What makes a cinema camera?
Comparing the Sony FX3 vs Fujifilm’s X-T4.
Recently, I borrowed a friend’s Sony FX3 in preparation for an upcoming production, and I naturally decided to run the FX3 through some tests to get acquainted with its: ergonomics, menu, and color science because you never take untested gear into the field. During these tests, I brought along my FujiFilm X-T4 and X-T3 to see how they stack up against a full hybrid cinema camera.
My previous experience with Sony:
While I am no stranger to Sony’s cinema line, I used to work with the original Cinealta back in the middle 2000s; Sony’s Mirrorless line was something I avoided when possible. I never liked the look I got from them, particularly the Sony A7s MkII; the skin tones were too red, and it always felt like it just took more work to get a pleasing result from them than other cameras. I am more open to working with Sony’s cinema cameras as they have always given me the impression of giving off a clean and neutral image. Nothing significant to get excited about, but nothing terrible either. And spec-wise, their cameras have always felt competitive compared to Canon and Nikon, particularly in Frame-rate, low light performance, autofocus, weight, and battery life. Their FX line of cinema cameras looks particularly appealing for documentary-style work. Out of the gate, their colors and skin tones just don’t seem to pop out as quickly as an image produced by a Canon or Red Cinema camera. That is, however, just my opinion, and there are many people out there who love them that I respect, like Philip Bloom. And a neutral image may not always be a problem; many videos get shot in a Log profile that will need adjustment in post anyway. It means you potentially have more control (and work) to make a look or style that you want rather than having one baked in for you that can’t be changed, in theory at least.
Now admittedly, comparing a $1,700 X-T4 hybrid mirrorless camera (or an $800 X-T3/X-T30) with an APSC sensor versus a nearly $4,000 FX3 budget cinema camera that boasts a full-frame sensor with dual gain should not be a fair fight. Fujifilm has no cinema line of cameras and has never made one yet. On the other hand, Sony has been making cinema cameras for decades and has multiple evolutions of their Log profiles. Fujifilm’s F-log2 (at the time of this writing) still isn’t supported in DaVinci Resolve.
Setting up the test:
For my testing, I did a latitude test to look for how the cameras handled overexposure and underexposure. Specifically, to find where the skin tones look the best and see where the image breaks apart.
For the over-exposure test, each camera started recording at average exposure, slowing the shutter speed and thus increasing exposure one stop at a time and recording the change up to six stops of over-exposure. The Overexposure test is specifically concerned with where the skin tones clip to pure white, where the color information and tonality of the pixels are lost.
The underexposure test used a slower starting shutter speed and stepped up from there to darken the scene up to six stops of underexposure. At that point, the image was virtually black. The footage was then taken into DaVinci Resolve, graded back to normal at each interval, and compared against the others. The Underexposure test looks for four things: firstly, what is the quality, texture, and tolerability of the camera’s noise, when does the noise start to eat away at the underexposed details, at what point does the noise get so problematic that it is uncorrectable through noise reduction, and finally, when do macro blocking compression artifacts appear.
You would be surprised at how far an editor can pull an image back from the darkness in post.
The decision to change the shutter over the aperture for the exposure test was to remove the potential of vignetting out of the equation, which all affordable lenses seem to share when wide open to some degree. Since the FX3 and X-T4 were not using the same lens, the difference could partially skew the results. Manipulating the shutter speed felt like the safer option.
With the Fuji X-T4, I decided to test 2 different modes on the camera, F-Log (at ISO640) and Hybrid Log Gamma (HLG) (at ISO1000), to see how the two modes stacked up against each other against the Sony using S-Log3 Cine (at ISO640). To facilitate the test, I set both cameras at their native ISOs. I had them record a simple scene of a person sitting in front of the camera while adjusting the shutter to simulate over-exposure and under-exposure. A second tier of testing ran simultaneously using an Atomos Ninja V. Using the Atomos, I lit each scene for the peak of the skin tones to just touch 60% in the waveform.
For more information on how to do a latitude test, CineD.com is an excellent resource for seeing how they test cameras. They have tried a lot of cameras from all of the major manufacturers. Yet, there is still value in doing your testing, particularly before investing in a camera yourself or taking a new piece of kit into the field.
I set both cameras to record in 4k at 24 frames per second at the highest internal quality possible. For Fuji’s X-T4, that would be recording in H.265, with the settings at DR400 and All-Intra. All in-camera settings like Color, Sharpness, and Noise reduction were reduced to their lowest settings. Internally the X-T4 camera can record up to 400mbps, with 10-bit color in all, be it a lackluster 4:2:0 chroma subsampling.
I also recorded a ProresHQ 422 recording using an Atomos Ninja V to see how the higher recording quality differed from the internal recording. I don’t own an external recorder, so this was also a chance to see if the presumed boost in image quality would be worth the extra expense, the cost of the device, and the extra storage needed.
For Sony’s FX3, the Camera it’s best internal coded is the XAVC S-I 4K, with all color, noise reductions, and sharpness settings reduced to their minimum. The X-AVC S-I codec records its 4k, 24p video at 240mbps with a superior 4:2:2 chroma subsampling. The Atomos was also used to record a ProresRaw signal from the FX3 to see how it performed.
Where the Sony FX3 won:
In many regards, the FX3 is the better camera. FX3 shooters are unencumbered by the 30-minute recording limits that X-T4 shooters must be aware of when documenting longer events. Most notably, the Sony is also capable of 4k 120p, which blasts past the X-T4’s top 4k frame rate of 29.97fps. Also, the FX3 is a full-frame camera with a similar megapixel count to the X-T4’s smaller APS-C-sized sensor. So, in theory, the photo sites should be a good deal larger and more light gathering compared to the Fuji, meaning that the Sony camera’s sensor should need less gain (or ISO) to peer into the same darkness. Less Gain/ISO means less noise. Thus, the FX3 should win in a low-light contest, which, to be clear, is different than how it handles underexposure. The FX3 is also a dual gain camera, meaning that at around ISO 12,800, the FX3 switches circuits and sees a reduction of noise, giving it a less noisy image at the higher ISO values.
Interestingly, the FX3 was at least 1.5 stops brighter than the X-T4 at the same ISO, Shutter, and Aperture. But there could be two explanations for this: firstly, the aperture size, despite being the same setting, is still relative to the sensor size. An aperture of F2 on an APS-C sensor may not let in as much light as an F2 on a full frame. Secondly, the two camera manufacturers may have different standards for assessing sensitivity; admittedly, 1.5 stops is a lot of light, so it’s probably a combination of the two.
Sony’s most significant advantage over Fujifilm cameras is its E-mount which is vastly better supported by 3rd party lens manufacturers today. Ever since Canon killed off their EF-Mount in favor of RF, it’s safe to declare Sony as the king of 3rd party lens support. Fuji recently opened up its X-mount to 3rd party manufacturers to develop lenses that can work with its autofocus system. Thus, the list of video-centric lenses is small and too close in cost to their full-frame counterparts to my liking.
In my opinion, Fujifilm’s stills lenses are the real disappointment here, coming from a company known for making high-end broadcast and cinema glass. Most Fujifilm/Fujinon still lens zooms don’t seamlessly step from one stop to the next, causing jarring exposure shifts while zooming. Their stills lens also generally seem to shuffle from one focus point to the next while manually shifting focus in a very stilted manner that is very distracting, making manual focus pulls a nightmare. Lens breathing on Fujinon glass is aggravatingly excessive. A lens that steps from one focus point or aperture stop to the next abruptly is forgivable in a photography environment, but in the video, these issues create severe roadblocks to creative expression. Most Fuji lenses also suffer from a relatively low aperture blade count (about seven blades), which means when stopping down, the bokeh quickly turns from smooth circles into jagged polygons. For example, Fuji’s 50mm f2 lens has far smoother bokeh at f2 than Fuji’s 50mm f1.0 lens when it is set to f2 due to its low aperture blade count. And worst of all, Fujifilm glass is generally the most expensive glass one can buy on the X-mount. To be competitive; Fuji and 3rd party lens makers need to make more video-friendly lenses with smooth de-clicked apertures and fast yet buttery smooth focus shifting that doesn’t breathe to appeal to content makers.
Personally, the X-T4 also suffers from bad pre-amps in its audio recording, and even with an external recorder feeding in a clean signal, the pre-amps are only good for scratch audio. The FX3 has a clear advantage with its top handle accessory that integrates a better pre-amp with XLR inputs. However, I did not test for it, as I would almost always prefer to use an external recorder and just feed in audio through its line out, but for run and gun style shooting, The FX3 has a leg up in audio if it has that top handle.
Sony also won the autofocus contest, which, to be honest, is no surprise; their cameras have been the best or among the best for autofocus for some time now. For my testing, the FX3 used a Rokinon 45mm f1.8 lens and maintained perfect autofocus the whole time. The Fuji was using a Fujifilm 35mm XC lens, and during one of the under-exposure tests, where my hand was moving in front of my face a lot, it lost focus about -3 stops down and never recovered. The Fuji 35mm XC lens also displayed significant focus breathing. Combined with the X-T4’s tendency to focus on a relatively static target, the breathing made the focus hunting very noticeable.
Where the Fujifilm X-T4 Won:
When comparing the $4,000 FX3 vs. a $1,500 X-T4, if the two cameras strike even, then that is a net win for the X-T4, in my opinion.
In the Over Exposure test: The X-T4 either won or tied the FX3.
In F-Log, the X-T4 clipped at about +4.6 stops overexposed and had the smoothest highlight roll off of all three. In HLG mode, the X-T4 tied the FX3 in the highlights, clipping at +4.3 stops of over-exposure. However, there were differences in the highlight roll-off. The skin tones for the X-T4 in HLG mode were generally a little more patchy and not as smooth as the S-Log3-cine, but the washout was more aggressive for the Sony camera. And granted, when hitting the clipping point, S-Log3 did look better due to its smoother roll-off; the X-T4 in HLG still had less surface area clipped. The point remains if you don’t clip your highlights on the skin tone, all three will look good.
Matching the cameras is also where I became impressed with their color sciences, but for different reasons. For one having the Fuji match and or beat the Sony in latitude and roll-off despite a smaller sensor is very impressive, albeit it was a narrow victory. I was also impressed with the FX3 for how improved the color science was over the previous generations. The skin tones are more yellow than the Fuji and needed some tweaking with color curves, but it was reasonably straightforward to get the two cameras to match together.
While I generally love Fuji’s color science, I have to concede that it is not perfect. F-Log (1) has this annoying quirk in that the redder parts of a person’s face tend to look more magenta than red, in other words having too much blue. It’s a simple fix using Resolve’s Hue vs. Hue Curve, but it took some trial and error to figure out the right cocktail.
Where the X-T4 won big over the FX3 was in the under-exposure test.
When grading the under-exposed footage back to normal, the FX3 got noisy and fast. At -2 stops, the FX3 started showing nasty Chroma noise, and at -3 stops, some really ugly macro blocking compression artifacts were displayed on flat textureless surfaces. The noise pattern completely eats away at the image detail at -3 stops making the image a mushy artifact-riddled mess. In my opinion, at -3 to -4 stops, the image is wholly broken depending on your tolerance.
When recovering from underexposure, the FX3 lost detail in the image appallingly fast due to noise and compression artifacts. If you look into the shadows on the FX3 footage (on the right side) at -2 stops, a weird red noise is beginning to wash over the skin tones. If the internet compression is too severe to notice in a still frame, below is an example of that same noise when pulled back at -6 stops.
All cameras displayed some form of macro blocking when graded back from -3 stops. F-Log seemed to display the least amount of artifacting at this level, HLG was a little worse, and for the FX3, the compression is eating away at the noise itself. Meaning the noise is a blob-like mess.
Depending on the particulars of your scene, and your tolerance, the X-T4 camera can handle -4 to -5 stops of underexposure and still be salvaged. In my opinion, chroma noise didn’t become a problem for the Fujifilm camera until -3 stops, which is a full stop better than the FX3. And while the macro blocking did appear at -3 stops on the Fuji, it was less severe than on the Sony. Of course, everything depends on context; images with more textured surfaces should be less prone to showing such compression artifacts; as always, test the limits yourself and find your preference.
The noise pattern on the X-T4 is vastly smaller and preferable to the FX3, which is more blocky and less random, resembling the noise effect in Adobe After Effects.
I must stress that the quality of Fujifilm’s X-T4’s noise was vastly more tolerable than Sony’s FX3's.
So how did the FX3, a cinema camera worth over 2x the value of the X-T4, fail so badly in the under-exposure test when it has the advantage of internally recording a 4:2:2 chroma subsample which is 25% better than the X-T4’s 4:2:0 chroma subsample? The best answer available is the bit rate. The FX3 records a paltry 240mbps in 4k 24p mode, which cannot be increased despite the camera and its CFexpress type-A cards being capable of higher data rates (but only at correspondingly higher frame rates). The X-T4’s internal recording caps at 400mbps. Simply put, at 24fps, The Fuji X-T4 has more information in the contrast values, as evidenced by the noise pattern.
The value of the Atomos Ninja V.
Initially, I was not impressed with the Atomos Ninja V as a recorder; on the over-exposure test, the Prores recording did not recover highlight information any better, and the video signal was always a touch noisier than the internal recording. The main advantage the Prores file had over the internal files in the over-exposure test is that color saturation can be pushed harder before the color values in the skin tones start to bleed out, particularly the red values, which made the skin tones slightly more even looking. Granted, this is a small gain for a piece of tech that will set you back at least $1,000 for the unit, media, batteries, cables, camera mounts, and eventually storage.
Where the Atomos proves its worth is when recovering shadows. And while yes the signal is a touch more noisy than the internal recording. The real value is in the macro blocking. The Prores422HQ staved off the macro blocking in both F-Log and HLG for at least for an extra stop. Meaning the image could be saved from nearly -5 stops of under exposure depending on your tolerance.
For the FX3, the test ProresRAW test was more problematic. While I am no stranger to handling RAW files from Red cameras, this was my first time using Apple’s Raw codec, and Premiere Pro did not translate it very well; the gamma shifted down three stops, and the hue shifted around horribly. Correcting these shifts created a very strong and fully saturated noise pattern on the red channel that increased significantly faster than anything else during grading.
Based on this experience, I would say that unless you were editing in FCPX, don’t use ProresRAW.
The ProresRAW recording confirmed one thing for the Sony FX3: the noise pattern was much finer than the internal recording, proving that the internal compression is eating away at the image.
The Sony FX3’s noise and compression issues are troublesome enough that even an $800 X-T3 holds up better in color grading than a $4,000 Hybrid cinema camera; the problem is even worse when you start to consider that the internal codec and data rates on the FX3 are the same as on Sony’s FX6 which costs nearly $6,000, and the Sony FX9 which costs $11,000, which means that the images will fall apart equally as fast on those vastly more expensive cameras.
CONCLUSION: F-Log vs. HLG vs. S-Log3-Cine
My takeaways from the three cameras are as follows. Generally speaking, HLG was the easiest to grade; in all honesty, most of my time grading clips was spent trying to get the F-Log and S-Log3 footage to look as good as the HLG clips. When clipping did occur on the skin tones, the HLG footage looked the worst. The HLG footage displayed macro blocking artifacts more prominently over the F-Log footage, adding another asterisk to its record. The HLG clips also tended to bleed color in the red channel faster than any other mode.
On the other hand, F-Log needed more work to get a usable result but had the best highlight retention, roll-off, and least macro blocking artifacts. It did require the most amount of saturation to be pushed on it, even compared to S-Log3. In all my testing, parts of the skin tone that should be red tend to display as magenta, which is correctible but takes some work using the Hue vs. Hue curves in Resolve or Premiere Pro.
My main problems with the FX3 are its recording quality, not its color science. S-Log3-Cine is a pleasing color space, and I had a decent enough highlight protection and roll-off. It is possible to get a beautiful image from the camera, and many have. The skin tones are a little too yellow out the gate, but this is also easily corrected in Resolve or Premiere Pro using the Hue vs. Hue curves. The internal codec is as such that it cannot survive any serious corrections from under-exposure. The camera is fine if you expose your subject correctly and don’t push too hard on the color grade. In other words, always be perfect and never make a mistake.
Generally speaking, all three modes give the user about +4 stops of latitude on the skin tones. As long as the (Caucasian) skin tones do not clip, they can be corrected back to the 55–70% range without much penalty in their look or saturation. The look has more punch in the contrast when skin tones are graded from 85% back down to 65%. It is also worth noting that each camera clips at a different point: in Fujifilm’s case, keep the skin tones under 100% or 1023-IRE. In Sony’s case, keep the skin tones under 90% or 896-IRE as the Sony FX3 clips off at the last 10% on either end of the spectrum.
For under exposure, Sony recommends not letting the skin tones fall below 42% or roughly 420-IRE; going below this is not recommended due to Sony’s poor internal compression. Fujifilm, however, can be more forgiving when it comes to under-exposure. My rock bottom would be 30% or roughly 256-IRE, but I wouldn’t do that on a shoot unless I had tested for it far more extensively. At -3 to -4 stops or 30% exposure on the skin tones, flat surfaces may be prone to showing macro blocking without the aid of an external recorder.
Sony’s documentation offers a helpful guide for exposure that even Fujifilm users can adopt. Sony recommends that on scenes with brighter backgrounds (like a sky during the daytime), their users opt to expose the skin tones at 42% to give more latitude for protecting highlights within the image. While on the other end of the spectrum, if the scene is dark, they recommend filming the skin tones at 55% to allow for more detail in the shadows. Still, given my tests, you could go safely up to 70% or even say 85% if the person is the brightest object in the frame to preserve details in the shadows. I think it is almost always better to add contrast in a post than needing to reduce it.
Is an external recorder worth it?
The camera that benefits the most from the quality boost of an Atomos Ninja V or Black Magic Design Video Assist 12g is hands down the Sony FX3; the latitude boost the camera gains in the shadows puts it more on par with the Fujifilm X-T4. The Sony FX6 and FX9 also need an external recorder to boot. The Sony cameras are too far-hobbled by their internal recording compression to survive any heavy stress from a color grade, particularly underexposure. The FX3 is more forgiving in retaining the highlights (+4) than it is in the shadows (-2). With Atomos, you could probably get to -3 stops or maybe even -4 stops of underexposure, depending on your preference and noise reduction software of choice.
Upon writing this test for the FX3, I checked CineD.com for a cross reference, and while they did not test the FX3 directly, CineD did test the A7s3, which shares most of the same specs as its cinema sibling, and their conclusions about the codec go to the same place. In my opinion, Sony’s X-AVCS-I codec is trash.
“In the underexposure test, the Sony a7S III is hindered by strange, large patches of chroma noise artifacts — if you don’t want to have those, the limit is reached at 3 stops of underexposure — rather similar to the Sony FX9 if you use the internal All-I codec.”
The Fujifilm fairs better with its internal recording quality. It gives the user about +4.3 stops to -3 stops of latitude, with the Atomos giving another usable stop or two in the low lights. The benefits, while nice, are more optional depending on your needs.
My opinion on the Atomos is mixed; adding an external recorder to a hybrid mirrorless rig is cumbersome: I don’t particularly like that the cables can be unplugged easily. But as mirrorless cameras continue to close the image quality gap with their cinema cousins, I think the time of the Atomos and BMD Video assist are nearing their end. We are now seeing mirrorless cameras with minimal rolling shutter, 13+ stops of latitude, 6k to 8k video, high 4k frame rates, and Prores and raw internal recording. An external recorder is not needed with cameras like these. The $1,000+ expense of an external monitor can be saved by getting a cheaper $200+ external monitor with good features like false color, waveforms, LUTs, and focus peaking.
If you are interested in buying a mirrorless camera to shoot video with, opt for a camera that records Prores422 and/or RAW internally. Doing so will guarantee you the best possible dynamic range and signal-to-noise ratio. External monitors are far cheaper and have fewer risks of losing the recording due to a cable disconnect.
Cameras like the Panasonic S1, S1H, GH6, Fujifilm X-H2s, Nikon Z9, Canon R5c, and the Black Magic Cinema camera (plus others) are all strong contenders.
We live in a fortunate age where there is no such thing as a bad camera right now where getting a good image depends more on the artist’s skill in working within the device’s limitations than anything else. The FX3 still has excellent autofocus, battery life, low light sensitivity, and more. Sony’s choice of internal codec for their mirrorless line is by far their weakest link. I hope that changes in the near future.
