Basti's Scratchpad on the Internet
09 Oct 2021

Comparing the Fujifilm XF 16‑80 with the XF 18‑138 Travel Zoom Lenses

When I bought into the Fuji system, I selected the XF 18‑135 f/3.5‑5.6 R LM OIS WR as my main zoom lens. This is a lens with a very wide focal range, that is commonly called a “travel zoom” because you could travel the world with just this one lens. And indeed I happily did. In 2019, Fujifilm released a second travel zoom lens, the XF 16‑80 f/4 R OIS WR. Ever since, I have wondered how this new 16‑80 compares to my 18‑135. But given that these lenses are somewhat similar, few people on the internet were ever able to compare them side by side. This blog post will change that.

The 18‑135 has served me very well indeed. Of course it is not the world's brightest lens, nor sharpest, nor smallest. But so long as I can get the shot, these limitations don't bother me. Except for two things: The transition between in-focus and out-of-focus can be a bit rough, and my lens extends on its own when carried on a sling. My hope is that the 16‑80 has a nicer rendering, especially for people photos with out-of-focus backgrounds, and a stiffer zoom ring that doesn't creep.

Picture of Squirrel, with hazy bokeh
Instead of smoothly blurring, the near-focus areas of the 18‑135 have this strange, hazy look

Physical

Physically, the 16‑80 is a slight bit smaller (89 mm vs 98 mm) and ligher (440 g vs 490 g) than the 18‑135, but the difference is negligible on camera. If anything, the 16‑80 actually feels a bit bigger to me due to its larger front element. Some people claim that the 16‑80 feels better built than the 18‑135. To that end, the rings on my 16‑80 turn more smoothly than my 18‑135's, but then that is comparing a new-ish lens to a well-used one. Autofocus speeds are also reputably different, but they feel similar to me.

The 16‑80 has a numbered aperture ring that allows for adjustments while the camera is turned off. On the other hand, the 18‑135 can switch to and from auto-aperture without losing its preset aperture, which is useful as well. The ideal lens for me would have both the auto-aperture switch and the numbered aperture ring. Oh well.

A picture of the two lenses from the side, both collapsed and extended A picture of the two lenses from the front
Size comparison of the two lenses. The 18‑135 is a longer, and the 16‑80 wider. Note the marked aperture ring on the 16‑80, and the Auto-Aperture/OIS switches on the 18‑135.

The 16‑80 has a 72 mm filter thread, while the 18‑135 uses 67 mm filters. This is somewhat annoying for me, as my filters are all 67 mm, which also fit on my 70‑300. I'd imagine users of the 72 mm 10‑24 see this differently. Anyway, using a thin polarizer with a 72‑to‑67 step-down ring works without issue on the 16‑80. My inch-thick macro filter however does vignette heavily until 23 mm.

Rendering

In terms of rendering, I find the 16‑80 to have a gentler transition from in-focus to out-of-focus, and indeed render out-of-focus backgrounds more smoothly than the 18‑135. At the long end, the 16-80 is actually a useful portrait lens, which is an unexpected but welcome feature for my photography.

A picture taken with each lens, with a blurry background
Rendering of out-of-focus elements. Especially the points of light in the top right have an asymmetric crescent shape for the 18‑135, but blur smoothly for the 16‑80.

At the wide end, the 16‑80 exhibits some significant distortion. While the camera or post processing programs can easily correct this, it leaves the image corners stretched, and renders the 16‑80's nice round bokeh balls as ugly ovals. Thus large-aperture shots at the wide end can be somewhat problematic.

Resolution

Next, let's compare the resolution of the two lenses. To do that, I printed a test chart, and took some test images. All images were taken in identical illumination, on a tripod, with the chart covering half the sensor height/width. The following graphic shows a resolution scale near the center of the frame and near the left edge of the frame. And just for fun, I've thrown in a similar analysis from two prime lenses as well. The resolution scales are in 200x line widths per picture height. On my 6000×4000-pixel sensor, the theoretical maximum resolution would therefore be a resolution number of ⁴⁰⁰⁰∕₂₀₀ = 20. All test images are straight crops from original images, reproduced at their original resolution.

A grid of crops of a test chart, with various lenses at various focal lengths and f‑numbers
Comparison of lens resolution with a test chart. Each image is a direct crop at the X‑T2's native resolution. Resolution numbers in 200x line widths per image height.

Center images were focused in the center, and side images were focused on the side. A two-second timer was used to eliminate camera shake. Due to the geometry of my room and the size of my printer, all pictures were taken near the close-focusing distance of the lenses. A red line indicates the limit of resolution of these lines, as judged by my eyes. The corresponding resolution numbers are generally consistent with the ones published by opticallimits and lenstip1, albeit I judged them a bit more conservatively.

The results of this test are surprising to me: the 16‑80 is actually my sharpest lens. It not only bests the 18‑135 at all settings, but also the 35 f/1.4. Only the 60 f/2.4 can reach similar resolution at the same f‑numbers. Generally, the 16‑80 is perfectly sharp right from f/4, while the 18‑135 has to be stopped down to f/8 to reach a similar resolution, especially on the image edges. Only at the long end does the 16‑80 benefit from stopping down.

That said, take these resolution measurements with a grain of salt. For instance, a “great” result of 15 (3000 LW/PH) translates to a blur radius of 1.3 pixels, while a “mediocre” result of 10 (2000 LW/PH) is instead 2 pixels. This is scientifically significant, but not at all relevant to (my) photography.

Additionally, the fact that the 60 f/2.4 macro lens scores so highly but the 35 f/1.4 does not is an indication that these measurements might be biased by being taken near the close focusing distance of the lenses. Thus the next set of images compares these lenses at more natural distances.

A grid of crops from a landscape shot, with the two lenses at various focal lenths and f‑numbers
Comparison of lens resolution with natural images. Each image is a 500x500 pixel crop from original photographs. The subject is ca. 100 m from the camera.

At this farther distance, and with a more natural subject, the differences are no longer as easily visible. What differences there are this time favor the 18‑135 instead of the 16‑80. Interestingly, I didn't see any significant differences between these pictures when looking at them “merely” side-by-side in Capture One. Only when I actually assembled these here graphics and looked at them at 200% did the differences become apparant.

Nevertheless, it remains curious that there would be such a difference between the two lenses. Then, someone mentioned that the 16‑80 might suffer from shutter shock, where the camera's mechanical shutter jolts the camera enough to upset the image stabilization system and induce a slight bit of motion blur. An issue such as this might explain the 16‑80's slightly reduced resolution in my test shots. So I created another series of images, but this time both, with the mechanical shutter, and with electronic shutter. In electronic shutter mode, nothing moves in the camera and there can be no shutter shock.

A grid of crops from a landscape shot, with the two lenses at various focal lenths and f‑numbers and with electronic and mechanical shutter
Comparison of lens resolution and shutter shock. Each image is a 500x500 pixel crop from original images. The subject is ca. 100 m from the camera.

From this comparison, I can see no evidence of shutter shock. It might have been an issue on earlier firmware versions of the 16‑80, but my camera (an X-T2) and lens (at firmware 1.05) does not does not exhibit shutter shock. Furthermore, this series of pictures shows the 16‑80 and 18‑135 essentially matched in image resolution.

All of that said, I must add that all of these comparisons used extremely tight crops of high-contrast geometrical features. Most of the differences here are all but invisible in actual photographs. From these resolution experiments, I see no reason to prefer one lens over the other. Both of them are perfectly sharp in everyday use.

Conclusion

So, how to choose between the Fujifilm XF 16‑80 f/4 R OIS WR and the XF 18‑135 f/3.5‑5.6 R LM OIS WR? My 16‑80 has tighter aperture and zoom rings, does not creep, and has a smoother rendering of out-of-focus background. On the other hand, I do find the increased telephoto of the 18‑135 very useful, and it doesn't suffer from wide-angle distortion as much as the 16‑80.

In terms of resolution, I did not find fault with either lens. Both are very sharp across their entire focal range and the entire frame. That said, the 18‑135 does benefit from stopping down for optimum resolution, while the 16‑80 is sharp right from f/4, and the 16‑80 might be sharper for closer subjects.

My tentative conclusion from these experiments is therefore that I would slightly prefer the 16‑80 for people pictures, where the close-focus sharpness and nicer background rendering are advantageous, and the larger aperture at 80 mm might make a difference. And I would prefer the 18-135 for landscapes, where stopping down is usually easy and the longer focal length comes in handy.

That said, the differences in rendering and resolution are really very minor, and the choice most importantly comes down to the focal range. Which is as it should be with modern lenses. And both lenses are of course very well-built, weather sealed, and have fantastic image stabilization. But you probably knew that already.

Footnotes:

1

multiply lenstip numbers by 2×16.7 mm to convert from lpmm (lines per millimeter) to LW/PH (line width per picture height)

Tags: photography

A Review of Darktable

With version 3.0, my favorite photo editing software Darktable started the journey towards a scene-referred editing pipeline. Which means that most edits are no longer bounded between a fixed black zero, and a white one, but can range between zero and infinity, like light itself. This is the norm in video editing and video games, but unique in photo editors at the moment. The scene-referred pipeline has brought changes to pretty much all parts of the editing workflow. I have been frustrated with it, too, even opting to use Capture One for a while. But with version 3.6, Darktable seems to have found a new stride. This article is about my favorite features in Darktable 3.6, contrasting it with Capture One's more traditional toolset.

A camera sensor records light. When there is very little light, we call that "black", but there is really no similar label for when there is a lot of it. It's just very bright. A very bright blue light is still blue, and could always be brighter. However, screens and print and analog film can not get arbitrarily bright. Instead, they have a limited maximum brightness, and the brightest color is always white. Which is why pushing things ever brighter in photo editing eventually turns them white, because that is the brightest thing possible on screen or print.

But this is a limitation of the display, not the editor or the recording. Thus it makes sense to edit photos with a physical understanding of light, where making things brighter does not make them whiter. Consider sharpening, or color grading, or contrast adjustments: All of these can push pixels brighter or darker, but really shouldn't suddenly whiten them at some arbitrary brightness. In the new scene-referred Darktable, colors are squeezed into a viewable or printable form only at the very end of the processing, retaining their physical properties for as long as possible. In most other tools, the squeezing is done early on, and edits hobbled thereafter.

In practice, this means Darktable can change tones with any number of tools without losing color information: the tone equalizer, or the rgb curve, or levels rgb, or the contrast equalizer, or sharpening, or color balance all manipulate tones without fear of losing highlight information. Only filmic rgb at the very end determines which bright pixels to turn white. In a way, the scene-referred part of the pipeline feels like raw editing, while the rest feels like editing a JPEG. Contrast this with Capture One, where only the sliders in the Exposure and High Dynamic Range modules can recover bright pixels from white, but e.g. Clarity, Levels, and Curve can not. After getting used to the scene-referred way of working, this seems like an arbitrary limitation, and really makes Levels and Curve less useful than they should be.

Darktable screen layout with tone equalizer and exposure module opened
Darktable's main development screen with the tone equalizer and exposure module

Interestingly, this leads to a somewhat different editing workflow in Darktable compared to Capture One. Since the scene-referred pipeline is mostly nondestructive, I can play with the overall tonality with abandon, safe in the knowledge that I will be able to recover any lost detail later. Thus I commonly use the exposure slider to expose for my mid-tones, and then reign in my burning highlights with the tone equalizer1 or filmc rgb white relative exposure2. In Capture One, digital overexposure is much more difficult to recover than underexposure, so I usually use the high dynamic range tools to underexpose slightly, and then recover mid-tone exposure with the high dynamic range shadows slider or tone curve. While these two workflows are largely equivalent, I find it interesting that Darktable can go either way, while Capture One can not.

A related area of rapid development in Darktable has been the color balance rgb module, and color calibration. Both of these allow for very fine-grained control over various aspects of color, handily parametrized by hue or tone or channel. There is some subtlety in the various controls for "colorfulness", "vibrance", "chroma", "brilliance", and "saturation" that took some learning. But it also taught me to see color in a more nuanced way, which is a good thing. In mixed lighting, I can now use several instances of color calibration modules3 for different image areas, producing more natural results that I am capable of in Capture One.

A particular pain point in Darktable 3.4 and Capture One was the rendering of colored highlights. In Darktable 3.4, bright colors often turned weirdly radiant, and then aggressively desaturated, with a sharp and unnatural edge in between. I don't quite know why, but this problem seems much reduced in 3.6. Both issues are now dealt with quickly and cleanly with a quick tug on the filmic rgb white relative exposure slider. A real boon that slider is. Now I just wish it didn't affect contrast quite so much and didn't also influence shadows, despite its name. Oh well, a man can dream. Capture One had the opposite problem, and sometimes wisted colored highlights into bright primary colors before desaturating, leading to cyan skies and yellow skin at the fringes of overexposure. This is still one of my major complaints about Capture One.

My last point of struggle with Darktable has long been the rendering of colors of my Fujifilm camera. Something about the colors always seemed off, in a way they weren't in Capture One. Greens too blue, browns too magenta, and skin too pink. After quite some experimentation, I found that this was simply due to Fujifilm depending on a bit of proprietary color magic for appropriate rendering. I now apply one of Stuart Sowerby's LUTs in chromaticity blending mode after filmic rgb. This leaves tonal adjustments fully to Darktable, but fixes the awkward colors perfectly. And it even allows me to play with the film simulations Fuji is famous for.

Darktable with lut 3D in chromaticity blend mode after filmic rgb
A LUT is applied in chromaticity blend mode after filmic rgb

Actually, this highlights a unique strength of Darktable: the order of modules in the image editing pipeline is customizable, and most modules can be applied with masks and blending modes. Only since I used Capture One for a few months did I understand how powerful this concept really is. The key difference is that "layers" in Capture One (et al) merely apply the image editing pipeline with different parameters in different places4, while Darktable's modules actually apply sequentially. This enables true incremental editing, such as fixing white balance correctly for one part of the image, then applying another level of white balance just for the blue parts. Or fixing colors in a LUT after all the tonal edits, without fear of destroying highlights or shadows in the process.

So overall, I am very happy with Darktable in its latest iteration. As always, there are plenty of nitpicks I could bring forward, such as being a bit sluggish in operation, and the UI could be streamlined a bit5. But it has once again displaced Capture One for my own work, and it makes me happy.

Footnotes:

1

similar to the high dynamic range module, but more fine-grained

2

essentially a tone curve adjustment

3

the new way of doing white balance

4

i.e. you can fully desaturate a picture to black-and-white, then "recover" saturation in a layer

5

the new quick access panel in 3.6 has already helped a lot!

Tags: photography

Camera Resolution

Some of my photographic lenses are reknown for their outstanding sharpness, others are said to be mediocre. But somehow I never quite saw a big difference in sharpness between them. Do older lenses have less resolution than the newer ones? Can a dedicated prime lens resolve more detail than a multi-purpose zoom lens? Let's find out.

These questions recently came to a head for me, in the choice between two compact cameras: A Fujifilm X100T, 16 MP, with a 23 mm lens, and a Ricoh GR III, 24 MP, with an 18 mm lens. In the last few months, I grew to like the X100T's 23 mm viewpoint, but did not enjoy how bulky the camera is in comparison to the GR III. So I wondered if I could replace the X100T with the Ricoh GR III and simply crop to 23 mm.

Being a scientist and all that, I set up an experiment: I printed out a resolution chart, set my cameras on a tripod, and took pictures with all my cameras such that the chart filled a similar portion of each image. Base ISO, two-second timer, processed in Capture One. White balance and contrast were equalized.

camera resolution comparison
Red bars indicate the point where some lines are no longer distinguishable, i.e. the limit of the system's resolution.

Each row in the above grid is one camera/lens/focal length combination, taken at the brightest aperture in the left column, and at the sharpest aperture in the right column. The text on the left shows photographic parameters, and the text on the right image parameters. The red line is the spot on the chart where some lines start to blend together, the limit of the image's resolution, as judged by my eyeballs. Depending on the exact framing, DPIs vary somewhat between the images, but generally at a factor of around √(̅2̅4̅/̅1̅6̅)̅ = 1.22 between the 24 MP images and the 16 MP images. The bottom row shows a synthetic image at simulated 24 MP / 380 DPI and 16 MP / 280 DPI.

From these images, the biggest contribution to resolution seems to be sensor megapixels. All 24 MP images resolve closer line pairs than any 16 MP image. This could also be a hint that lenses seem to generally outresolve their sensors.

Looking closely at the images, another hint becomes visible: aliasing is visible in all images. This is only possible if lens resolution is higher than sensor resolution.

Frankly, this is a rather big surprise to me. Reading discussions on the internet had predisposed me to believe that some lenses should be much higher resolution than others, and my big travel-zoom in particular should be terribly soft. From my measurements however, I see no evidence of this claim. All of these lenses outresolve their sensors, and differences between lenses is rather miniscule in general.

Another common wisdom is that lenses wide open are less sharp than stopped down. In this test sample, this seems to be true for the X100T's, and to a lesser extent for the LX100, and the Fujifilm XF 60. All other lenses do not show significant (!) differences between apertures. And even for the affected lenses, the difference in resolution is not dramatic.

However, resolution is not sharpness. While line pairs may remain resolvable, their edges do lose definition at wide-open apertures on the Fujifilm XF 18-135 at some focal lengths, and the X100T. These lenses definitely need at least additional sharpening at wider apertures. Contrast is also visibly different, but I chose to edit out contrast differences to keep the graphs legible. Generally, the Ricoh GRs' lenses (particularly the GR I) were by far the most contrasty of these lenses.

Another interesting comparison can be drawn between the synthetic images in the bottom row and the camera exposures above: None of the 24 MP images reach the resolution of the synthetic image, perhaps due to demosaicing artifacts. However, the 16 MP exposures are very close to the synthetic image. I don't yet know how to interpret this.

Speaking of demosaicing however, there does not seem to be a difference in resolution between the X-Trans Fujifilm sensors and the Bayer Ricoh/Panasonic sensors. I also checked whether a different raw developer would make a difference, but did not find anything noteworthy.

Overall, this experiment has taught me that all my lenses are high resolution, and that lens resolution in general is probably not a topic worth fussing over. Subpixel patterns are equally unimportant. Instead, lenses do differentiate themselves in contrast and sharpness, and camera sensors in megapixels. And finally, the tiny GR III in crop mode can indeed outresolve the X100T, and out-contrast and out-sharpness it as well.

Tags: photography

Books of 2020

Chickenhawk

chickenhawk cover

Most first hand accounts of wars I have read are written by a somewhat amateur author. Not this one. Robert Mason infused his terrifying tale with plenty of drama and humanity, and managed to write one of the best personal military history I have ever read.

Part of it is surely the subject matter, as I am a huge fan of aviation and its lore. And part of it is that I haven't yet read a lot about the Vietnam War.

But either way, I could not put this book down. Every mission seems to be more dramatic than the last, and you can see Mason's flying and survival skills just barely keeping up with the challenge. You can also viscerally feel the surviver bias at work, with plenty of close calls and dead friends. It's terrifying and thrilling at the same time.

And towards the end, the book truly surprised me with a very frank account of PTSD and the life of a veteran, which I hadn't read anywhere else in this level of clarity.

I cannot recommend this book enough. Perhaps a bit depressing at times, but an utterly compelling story, and without a doubt one of the best books I have ever read.

Mountain Light

mountain light cover

Quite simply, the best photography book I have ever read. From the era of color film, that short time between the black-and-white darkroom and photoshop, where pictures really couldn't be edited. It all came down to the skill of the photographer, which makes a book from this era much more useful than its darkroom-edited predecessors or photoshopped successors.

Besides that, the story of an avid adventurer who took part in several expeditions to climb the most difficult mountains in the world for the first time, and document life in the most remote places on earth before man could ever touch them.

All in all, a fantastic book of outdoorsmanship, adventures, and photography, three things I love dearly.

Two Years Before the Mast

two years before the mast cover

The story of the young author joining a trader ship to the US west coast, in 1834, when it was still largely uninhabited and wilderness. They traveled to many a well-known place, such as San Francisco or Santa Barbara, but these places were still merely trading outposts and missions.

At the same time, the author chronicles a sailor's life on board the last generation of sailing vessels, just before they were replaced by steamers. He joined the ship on a two-year-or-so contract around the American continent, on an entirely self-sufficient ship with only the occasional sighting of a fellow ship or the lighting of cargo in a trading depot. Which is an utterly fascinating historical perspective considering that the story happens more-or-less at the same time as the comparatively modern-seeming Sherlock Holmes.

But a true eye-opener happens at the very end, where the author re-visits San Francisco later in his life. The city by then has grown to a bustling trade town with regular ferry routes and rail connections, churches and pubs, and all the amenities of modern life. It is hard to believe how such industry would grow from these humble beginnings in just a few decades.

Bobiverse: We Are Legion / For We Are Many / All These Worlds

we are legion cover

Like good candy, I just couldn't stop reading. The story of a regular guy, uploaded into a colony ship as an AI, roaming the galaxy, and exploring the boundaries of what it means to be human. Which sounds much more philosophical than it actually is. The book is written in a lighthearted and fast-paced style that I could hardly put down. I read the first book in one week, and immediately followed through with book two and three.

And it actually comes to a satisfying end in the third book, instead of methastesizing into a franchise, which I can't praise highly enough. Perhaps a little bit superficial and handwavy at times, but a thoroughly good time for a science fiction fan.

Tags: books

A Review of the RAW Photo Editor Capture One

About half a year ago, my second child was born, and all that precious free time I had used for photo editing evaporated. So I started looking for faster photo editor, as a temporary replacement for Darktable. Based on my research last time, I chose Capture One. During the last six months, I got to know the program well, and want to share my thoughts on it, and how it compares to Darktable.

Let's start with a bit of background about my photography, to put my views into context. I learned photo editing on Linux, with Darktable, mostly for documenting my travels and family. I gather that a photographer who does not know Lightroom is a bit of an oddity. But it should provide an interesting perspective for the purposes of this review. Furthermore, my interest in post processing is not purely artistic. Being a signal processing scientist, I can't help but analyze the underlying image processing algorithms, and am perhaps more annoyed by certain editing artifacts than most people.

In terms of photography, I tend towards quick reactionary shooting of the fleeting moments of my family's life. Mostly because I prefer to spend my time playing with my kids instead of pointing a camera at them. I am therefore perhaps a bit more tolerant of slight imperfections in terms of focus and exposure than is usual. This also means I value my camera/editor's capability for recovering misjudged exposures or imperfect lighting very highly.

Most review articles of photo editing software are conducted from only a short glance at the contesting programs, since spending the months necessary to get to know each program is usually neither practical nor affordable. But due to my particular circumstances, the present text actually comes from a few months of more-or-less exclusive use of Capture One, on about 30 thousand pictures, of which I edited about 2000 (all new, not just re-edits of old ones, and implicitly the incentive to recreate another program’s rendering).

Capture One screen layout
Capture One's main develop screen

So, Capture One. It is very fast. You move a slider, you see the results of that change immediately, even on my lowly Surface 7 Pro tablet. But this focus on speed actually goes deeper than just slider movements: I discovered that many parts of the application seemed actively optimized for quick operation, which allows me to process pictures much more quickly. This was frankly a revelation in the time-starved months after my second baby’s birth.

As an example of how much quicker Capture One's default workflow is than Darktable's, here's my usual editing flow in Darktable: I start with fixing exposure, in the Exposure module. Then I adjust shadow density and recover highlight color with the black and white sliders in the Filmic RGB module. Then I adjust white balance to taste in the Color Calibration module. Then I crop in Crop and Rotate, then add additional adjustments such as Color Zones, Color Balance, Tone Equalizer, Denoise (Profiled), or Contrast Equalizer. Last come local adjustments if necessary. The thing is, more or less every image needs at least Exposure adjustments, some Filmic RGB tweaking, and cropping. All of these items are situated in different modules in Darktable, each requiring multiple clicks to access the module and then change the value. This is supposed to be improved in the next version of Darktable, which will let you put all your favorite sliders in a custom module. I am very much looking forward to that.

Darktable color calibration Darktable crop and rotate Darktable filmic RGB Darktable tone equalizer
Darktable's basic controls

In contrast, a similar workflow in Capture One happens entirely on one screen, just by going from one slider to the next:

Capture One basic controls
Capture One's basic controls

Furthermore, common tools such as cropping, rotating, and the white balance picker are accessible at all times with highly memorable keyboard shortcuts (C, R, W, respectively). Taken as a whole, this allows me to positively blaze through images compared to my Darktable workflow. It took me a while to appreciate how significant this difference was in my use: I now sometimes do a few days’ edits in a spare half hour, which used to be an all-evening affair in my usual Darktable workflow.

Which is not entirely Darktable's fault. Assigning custom keyboard shortcuts in Darktable can speed things up, and of course Darktable’s module system is infinitely more powerful, so there's a reason for its complexity. But the above example highlights a bit of a philosophical difference between Darktable’s unflinching priority on user control, and Capture One’s compromise between power and speed. There are upsides and downsides to both, but at this moment in my life, I begrudgingly value speed over power.

However, there were also a number of occasions where I missed Darktable’s deep control. Most notably, Capture One’s High Dynamic Range sliders and Clarity controls feel a bit restrictive and oversimplified: It can be hard to control which parts of the image are affected, and the tools are prone to produce artifacts such as halos if not managed carefully. In contrast, Darktable's continuous Tone Equalizer mask gives very precise and adjustable control over the affected area. Similarly, Darktable's Contrast Equalizer can control local contrast at arbitrary wavelet sizes, not just the very small ("Structure") or very big ("Clarity") ones, for example for specifically highlighting tree trunks or bird feathers.

Darktable develop screen with contrast equalizer
Highlighting feather detail with darktable's contrast equalizer

Another annoyance with Capture One can be found in colors at the edges of the tone curve: When pushing exposure, Capture One tends to turn brightly colored highlight towards primary colors such as cyan/magenta/yellow before desaturating them into white. Similarly, shadow recovery sometimes pushes brightness and saturation a bit more strongly for brightly colored shadows than for dull ones, leading to an unnatural glowy effect. These cases are usually easy enough to fix with a quick tug on the shadows/highlights slider or by reigning in the offending color's lightness in the Color Editor, but they smell a bit of a runaway algorithm, which bothers me. Although in direct comparison they bother me less than the various rocks and hard places in Darktable's Filmic RGB chrominance preservation modes.

On the topic of bothersome details, Capture One's repair layer seemingly has a mind of its own when deciding whether to create a new control point, or adding to an existing one. The Highlight and White slider in the High Dynamic Range module sometimes can't quite decide whether to move white point, and sometimes lead to lightness reversals around bright objects. More on the algorithmic side, it annoys me that only the Exposure and High Dynamic Range modules have access to the full dynamic range of the raw file, while all the other tools apparently come after the screen transform, and therefore can't reach beyond the black point and white point. This makes Levels and Curve a bit less useful than I'm used to. Demosaicing is also not quite as detailed as Darktable's, although only by a tiny margin.

One surprising limitation of Capture One is that its support for cameras and lenses is a bit hit-and-miss. Film simulations are only supported for my Fuji X-T2, but not for the older Fuji X100T, nor the Ricohs or Panasonic. Some of their lens profiles are also laughably bad, and leave obvious reverse-vignetting or barrel distortion when enabled. Frankly, Darktable does better in terms of lens support, despite relying solely on volunteer support.

In terms of UI, Capture One is generally well-organized and easily configurable. But there is a constant stumbling block in the Layers module that I find very annoying: many of Capture One's tools automatically create new layers, but deselecting the tool does not deselect the layer. As delightfully easy it is to press B for the brush and paint in some Clarity, as unnecessarily laborious it is to then spend three clicks to slide open the Layers module and select the background layer again to resume editing. Or alternatively, wonder why your edits don't work while you don't have the background layer selected. It's a true pain.

Capture One layers
With a layer selected, edits only affect the layer

As the last point of the nit picking, I was disappointed by the number of minor technical bugs I encountered in my use of Capture One: Half the time, Capture One starts half zoomed-in, and leaves me scratching my head at what weird compositions I chose for a few seconds. And the main, zoomed-out view is weirdly blurry if you use two displays with different DPI. Both of these issues are well-documented on their forums for several releases, but have not been fixed. For the exorbitant price that Capture One commands, such issues and customer communication are frankly unaccetable.

Finally, a few words about the library module and file organization. At first glance, I hated Capture One’s library. You have to import every single directory manually (I organize my images in daily directories), all the edits go into a central catalogue and nowhere else, and the sidecar files contain no editing information. But then someone told me a much better way: Instead of using Capture One’s Catalogue, create a Session, but ignore all those pre-built input and output directories, as well as the import button, and instead simply navigate to any old directory on your computer with the sidebar file browser. This is clearly not how sessions are meant to be used. But it actually works reasonably well, and puts editing information in a subdirectory next to the raw file.

On the whole, I grew to quite like Capture One, despite its flaws, mostly for its streamlined user interface and speed of operation. In terms of image quality, I honestly didn’t see much difference between Darktable and Capture One. But perhaps I am not the most discerning of users, either, as my focus is not on crazy detail recovery or the more technical arts of macro or astro. When it comes to control, I occasionally felt restricted and, dare I say, patronized by Capture One. But the speed of operation and general good-enough-ness of the image quality are still hard to argue with. I just wish they fixed their UI bugs, and improved their algorithms a bit.