I have spent quite a lot of time over recent last days getting to use, and hopefully understanding something about the use of the darktable module ‘filmic’. I have done this by comparing filmic results with how I would have otherwise typically have processed my RAW data using other darktable modules, and I tested results over a very wide range of my typical RAW data.
My personal finding is that when filmic is applied to a ‘well exposed’ data set it can produce, rapidly and with minimal use, even without the use, of other modules, a satisfactory overall result that can closely emulate a JPG directly produced by the camera’s firmware. If however the RAW data is ‘difficult’ then filmic requires (for me) more time and effort than my normal process and requires substantial intervention with other modules.
When I encounter a set of similar images from a session, in my normal processing, I can often cut and paste basic ‘starter’ settings over the batch … filmic does not appear to respond well to such usage.
I shoot RAW as a part of an overall creative process much of which occurre within the darktable process. My output image products are not intended to emulate the work of the camera’s firmware nor are they necessarily intended to look exactly like the original subject matter. I never treat my images as a single entity but as a collection of elements each one needing individual attention; that is something that filmic’s process appears is unable to provide.
For those looking for simplified processing, filmic may provide a solution for specific instances … for me I would be concerned that the full potential of the RAW data set may be missed by what I see as a short cut solution.
I did not consider that filmic met my measure of success at all. There are a number of ‘styles’ that would have produced, in my opinion, a starting position that would have been equally as good. Trying to ‘lock-in’, successfully, a position for exposure/contrast/saturation within a single module requires a lot of optimism.
show me your curves and I will tell you if your settings seem legit
reproducing JPEG from the camera has never been a goal (it’s not called “jpegic”)
filmic works in Prophoto RGB in a pipe that is in Lab. Stripping down every bit of Lab in darktable is the next step to ensure color consistency, but that won’t be done overnight and there is 10 years of legacy to deal with. Then, you will be able to chose a relevant RGB color workspace.
Hence the 5 color-pickers, to measure quickely in the picture and adjust the setting. If you want a one-size-fits-all setting, shoot a color-checker on set and build a filmic preset on the color-checker.
Also, filmic is not a creative adjustement but a technical parametric view-transform, meaning you take an high dynamic range, and you try to force-fit it inside the 8 EV of your output (screen, file, etc.) in a similar way film would do it, so your settings should retain some optical consistency.
@davidvj By reading your findings, I got the impression that you’re trying to compare two different approaches: free, creative editing, full of local adjustments, not necessarily aiming to reproduce what would have been the original scene, and scene referred editing, afaik, a way to preserve colors as close to the scene as possible.
From that, I don’t agree that filmic is a tool for simplifying the editing process. It seems to me that its goal is to compress dynamic range, still preserving colors as close to the scene as possible (as Pierre said, “you take an high dynamic range, and you try to force-fit it inside the 8 EV of your output”).
I also think that what you coin as “full potential of raw” can be thought differently, depending on what you want from raw editing. Do you want a creative, free approach or do you want a scene referred approach? I think both ways profit from the “full potential of raw”.
Anyway, I feel this thread risks becoming an infinite discussion, unless you bring concrete examples, as paperdigits mentioned.
There are different takes and implementations of “filmic” but their goal should be the same: to compress the dynamic range to a manageable size without making it unnatural and allowing the possibility to edit the image linearly.
I would say that some implementations are easier to use than others but I am thankful that @anon41087856 donated his time and effort to do one for dt, even though I won’t be using it because I don’t typically use dt. As he and others have pointed out, the implementation isn’t easy because of some of the choices that dt has made, which would need to be sorted out for the filmic module to truly shine.
Based on the feedback and discussion however. I do see the module as potentially confusing for the average user. A programmer, engineer or scientist could be top-notch in their discipline but ultimately it all comes down to the lowest common denominator, which is the user. If the user doesn’t get it or cannot use the tool to its full potential, then the tool isn’t being used as it was meant to be, which is a shame.
That is all to say that concrete feedback with examples is of utmost importance to its improvement, even if it is just specific to your workflow. It is entirely valid that there are cases where you wouldn’t want to use it. In the end it is just a tool.
Here is an image where the left version was processed easily with dt modules that I would conventionally use. The right image is an attempt to use filmic alone.
The foreground and the bag are processed separately in the left image.
I did spend some considerable time thrashing in an effort to obtain a reasonable match but the time spent simply did not achieve anything near to my goal. Pushing the contrast slider further did not appear to really help the problem. The log curve and my settings are shown.
On several images I have tried to use the ‘auto tune’ without finding acceptable results (or even a starting point) on any of my test images.
Ok, I think we also need some ‘screenshots’ of the module and a good s-curve what you should aim for in the documentation. I will try to create one with my color checker and update the docs. We also need to bring it upstream soon.
As requested, here is the RAW cropped as in the example. There are any number of ways to play this data … my interest is is looking at results and settingsVincent_Jones_Preliminary Layout.pdf (36.7 KB)
@davidvj I think the way you uploaded it prevent us to handle the file as a raw file. It is actaully another jpeg.
That file is already demosaiced.
If you upload a real raw file, it won’t show any image, it will show only as an attached file to the post.
I think you have to upload it the way you got it from your camera (a format like .dng, .nef, .arw… depending on your camera model)
Unless you haven’t originally shoot in raw…
OK … will try again. I had sent a cropped TIFF that I had assumed would work … apparently not, so here goes for the full RAF file.20181020_0038.RAF (48.1 MB)
The image that shows right above this sentence was obtained by using filmic? If so, can you provide the xml file?
This one is tough, I couldn’t get close that late image you’ve uploaded.
The image that I sent earlier was NOT using filmic … that was worked using my conventional processing. What I was hoping was that you were going to demonstrate to me how filmic was able to provide at least the basic process on an image of this type.
Yes, this is a tough image to work on and I also have a series of similar shots, many of which are under even more harsh lighting conditions.
I understand that filmic is still being tweaked … I will watch its progress with interest.
I believe that the module is considered finished but there is always room for improvement! Keep the feedback coming; in particular, how you approached it (i.e., your step by step experience). The reason I encourage that is because showing the final slider values doesn’t tell us the whole story.
Just looking at your settings, you have 25 EV of dynamic range… No camera reaches that, we are at 14.9 EV max at 100 ISO for the best workhorses… That’s exactly what I meant by “settings should retain some optical consistency”. It’s written in the doc that the dynamic range inside the software is bounded by the physics of your actual sensor. So you didn’t RTFM.
Also, I have been doing tests for 2 weeks to add a chrominance handcuff. This ensures the hue and saturation are retained between the input and the output of filmic. I still have some problems with that, but here is a preview with color balance and filmic:
The reason I encourage that is because showing the final slider values doesn’t tell us the whole story.
