No. Where is it?
Base primaries determine where the corners (primaries) of the triangle (color space) are to begin with.
Purity/Attenuation moves the primaries closer to the white-point.
Hue/Rotation rotates the primaries around the white-point.
Tint changes the hue of the white-point.
This graphic is for both sigmoid primaries and rgb primaries.
12 Likes
1 Like
Yes.
Still sounds like CA, but please do post a raw + xmp that shows the issue.
Thanks!
Can also be clipped highlights and highlight reconstruction (the module, not in filmic) failing. Try this: Magenta highlights vs raw clipping indicator vs filmic white level - #84 by 01McAc
With highlights saturation mix at 0:
+50%:
-50%:
It’s not caused by filmic, but the module makes the problem more visible. Exposure dropped, filmic off:
Contrast EQ sharpening to bring out the sharp transitions:
CA is similar, filmic can make it more evident. It goes away / is much reduced with highlights saturation mix = 0, or filmic in v6 mode and norm = luminance Y or no preservation.
3 Likes
I notice this often even when it’s not caused by CA.
In the real world we look up and see only a clear blue sky, and dark branches against it, and nothing in between. If the filmic white point is set high enough the sky is all blue; if it’s set lower it transitions from blue to white which usually looks very obviously wrong, and if it’s set even lower the whole sky is blown out white, which might look ok in some cases.
There will always be some pixels along the edges of the branches where both the sky and branch contribute to the color of that pixel. If the filmic white point is set low enough these usually end up as ugly blue halos around the branches against the white sky. Or even if the white point is higher so the sky is blue the chroma preservation turns those edge pixels into something like a dark saturated blue that looks wrong.
I’m looking for ideas for the best way to mitigate this, as it happens so often…
1 Like
Try the details slider and primaries will likely help… but I found using CC instance, blue channel in gray tab between .5 to 1 and also use opacity and mask with details threshold… you can get a nice desaturation of the blue along details like branches…
I’d still like to see some samples.
1 Like
Yep, I agree. This is one of those classic examples… (in my view).
CA can often contribute, but even without that it can still look ‘wrong’. To me. But not everyone thinks so. I think it must be a bit of a opinion/perception thing.
Yup, this sounds like a great candidate for a Play Raw. It sounds like CA to me, but seeing a RAW will help us determine whether that’s the case or not. And regardless of whether it is CA or not, we can all provide our attempts at correcting it.
Just want to add some stuff. I’ve been editing hundreds of shots with both filmic and sigmoid. My conclusion is this: you can get very similar results with both. However, sigmoid has very nice defaults, especially when you want to shoot shots like those with bokeh circles that are overexposed and the like.
To get a similar result with filmic, you have to push the dynamic range scaling to the right and the highlight reconstruction “transition” slider to the right also. You might also need to do other tweaking and even then there might be some harshness unless you get it exactly right. The overall workings of the sliders feel like alchemy a little bit.
Sigmoid seems to get exactly right upon activation, and a little contrast and skew adjustment seems very logical. Overall, I think sigmoid is better for the vast majority of images though I understand why some people like filmic.
Personally, sigmoid + curve adjustments are much more natural…
4 Likes
And that might be the main difference between filmic and sigmoid…
With filmic, the only curve adjustment I do outside filmic is in the tone equaliser, and even there only for some images.
So you’d have to do similar corrections (logical), but you do them in different places.
That is true. However, I am not a fan of the tone equalizer. It’s not fine-grained enough compared to using actual curves like RGB curve.
1 Like
But that’s partly a matter of personal preferences, experience, and needs.
Which makes the whole discussion about which is better, shall we say, complicated. Filmic, sigmoid and basecurve are different tools for basically the same job, with different approaches and trade-offs.
Personally, I don’t think one of them is always better. But they all differ in what has to be done outside that particular module. Sigmoid is simpler as a module, but it requires some actions (or different actions) with other modules, which for me doesn’s make the overall process easier or faster.
Personal pref and needs of the image.
Sometimes one gives better results, sometimes the other one. Simple as that.
Normally, Filmic is all-around module. But sometimes the other gives better results.
And it doesn’t have to be. Its main purpose is to act on tonal ranges. When increasing contrast, its special feature, detail preservation, is not that visible. But try compressing the dynamic range. Since it does not decide what happens to a pixel based on the value of the pixel, but rather based on the value of the (blurred, averaged) mask over the pixel, it will treat neighbouring pixels in a similar way, even if they have quite different values. When lifting shadows, mid-tone pixels in a dark surrounding can be lifted along with the surroundings, keeping local contrast; when lowering highlights, a pixel having the same (mid-tone) value will be darkened together with the bright pixels surrounding it, again, maintaining local contrast:
Lifting the shadows by 1 EV:
dark pixel with value -6 EV (relative to mid-grey) → lifted to -5 EV (below mid-grey)
midtone pixel with value 0 EV → lifted to 1 EV
Their difference is maintained.
Compressing the highlights in the same tone EQ instance by 1 EV:
midtone pixel at 0 EV → lowered to -1 EV
bright pixel with value +3 EV → lowered to +2EV
Again, their difference is maintained.
You cannot do that with a pixel-wise curve: you can lift -6 EV to -5 EV, leave 0 EV as 0 EV, lower 3 EV to 2 EV, but the difference between the midtone pixel and its darker/brighter neighbour will be diminished.
The big downside is it can introduce halos, so blurring and edge-awareness have to be managed.
3 Likes
What do you mean exactly?
With tone eq, using the masking options, you can target very precisely the luminance ranges you want. And, as @kofa says, you get very good control over how much local contrast is affected. It does require a good deal more understanding and experience, granted.
AP, the creator of tone equalizer, stated that TE is designed for local tone mapping, although it can also be used for global tonal mapping. Sometimes global usage requires two instances to cover the dynamic range, though.
Thanks for a good explanation of TE’s maintaining of local contrast. I’ve come to realize that masked instances of TE are better for dampening too salient details in an image rather than Exposure that may end up in a uni-toned blob. Now, I understand better why and where the blur comes into the picture. (How to read the guided mask of TE, I’m not quite sure about yet, but regrettably we’re in the wrong thread to ask about that …)
Yes, I think this is key. It’s a good example of a module where lots of practice and experience pays off. It can be frustrating at first and I understand why some people don’t like it (especially if they are coming from more classic tone curves). But once you get a deep understanding of it, its power really shines through.