the first step is to use the diffusion in the top layer on the green channel and set this layer to “lighten only” (the diffusion is green when a pixel becomes brighter)
the second step is to use the diffusion in the middle layer on the red and blu channels and set this layer to “darken only” (the diffusion is green when a pixel becomes darker)
I just noticed that Ricoh added a HDF filter to their renowned GR III.
I experimented with the suggestions above again and found that they an easily replicate something very similar. Specifically, a not too wide radius span, 1st order diffusion, and edge threshold to taste.
Looking back at 2024, I think that this topic had the largest impact on how I develop images in Darktable.
First, I started noticing the look associated with black mist (& friends) filters in movies. In contemporary cinematography, I find it nearly ubiquitous. If you know what to look for, it becomes challenging to find a cinema image without it.
Then I developed some styles that blunt the edges a bit for images I take, especially with backlight or specular highlights. I apply these selectively, often with an edge preserve threshold, and 50% opacity. Maybe 25% 1st and 2nd order speed, and one iteration, tweaked to taste, and maybe masked to exclude the main feature of the composition. Now my images look more “natural”, even my cheapo primes were too sharp.
Also read through this thread again recently after watching the following nicely produced video that suggests using a quarter strength mist filter to try to blur the hard edges of clipped light sources in night photography.
Incidentally, I think he’s wrong on the slowest shutter speed you can use, given stabilised sensors and optics, though of course you might need higher speeds to freeze moving bodies.
It is unfortunately overused. I think younger cinematographers have a romantic view of how older films used to look like, based on watching older scans/transfers.
Some films do grant the look, and even back then had these effects exaggerated by using old(er) lenses, vaseline on the lens, etc. A good example is Wong Kar Wai(Christopher Doyle as cinematographer).
But a lot did not, especially when shot on 35mm or larger. We only need to watch one of the recent scans of Lawrence of Arabia, as probably the most absurd example of pristine picture quality. Some modern films shot on film also look way more “pristine” than digitally degraded ones, such as Dunkirk, Tree of Life, A New World, Samsara, etc.
Hopefully this trend will slow down and it will be reserved for films that really need it and are enhanced by it. And I say this as someone who has glimmer glass stuck to his x100v
Indeed it is excellent for that day and age, but lenses have developed since then. Now extremely sharp lenses are well within the enthusiast price range, which creates a look that most people find straining. A tiny amount of “mist” tames that (of course there are other ways).
Indeed. One thing I love about night time street photos is the ability to blur people with 1/10s handheld shots using a stabilized body. Or if I have a mini tripod, any surface will do and I can do 2–5s exposures, which I find ideal for moving crowds, as it preserves some sense of people being there without making them recognizable.
In a lot of cases this can easily be fixed in Darktable by adding some exposure to specular highlights before diffusion. Eg here I masked the LED lights on our Christmas tree, which were clipped (I found the reconstructing in LCH gives the smoothest results), added +2EV, and then applied D&S.
This does not change the appearance of these highlights since sigmoid maps them to 1 anyway. But in the linear part of the pipeline, they add more “mist” with D&S.
This idea can be varied to taste, eg you can also color these lights to get a reddish haze, etc.
(Not starting a new topic for this, so I will add it here in case anyone is interested.)
I have always wondered how much mist filters dampen the highlights. This is relevant for me because, replicating them in post, I often underexpose the image so that I have light to diffuse into that desired halo. (The reconstruction method I proposed above is fiddly and I like to save time).
This week I experimented with some mist filters (K&F black most 1/4 and 1/2, but these things are not standardized) that I borrowed from a friend. I shot some point-like light sources (decorative LEDs) and also wider ones (a LED light bulb with a diffusion cover, from close up so that it takes up about 10% of the image width).
The method is the following: using a tripod, ETTR using full manual aperture, ISO, and shutter speed. Then put on the mist filter and expose with the same settings. Measure the points with no tonemapping or any other transformation, just demosaicing in Darktable.
On point-like sources which are tiny, I found that the filters reduce about 1–1.5 stops. Curiously, I did not find a large difference between the 1/4 and the 1/2. On larger light sources, the reduction is about 1/3 stops or less.
But, to get a convincing halo in Darktable using diffuse & sharpen in post, I don’t need full ETTR. It is enough if the light source is about 1–2 stops brighter than its surroundings. This is a relief since LEDs can be very, very bright when not diffused and ETTR would require -3 stops, making my shadows very noisy.
D&S is wonderful, it allows me to achieve the look I want. I often keep the central radius large to preserve local contrast, or use harmonic mean.
