Red filter, or software?

Neither tool can multiply zero

Both a red filter and a software correction do the same arithmetic to your picture: they change the balance between the channels your sensor recorded. A filter does it before the exposure by holding back blue and green, software does it afterward by lifting red. Either way the red in the result comes from red that arrived at the lens.

So both run into the same wall, and they run into it at the same place. Measured off color charts at roughly 10 meters in clear tropical water, the red channel read exactly zero - not dim, not noisy, zero, in both frames and across the red, yellow and skin-tone patches. There is nothing there to rebalance. A filter cannot hold back enough blue to reveal a red that never arrived, and no correction can rescale nothing into something.

An illustration, not a measurement - the measured version is the bars above. A filter divides the other channels down and a correction multiplies red up. Both change the same ratio, and neither operation moves a zero.

The threshold is a distance, not a depth

Advice about filters is almost always phrased as a depth - take it off below fifteen feet, put it on above ten meters. Our measurements say the depth gauge cannot predict this. Red died by 3.4 m of water at the shallowest site we measured and survived past 10.2 m on a clear Red Sea wreck, a range of three to one that has nothing to do with how deep either camera was.

What governs it is the water itself and how much of it sits between your subject and your lens. In murky green coastal water a filter can be doing nothing useful at three meters. In clear tropical water it can still be earning its place at ten. The same filter, the same depth, opposite answers.

Each tool wins somewhere, and the honest answer is both

A filter is worth carrying when you are shooting in available light, in blue water, at distance, and you would rather spend the light than the editing time. It costs you one to two stops, it is wrong the moment you turn on a strobe or video light, and what it does cannot be undone afterward.

Software is worth having because it works on the footage you already came back with, including everything you shot before you owned a filter, and because it can be dialed back or discarded without touching the file your camera wrote. It cannot invent light that never reached the sensor.

Neither is a substitute for the one thing that actually changes the physics.

Closing the distance beats both

Every meter of water you remove between your lens and your subject is a meter that cannot take red out of the picture, and the effect is far larger than the difference between any two correction methods. Inside one measured photograph, spanning 6.5 to 29.7 meters of water, the water's effect on the blue channel varied by 254 times - within a single frame, before anyone touched it.

This is why underwater photographers are told to get close and then get closer, and it is the one piece of common advice our measurements support without qualification. A filter helps at the margin. Software helps afterward. Two meters closer changes what arrives.

How this was measured

The numbers here come from color charts photographed at known distances on four dive sites in a public research dataset, two in the Mediterranean and two in the Red Sea. We read the channels off the charts rather than estimating them from the scene, which is what lets a figure be quoted as zero rather than as nearly zero. Anyone with the dataset can reproduce them.

Dataset: Berman, Levy, Avidan and Treibitz, “Underwater Single Image Color Restoration Using Haze-Lines and a New Quantitative Dataset”, IEEE Transactions on Pattern Analysis and Machine Intelligence, 2020. 10.5281/zenodo.5744037

Questions this raises

Do I still need a red filter underwater?

A red filter and software are different tools and neither retires the other: a filter works at the moment of capture, in blue water, and costs you light, while software works afterward on footage you already have and can be undone. Neither recovers red past the distance where it has reached zero, measured at between 3.4 and 10.2 meters of water depending on the site.

At what depth should I take the red filter off?

Depth is the wrong variable. What removes red is the distance light travels through water between your subject and your lens, and the kind of water it travels through, so a filter can be useless at three meters in green coastal water and still working at ten in clear tropical water. Advice phrased as a depth in feet is a rule of thumb for one kind of sea.

Can software recover color a red filter would have caught?

Only where some red survived. Correction rescales what the sensor recorded, so it can lift a weak red channel a long way, but a channel that measured exactly zero holds no information to rescale and no tool returns it. This is the same limit a filter runs into, reached at the same distance.

Does a red filter work with artificial light?

No, and the two should not be combined. A strobe or video light puts full-spectrum light on a subject a short distance away, so the red is already there and a filter in front of the lens only takes it back out. Filters are for available light in blue water.