Replicating black mist filters in darktable

Has anyone tried to replicate in darktable the effects of black mist filters? How would you go about making highlights glow like that? In my opinion, the most notable difference wrt to the bloom effect is the hue of the glow around the highlights. For me, it’s very difficult to emulate.
(in case you are not familiar with such filters I recommend this video)

What I tried is to use the bloom module and combine it with a separate instance of color balance rgb to add the hue in the highlights combined with some draw+parametric masks. The results were mostly okay, but all tedious manual adjustments are needed every time (hue and masks). It would be awesome to have something easier.

I would try the diffuse or sharpen module: darktable 4.0 user manual - diffuse or sharpen

If you want the most pain free way… Just buy the filter :slight_smile:

1 Like

I’d have a go with the lowpass filter trying differents overlay mode (lighten only) with high radius, high contrast and play with saturation …

If you want the most pain free way… Just buy the filter

I am a fan of Nisi filters, but they are not cheap. Here in the UK the 1/4 stop, circular black mist filter for my standard lens is £95…

From here

For me, the way to do this effect has been to observe the way in which it happens naturally on film and then emulate that process. I observe that:
1. The halation has the characteristic of flaring, and not of brightening, around bright objects. This is why it is most noticeable when the bright object is surrounded by black (i.e. traffic light at night). When the background gets brighter (i.e. traffic light at day), the halation is less or not visible. The "spillover" effect still happens, only we don't see it (the overexposed silver halide doesn't "know" if the ones next to it are also overexposed).
2. The brightness of an object modulates both the luminance and the width of the halo.

It follows then that:
1. The emulation needs to be processed in scene-linear (gamma 1.0) to represent the flare.
2. Its intensity is dynamic in terms of nits and pixels. That means neither qualifiers, nor edge detection will be useful.
3. It's alright if it affects the whole image, as long as it's proportional to the intensity of the light and the darkness of the surrounding pixels.

Here is my recipe:
1. Convert the image from the working space curve to linear (gamut stays the same).
2. With curves, force all except for the super bright spots (the "halo emitters") to 0. Being in linear-light, the curve has to be quite aggressive because the highlights are way outside the usual 0.0-1.0 range (see image).
3. Give it the orange tint.
4. Blur. The blur amount is the maximum flare width.
5. Add the result back to the original (linear) image.
6. Convert from linear to working space curve.


Reddish glow

I’ve used gimp and I have inverted step 3 and 4 , I’ve used rgb curved for the orange tint

Another interesting discussion

Thank you very much, that is very useful! I will respond in this thread if I succeed in replicating this formula in darktable