A Commercial Test Flickers. Change the Shutter—or Change the Light?
A Commercial Test Flickers. Change the Shutter—or Change the Light?
You have a test clip and a moving band of brightness in it. There are two families of fixes, and both of them are real fixes. One moves the camera's timing. One moves the light. Whichever you take, something the test was set up to show comes back different.
That isn't a reason to avoid the decision. It's the shape of it — and it's the reason the answer can't be a setting. The useful question isn't which route removes the banding. Both can. The useful question is which route leaves the clip still answering the question the test was asked.
What follows is bounded on purpose. It assumes a test you can re-run, a camera whose movie modes you can read about, and a light you can control through documented means. It does not include electrical work on the fixture, a promise that one shutter value cures every LED, or recovery of missing exposure after the fact.
Be sure the banding belongs to the light
Brightness variation in a saved file has several possible authors, and they don't share a fix.
The light is one. The camera's own exposure logic is another: if the meter is free to move, a hand entering frame can provoke a whole-image shift that looks like flicker and isn't. Lock exposure and shoot a take with no movement at all. If the band is still there, the light is implicated; if the file settles down, you were watching an automatic exposure decision.
The subject can do it too. A reflective surface rotating through a source, a screen in frame showing its own refresh, a practical whose own driver is doing something the key isn't — each of these can produce a variation that follows the object rather than the frame.
And some defects that arrive during movement belong to a different family entirely. Bent verticals through a whip pan, a leaning tower as the camera travels, is a readout-geometry problem. It shares a cause with banding in the loose sense that both come from sampling a scene that won't hold still, but it does not respond to a shutter change or a fixture swap the way luminance bands do. Chase it here and you'll spend the day tuning the wrong end of the camera.
Before any of this, keep an untouched baseline and write down the setup: camera mode, frame rate, shutter value, the light configuration, and the fixture's own settings — dimmer position, mode, anything its manual names. The fixture's settings are a variable, and if you don't record them you can't tell later whether your fix worked or your dimmer moved.
Then check the file properly. Sony's support article on reducing camera flickering — read in September 2026 — warns that what you see on the camera's monitor can differ from what was recorded, and that page covers Sony's own features and documented modes rather than your fixture. The warning travels well. A monitor can soften a moving band into something you dismiss, and it can also show a faint rolling artefact that the file doesn't contain. Scrub the recorded file at full size, frame by frame, and compare it against a take with the suspect light switched off.
Why there are only two levers
A light's output is not flat at every level. Mains-driven sources commonly vary with the supply, and dimmers and LED drivers add patterns of their own — some slow enough to read as a pulse across a whole frame, some fast enough that they only show up in the interaction with the camera.
The camera samples that output twice over. Each frame's exposure is a window laid over the light's waveform; the readout assembles the frame from lines that opened that window at different points in the cycle. Many movie cameras read the sensor line by line rather than all at once, which is why a variation occurring during readout lands at a different phase on different lines and shows up as a band that can drift, sit still, or crawl.
Where the window stops sets the depth of that band. Stop it on a whole number of the light's cycles and every line integrates the same total, whenever it happened to open. Leave a fraction of a cycle hanging off the end and lines that opened at a different phase integrate a different total; the band is the difference between them. A longer window does not average this away — the leftover fraction sets the depth whether the window is long or short — and the widest split comes when the window stops halfway through a cycle, with one set of lines catching the bright part and another the dark. So the values worth finding are the ones that close on whole cycles, and finding them is a matter of measurement rather than arithmetic, because the fraction is measured against the fixture's own cycle and not against the frame rate.
So the interaction has exactly two ends: the light's pattern and the camera's sampling. A shutter change moves the sampling — how long the window is and where it stops. An illumination change moves the pattern. That's the whole menu.
It's worth being explicit about a third thing people reach for, because it looks like the same lever and isn't. Changing frame rate also changes temporal sampling — cadence, strobing, how much slow-motion headroom you have, and what the delivery format will need. It may genuinely alter how visible the banding is. But it is a different test with different consequences, and folding it into a shutter trial produces a log you can't read afterwards.
One more reason not to treat mains multiples as a cure: the article's remedy list is deliberately not a formula. A setting tuned to one ripple frequency has no particular claim on a fixture whose driver is running at a few hundred hertz, on a display in the shot, or on a location with two different fixture types in the same frame. The vendor split is instructive here. Sony's page distinguishes a still-only anti-flicker timing function from variable-shutter features usable for movies on supported models — which means a habit you built shooting stills may not exist in movie mode on the body you're holding, or may exist under a different control with different behaviour. Read the actual camera's movie-mode documentation; don't port the fix.
Route one: change the timing, and pay in motion
Take the baseline light configuration and hold it. Then work through the shutter values your camera's movie-mode guidance actually supports, changing one condition per take. The value to look for is the one whose exposure window closes on a whole number of the fixture's cycles; it is not the shortest setting the list offers, and a shorter exposure can band worse than a longer one. Where the body documents a fine or variable shutter adjustment, that control places the window more exactly than the coarse list can. Log exposure compensation as a separate line item, because a shorter exposure costs light and something has to give. Neutral density is the cleanest compensation; ISO changes noise, and aperture changes depth of field, which is another way of changing the shot while telling yourself you only touched the shutter.
What you're buying is a reduction in the brightness problem. What you're paying is motion blur, because exposure time is motion blur. Shortening it hardens the edge of everything that moves. That matters most exactly where a treatment has chosen a 180-degree shutter for a reason — a soft hand entering, a pour that reads as liquid rather than as a series of shapes, a rotation the eye is supposed to follow without effort.
The size of the bill is easy to underestimate because it's invisible in a still. Going from 1/48 second to 1/100 second is a little over a stop, and it takes an edge that carried a soft smear and turns it crisp. Nothing else in the frame appears to move. The one thing the test existed to demonstrate now has a different character.
Two further cautions on this route. First, a shutter value that suppresses banding on one fixture at one dimmer position is evidence about that fixture at that position. It may hold for the second panel two metres away; it may not. Second, the trial is not as single-variable as it looks, because the compensation rides along. If you can't compensate with ND, you're changing two things and should say so in the log.
Route two: change the light, and pay in the look
Return to the baseline capture timing. Now change the illumination: remove the source from the take, replace it with a different type, or use controls the fixture's own documentation names — a different dimming response, a high-frequency or flicker-free mode, battery operation, whatever its manual actually describes. Modifying mains hardware is not a capture correction; it's a different activity with different risks, and it has no place in a test you intend to repeat.
What you're buying is your chosen motion rendering, intact. Every blur characteristic, every cadence decision, survives untouched. What you're paying is the photograph.
And the payment is larger than "the light is a bit different." A dimmed panel raised toward full output is not the same source at a higher level; it's a different quality, with a different contrast ratio and a different relationship between highlight and shadow. A swapped fixture changes falloff, colour and the shape of the reflection it throws into anything glossy. The source itself may be visible in frame — a practical in shot is a composed bright object, not merely illumination, and changing it changes composition.
That is the dangerous version of this route. If the light relationship is what the test exists to show, swapping or reconfiguring the light quietly deletes the subject. A location test that replaces the location's dimmable practical with a clean panel is no longer a location test; it's a studio test wearing a location's clothes.
The same two routes on one invented setup
Stipulate the following. It is a hypothetical written to make the decision legible — nobody recorded it for this piece, and no fixture was observed.
One bottle on a matte tabletop. A hand enters frame right, rotates the bottle a quarter turn, withdraws. About two seconds, four takes. The proposed key is a single LED panel, low and wrapping, dimmed to roughly a quarter of its output because the treatment wants the matte finish to read soft and the bottle's shoulder to stay off the highlight. Camera in movie mode, 24 fps, 1/48 second — a 180-degree shutter — manual exposure locked, white balance fixed. Its dimmer runs at roughly 250 Hz at a quarter output: a cycle every four milliseconds.
The saved baseline shows wide, soft-edged bands drifting up the frame. A take with no hand in it shows them too, so the exposure isn't hunting. At full output the panel shows almost none. On the camera's monitor the bands looked like a faint shadow; in the file, scrubbed frame by frame, they're unmistakable.
Route one: hold the panel at a quarter and sweep the shutter values the camera's movie mode lists — 1/60, 1/100, 1/125, 1/250, against the 1/48 baseline — one condition per take, compensating with ND and changing nothing else. The log does not slope in one direction. At this fixture's cycle, 1/48 stops a fifth of a cycle late and 1/60 a sixth; 1/100 stops exactly halfway through a cycle, where the split between the lines that catch the bright part and the lines that catch the dark is widest, and that take bands as badly as the baseline. 1/125 stops on two whole cycles and 1/250 on one, and both are clean. Which makes 1/250, the shortest value on the list, also an unnecessary one: it buys the same clean frame as 1/125 for a stop of blur you didn't have to spend. Take 1/125. The bands go, and the light stays where the treatment put it.
The bottle's rotation is the same turn, and not quite the same edge: the label travels through a shorter exposure, so it sharpens a little too. The hand is where that shows, because the hand moves and the bottle barely does — a quarter turn over two seconds against a crossing of frame right in a fraction of the same two seconds. At 1/48 the hand's leading edge carried a soft smear; at 1/125 the same edge is crisp and slightly stepped. The treatment wanted the hand to enter like weather — close to abstract. That quality is gone, and the framing was built around it. What route one leaves intact is the photograph: the panel, its level, the soft wrap, the small pool of shadow under the bottle, the falloff across the matte.
Route two: return to 24 fps and 1/48, and raise the panel to full output. The bands mostly go, because this fixture's modulation is worst where it's dimmed; the cycle is still running, there is simply less left in it for the leftover fraction to work on. But the key is now bright enough that the contact shadow beneath the bottle stops being the thing you notice. The treatment's entire reason for the dimmed key was that the bottle should sit in its own small pool of shadow. You have a clean clip of a different photograph.
Now the decision, stated as questions rather than settings.
If the test was asked does this rotation read at 24p, with this much blur, the only version that can answer is the one that kept 24p and 1/48. That's route two — and the price is that the clip now shows a lighting idea you didn't propose, so the lighting question is still open.
If the test was asked what does this dimmed key actually do on camera here, the only version that can answer is the one that kept the light. That's route one — and the price is that the motion in the clip isn't the motion you'd shoot.
If the test was asked both at once — which is the common case, because a test day is expensive and one setup is cheaper than two — then neither route answers it. Both routes are best understood as answering one question at the cost of the other. Split them. Keep the timing-change takes as a motion reference and the illumination-change takes as a lighting reference, label them as different documents, and don't hand a single clip to the room as though it settles both. The fixture's own manual decides which controls are available; this comparison can't supply them.
Choose on the files, then say what you gave up
Watch baseline, timing change and lighting change through the full repeated action, at full size, on a computer. Compressed playback hides band drift, and a phone screen mutes it entirely. A fix that looks solid across a two-second loop may reappear at the fourth repetition when the fixture warms.
Then write one sentence per version — what it keeps, what it changes. "Keeps the light, changes the hand's edge." "Keeps the hand, changes the key's falloff." Naming the relationship is what turns a setting comparison into a decision, and it's the part that survives into the edit and the client conversation.
There's an asymmetry worth weighing before you commit. A capture-timing change applies to everything you shoot under that setting, across every location and every fixture. An illumination change applies only to the shots that use that light. If one banded fixture appears in one setup, the local fix is cheaper. If banding follows you from location to location, the shutter habit is the more portable decision — and by far the more expensive one, because it rewrites the motion of the whole piece rather than the look of a single shot.
If a named value works, remember what it is: evidence about that setup. And keep post out of the plan as a repair route for light you didn't capture. Banding can sometimes be reduced later; exposure that was never recorded can't be restored by it.
So make the call and say it out loud, in one sentence, in the room. We kept the hand and changed the key; the light in this clip is not the light we proposed. Or: We kept the light and changed the shutter; the motion in this clip is not the motion we'd shoot.
Both sentences are usable. The unusable outcome is the clean-looking clip and the silence about which of the two things it stopped being.
Frequently asked questions
If a test clip shows a moving band of brightness, what are the only two families of fixes?
Change the camera's timing, meaning the exposure window and where it stops, or change the illumination pattern. A third move such as changing frame rate also changes temporal sampling, but it is a different test with different consequences, so it should not be folded into a shutter trial.
How can I check that the banding belongs to the light and not the camera or subject?
Lock exposure and shoot a take with no movement. If the band remains, the light is implicated; if the file settles, you were watching an automatic exposure decision. Reflective surfaces, screens, and practical drivers can create variation that follows the object. Bent verticals or leaning during a pan belong to readout geometry, not luminance banding.
Why can a shorter shutter make banding worse?
Band depth is set by the fraction of the light's cycle left hanging off the exposure window, not by window length alone. The widest split comes when the window stops halfway through a cycle, so a shorter value can band badly. The useful values close on whole cycles, and finding them is measurement against the fixture's own cycle, not frame rate arithmetic.
What do I trade when I fix banding by changing shutter versus changing light?
Changing shutter preserves the light but changes motion blur, because exposure time is motion blur; it also costs light, with neutral density as the cleanest compensation. Changing illumination preserves the chosen motion rendering but changes the photograph: dimmed versus full output, falloff, colour, reflections, and whether the source itself is visible in frame.
What if the test needs to answer both a motion question and a lighting question?
Neither route answers both at once. Split the test: keep timing-change takes as a motion reference and illumination-change takes as a lighting reference, label them as different documents, and do not hand one clip to the room as though it settles both. Then write one sentence per version naming what it keeps and what it changes.