Some preamble so you know where I’m coming from.
Bought a Fuji digital fixed lens zoom bridge camera about 15 years ago, and it was only in 2020, during one of the lockdowns that I found the time/energy to pursue this interest further and learn about cameras, and photography properly, enough to start taking photos in manual mode, and more recently with a Sony mirrorless APS-C, a much more recent aquisition, now using other modes like aperture priority.
Over the last year, invested a lot in learning darktable, for only one reason. In a quest to find a good photo editor, that supports raw files, which I did not have to pay a subscription for, as photography is still a hobby for me, and sometimes I do take a hiatus of several months away from this love of mine - especially when I get frustrated that my pictures are not as good as I would like them to be.
This frustration in itself has been a difficult learning, starting, abandoning out of frustration, and gathering the energy to start or continue again, and over time, inch by inch, the pictures are getting better.
Why the improvement? I understand a lot more about light, the essential without which there is no photography, and I have become adept at using more of the functionality of darktable. The two important things I love about darktable are it is super stable, almost no crashes, and the quality of the image display within the tool is second to none, in the open source world. What you see is somewhat much better than the final export to a jpeg, and the benefit of its 32 bit floating point processing is there to be seen, if only we did not have this cludge of an export format called jpeg, which always degrades the quality - such is life.
In more recent times, for about the one year I have been investing in learning photography and darktable, one of the darktable authors, Aurelien, has invested so much time and energy to create a module called filmic-rgb, which aims to solve some of the shortcomings of a display referred approach to processing, and provide a better solution for processing images using a scene referred approach. The benefit of the scene referred approach being it aims to process the image more like an optical device, rather than a digital device.
The de-factor mode of processing, the display referred used a module known as the base curve, to transform the image from the raw input, to one which we can see in RGB, and display. Aurelien indicates that there are limitations with using the base curve, such as the way in which at the extremes, such as highlights the relationship between chrominance (colour) and luminosity(brightness) breaks up, and this display referred processing does funny things to the image, when pushed to extremes.
The base curve is similar to the camera profiles in Adobe Lightroom, and includes profiles for different manufacturers e.g a Nikon profile, and a Canon like profile, or a Sony like profile. Not exactly like Adobe where the profile is somewhat individualised for specific cameras, but in darktable these base curve presets are generic for supposedly similar styles that each camera manufacturer is deemed to favor.
By and large, I have learnt to get decent results with either the scene referred approach using filmic-rgb module or the display referred approach using the base table. Not perfect results, cos nothing is ever perfect, but good enough, not to be ashamed of my photos, and able to share them with others.
There is another approach which I have also used, which relies on LUTS as the primary transform method to convert RAW to something we can see. A most strange use of LUTS, but it does produce decent results, when you know how - lots of caveats cos this is a somewhat unorthodox use of LUTS, as the input to LUTS is typically a log image in something like Sony S-log2 or Canon - C-log, not a raw photo file.
Why all this palava? As Aurelien explains, our cameras can capture in typically at least 12 bits, which have to be converted to 8 bits for most of us to view the damn thing on a screen. Furthermore the data must be converted from a linear raw format, using some appropriate curve bending to bring it into a format that can be visualised.
This is not just a problem for photo editor apps, but also one which has to be solved whenever a camera produces jpegs, and this includes the embedded jpegs included in all Raw file formats. So - as many others know, and I only just began to understand this, every camera does this in a non identical manner, wherein lies a lot of their claim to fame, some of this based on their sensor, but a lot of this is also based on the magic of their algorithms, to transform the image into something that highlights aspects of the image considered a priority, such as human skin tones.
Not just in photography, but in video, we have the same “intellgence” with expensive cameras like Arri and RED, all touting their special “colour” science, a proprietary set of algorithms, which transform the RAW file (or raw image captured by the sensor) into something we can see, even if that process takes it through intermediary stages like conversion to a log based format.
For the vast majority of photo and video editors and users, this transform from RAW (or sensor capture) to jpeg, compressed video formats, or via intermediary log formats, involves an unavoidable encapsulation of some pretty hefty know how, by which each manufacturer has baked in a whole lot of knowledge and experience of how to do their own version of this transform. Taking a 12 (or 14 or 16) bit file down to 8 ( or 12 or 10) bits and going from the raw linear format to what we can see, is not a straightforward conversion, otherwise everyone would do it the same way.
While the base curve at least attempts to bake in some pretty decent intelligence into this transform, the filmic-rgb module, leaves the end user pretty much to their own devices, and therein lies the problem.
I can imagine that the profiles/presents included with the base curve are based on some comparison of RAW files and corresponding JPEGS, from each manufacturer and an attempt is made to provide an approximate version, through this approximate “reverse” engineering, to make an intelligent guess at the transfer function., so that once you select the right preset, you can reasonably expect that your image will turn out about ok, of course assuming you have got the right white balance and exposure.
With base curve you have a safety net. With the new filmic module, however well intended, no matter how many parameters are included in the tool, you are still somewhat left to the decisions included in the algorithm, by the developer, and your own intuition. Whatever result you end up with, is a bit of a stab in the dark. However good it is, this transform cannot provide the kind of safety net that is built into alternative solutions like Adob Lightroom, Adobe Camera raw, and Capture ONe, etc, etc. Raw therapee also attempts to derive a transform, from the embedded jpeg in the raw file, somewhat a crude effort, cos the quality of such reverse engineering completely depends on the quality of image taken., light conditions and range of colours within the RAW and accompanied/embedded Jpeg file - something that cannot be guaranteed.
No matter how accurately one creates image processing tools, as long as the image is not transformed using a method that prebakes the look, which is what I finally understand and appreciate is the value of tools like Adobe Lightroom and Camera Raw, providing a decent stating point, with the “look” baked in, it is pretty difficult to aim to do anything as good or better than the visual transform intelligence that is baked in in all modern digital cameras, and software like Adobe’s.
I have found the base curve which comes with some starting points a much more reliable manner of developing raw photos in darktable. The filmic module leaves you completely on your own, a great tool. but its like having to make make your own suit or yoru own shoes.
The whole point of a tool is to help you, not give you another set of raw materials as a starting point, or tools which you have to further learn. With Adobes tools for example you get an image in decent condition, and all you are doing is adjusting it, and so much of the final look is already baked in, before you even start tweaking.
With the filmic rgb, you invest so much effort into creating this baked in look, and have to do this for every single photo/image…
I can imagine it took thousands of hours and millions of dollars in tools, to discover the image preferences baked into the top camera products or into Adobe tools.
What filmic needs is not anothr set of levers presented to the user, but a set of look profiles as presets, similar to base curve, as a starting point.
I appreciate that the base curve module may have innacuracies, but filmic is another layer of tooling and not a replacement for base curve module. What we need is a scene referred version of the base curve module, with a similar bunch of presets, and then one can tweak this further with a tool like filmic (as an optional next step) or any of the other modules of darktable.
A tool like filmic can rather than solve a problem, which is give you a decent starting point, similar to the in-camera generated Jpeg, may lead you down a long path of your own making tweaking settings, and it is only as I write this that I ask myself, why do we need this additional headache as the starting point of raw development.
I like the idea of base curve - pick a starting point, and tweak from there. With filmic, we should be able in a similar way to base table, given a well exposed photograph, filmic shoul have defaults, or default set of presets we can choose from, which give us a pretty decent starting point, taking our raw input to a nice presentable well developed image with no further tweaking needed. But at this time the default filmic settings rarely work well with many photographs, and it just becomes a guessing game, having to tweak about 20 different options in filmic, to get a starting look, for further editing in other modules.
Maybe what we need is a project to develop such presets for filmic that work well with any well exposed image, and these presets can then be included in the darktable presets that are installed with filmic.
Does anyone find value in such an effort to rescue filmic, and repurposing this tool, via the addition of a very good set of presets. The advantage of such presets, is that the developer has nothing to do, we the end users can collaborate and share and contribute this back to darktable - interesting idea, cos what we are contributing my not be computer code, but is identical to the thousands of years of image intelligence that is behind various film chemical compositions in the days of film, and in the baked in visual transforms behind all the worlds succesful digital cameras.
Especially if there is some sort of attribution and if the contributors also get a special mention, that would be enough reward for the effort.
Is anyone interested in working with me and possibly feeding back these presets to the darktable team to improve the filmic module with in-built presets that work well as a decent startig point? And like Adobe, I can imagine these presets improve over time, and the developer of filmic gets insight into what features provide the best benefit in the filmic module as feedback for any further improvement.
I was looking at the filmic module and guess what - right in the module it uses the same words - color science v3 and v4. so color science is a valid idea.
But rather than giving us a decent set of color science options as starting points, like every other similar tool, what we get is a set of engines and tyres to go and create our own car. I hope my ideas of creating presets for filmic are valid and produce a positive roadmap for the product.
Pretty much like when you pick up any camera and you have different kinds of looks - vivid, standard, etc. That’s what effective color science is about - a bunch of 5 to 10 great simple starting options, not a menu that involves over 500 possible combinations of different settings !