Rapidly changing image sequences, including fast cuts, are provocative if they result in areas of the screen that flash; the same constraints apply. A sequence of flashing images lasting more than five seconds might constitute a risk even when it complies with the flash-rate tests, which is the document's own caution on duration.

Ofcom requires UK television broadcasters to take precautions against flashing lights and patterns. The Harding Flash and Pattern Analyser (Harding FPA) is the implementation most UK delivery specs still print. UK DPP language now says "PSE test"; operators still say "run the Harding."

WCAG 2.2 Success Criterion 2.3.1 (Three Flashes or Below Threshold, Level A) applies when the piece plays on a web page: the page must not contain anything that flashes more than three times in any one second period, or the flash must sit below the general flash and red flash thresholds. An owned player on a brand site is a web page. SC 2.3.2 (Level AAA) removes the threshold exception.

Do not swap these. ITU-R BT.1702 is a broadcast recommendation with area, luminance, red, and spacing rules. WCAG 2.3.1 is a web success criterion. Harding FPA is a tool that implements Ofcom's guidance. They agree on "no more than three flashes per second" as the headline, and they disagree in the footnotes. Test against the spec the delivery names.

A flashmeter pass on the conformed cut

Run the analyser on the timeline you will deliver, at the frame rate you will deliver, after the grade. A 24fps generation viewed at 25fps has a different flash spacing than the file you thought you tested.

Conform first. One rate for the show. If the sequence is 25fps, test 25fps. A preview at another rate will move the leading edges.
Export a QC proxy or the mezzanine. The analyser needs a file, not a timeline playback.
Run a PSE analyser that implements ITU-R BT.1702 / Ofcom. Harding FPA is the name most broadcast specs still use. Read luminance flash, red flash, and pattern warnings, with timecode.
Treat "pass with caution" as a note, not a skip. Area under 25 percent can still warn. Duration over five seconds is the other caution BT.1702 flags.
Fix in the timeline, then re-test the failed section. Do not re-test the whole show until the reported window is clean, unless the spec requires a full pass.

The Versely editor is not a flashmeter. Assemble there, then analyse the export. The editor's preview: true pass renders at 480p and carries a short per-user cooldown. 480p is the wrong fidelity for a PSE test: area calculations and luminance differences both change when the image is that small. Use the preview to confirm that a recut still makes sense. Run the analyser on the real file.

Check a video before you publish it is the last assembly pass. Put the flashmeter ahead of that, in the same folder as the legalizer and the loudness report. A show that is legal, loudness-correct, and PSE-dirty is not deliverable.

Recut or dim without regenerating the shot

The generation is almost never the cheapest fix. The flash is a timing and luminance problem in frames you already have.

Lengthen the gap. If four impact frames sit inside one second, drop one, or hold an extra four frames between them so leading edges move past 360 ms at 25fps (nine frames). The cut still reads. The analyser sees three flashes, or two.

Dim the bright frames. A gain-down, a gamma pull, or a few frames of opacity on the white hit will drop the luminance difference below the 20 cd/m² threshold without changing the edit. This is the right tool for a lightning plate that is otherwise the correct picture.

Hold a frame. Replace a strobing run with a single bright frame and a decay. Isolated single, double, or triple flashes are acceptable under BT.1702. A hold is a recut, not a new generation.

Shrink the flashing area. Crop, a darker grade in the corners, or a graphic that covers part of the frame can pull the concurrent flashing area under 25 percent of the screen. Use this when the flash is motivated (a practical in the set) and you cannot dim it without losing the scene.

Pull the red. Saturated red strobes fail even when luminance is legal. Desaturate the red channel on those frames, or recut so a red hit is not flashing against a non-red frame more than three times a second.

Trim the burst. Trim the clip or cut a section out so the strobing tail never makes the timeline. A two-second generated lightning loop does not owe you all two seconds.

Regenerate when the content is the flash: the model drew a strobe you cannot dim without losing the subject, or every frame of the shot is a luminance hit. That is a new prompt ("continuous overcast lightning on the horizon, no on-camera strobes") or a different plate. It is not the first move.

Do not "fix" a fail by adding a verbal warning and shipping it, unless the relevant regulator's code gives you that out and the producer has taken it. Ofcom allows a warning in limited editorial cases. ITU-R BT.1702 is guidance for reducing risk in the picture. WCAG 2.3.1 does not have a warning exception. Know which document you are delivering against.

FAQ

Does a social cut-rate mean I can skip the test?

No. A Reels or Shorts encode does not exempt the file from WCAG if you also put it on a brand site, and it does not exempt a later linear or FAST version. Cut-rate is a creative choice. Flash rate is a measured property of luminance and area. They sometimes coincide. Test the version that will actually play.

Can I trust a pass on a 480p preview?

No. Area is a percentage of the displayed frame. Luminance is a display-referred number. A 480p preview with a short per-user cooldown is useful for checking whether a recut still makes sense as an edit. It is the wrong input to a PSE analyser. Test the delivery file, at the delivery rate.

What about patterns rather than flashes?

BT.1702 Guideline 2 covers regular high-contrast stripes: more than five light-and-dark pairs, over 40 percent of the screen if stationary, or over 25 percent if they oscillate, flash, or reverse. Generated graphics, test-card gags, and fashion stripes fail this more often than live-action. Harding FPA reports pattern separately from luminance flash. Fix by softening contrast, reducing spatial frequency, or cutting away from the pattern.

If the analyser fails one second of a thirty-second spot, do I re-export everything?

Fix the second, re-export, re-test that window, then run a full pass if the spec requires one. A one-second fail is usually a trim, a dim, or a held frame. Regenerating the surrounding shots to "be safe" is how a PSE fail turns into a new colour-match problem.