Before heading for a shooting trip to Croatia for a music video (to be released later this year), I decided to take both the preproduction KineRAW Mini and a 5D Mark III (Magic Lantern firmware for raw recording) with me to do some comparisons.
The wonderful and equally talented Nicola von Leffern (a very talented camerawoman in her own right – seriously do check out her videos and showreel by clicking here) agreed to be my camera test victim. Thanks Nicola!!


My friend Johnnie Behiri has already shot a proper Video Review with the KineRAW Mini (a beautiful documentary about a sailing team training in Southampton – check it out here if you have missed it!).
Usually I am not a big fan of technical tests of cameras, but this camera had me intrigued – for a surprisingly low price of €3800 (including tax), the KineRAW Mini by Kinefinity is capable of recording 12 bit RAW video. And that puts it right next to the 5D Mark III, which is able to do the same in 14 bit with the ever-evolving Magic-Lantern hack, which we have covered extensively on cinema5D (click here to read the complete and constantly updated guide to 5D Mark III RAW recording, thanks to our very own Sebastian Wöber.)

So the KineRAW Mini and 5D Mark III are priced very similarly ($3,500 for the 5D Mark III), and both can record raw video – that really called for a comparison.


More dynamic range!

The advantage of raw video is the fact that it opens up a multitude of post production options that are impossible with heavily compressed footage (which we usually get from DSLR and even many other more professional video cameras). That comes with a price, of course – the file sizes are enormous and hard to deal with on every job day in and day out. Nevertheless, there are a lot of applications where raw is a real asset even for everyday jobs – for example, in occasions which require a lot of latitude.

Let’s say you have a talent moving in front of very strong backlight. You would have to light the foreground considerably (one way or another – with a reflector or simply a very powerful light) in order to be able to see the subject in-camera and at the same time exposing correctly for the background. If that backlight is sunset, it becomes even more difficult – the light is changing rapidly, a reflector often looks unnatural and will blind the subject.

With raw video, you are able to lift the shadows and regain detail in blown-out highlights. A scene like this is probably as bad as it can get with regards to dynamic range. Extremely bright and extremely dark spots in the same image are a real hard challenge for any sensor.

For this test, I decided not to use ND filters. Why?

Two reasons: for one, I did not have the same brand of ND filter twice with me – which would have meant that inconsistencies in shooting could be a result from different qualities of filter glass. I did not want to risk that.
Secondly, I wanted to achieve deep focus, which also meant I needed to step down considerably. Shallow depth of field would have meant that also the highlights in the background would have behaved differently – if they were out of focus too much, this would have resulted in highlights that are much harder to reconstruct (and judge at all, because after all, they are out of focus).

The decision not to use ND filters of course also meant that I had to use a very low ISO. The lowest ISO on the KineRAW MINI is ISO 320. I know that both the KineRAW MINI and the 5D Mark III are rated at a higher native ISO (around 800 ISO depending on the camera), which means that generally they have their ideal dynamic range in this setting. It was a compromise I was willing to make – but keep in mind that dynamic range might even be BETTER than shown in the video (in both cameras) when shooting at ISO 800.

photo 2

The setup & workflow

The setup was simple: I would stop down from f/9 to f/22 stop by stop while Nicola would move back and forth in front of both cameras. Thereby she would block the sun from directly hitting the cameras at some point, which means we would be able to see both the unblocked sunset and her silhouette blocking the sun in the same shot.

Only the first f/9 shot is exposed more or less for the foreground (Nicola), while all the remaining shots expose more and more for the very bright background, leaving nothing but a silhouette of Nicola.

I would quite literally not be able to see her any more on any camera. As you can see in the recovered shots, it’s unbelievable how much detail can be recovered through the raw footage on both cameras.

All the highlight and shadow recovery was done through Adobe Photoshop – I simply opened all the DNG files of a sequence, then corrected one photo with the RAW Import Converter (“Highlight Recovery” turned to the maximum on all shots, “Shadow Recovery” increased with higher aperture). The setting for one photo was then synchronized for the whole sequence. The photos were then saved as uncompressed TIFFs and imported as an image sequence in QuickTime 7, where I saved the sequence as an uncompressed MOV file. The MOVs were imported into Final Cut Pro X and dropped into a ProRes 422 timeline. Minor saturation adjustments were done to the KineRAW MINI footage as the KineLOG produced considerably less saturated colours than the 5D Mark III .

EDIT: Readers have correctly pointed out that you can’t “bake in” picture profiles into raw footage – which means that the CineStyle setting in Magic Lantern is irrelevant. Sorry for my stupid newbee mistake!


As you can probably see, both cameras have their advantages when it comes to dynamic range, and it’s a really close call. Due to the considerably larger sensor, the 5D Mark III seems to be better in low light. The shadows are simply better protected and there is more detail. When completely underexposed and pulled back up (e.g. at the f/22 shot), grain is very visible from both cameras, while the 5D Mark III seems to be a tiny bit better. However, it seems like highlights are more easily recoverable with the KineRAW MINI in KineLOG mode. Look at the horizon in the shots and you will see much more detail in the blue.

So all in all, both cameras have impressive dynamic range, yet the 5D Mark III seems to be better in low light and the KineRAW MINI is better with highlights. Both are definitely worth quecking out in more detail!


3800€ (inc. tax) available here:
(shipping to US is 89€)

The 5D Mark III is available through B&H:

Recommended Retailers:

US: $ 3,399


Watch it on Vimeo

Silton Buendia September 13, 2013

On the wider open shots the KineRaw looks much more under exposed or shall I say the sensor doesn’t seem to have the same sensitivity as the 5D looks 2-3 stops off. We see more lens flare with the smaller sensor to as well as the star effect. Its really hard to get the star with the larger sensors because of how shallow the footage is. Also the I’m not liking how the KineRaw makes the skintones brownish black especially when underexposed. Reference the F11 shot especially

Rush September 13, 2013

5D Mark III RAW Video ISO320 is ISO400 for real. In RAW video intermediate ISOs don’t work!

Picture Style don’t do anything to RAWs. It affects only image on camera’s screen.

ISO100 is 0.3 EV better in Dynamic Range compared to ISO400, according to DxOMark.

Nino Leitner Reply
Nino Leitner September 17, 2013

Thanks Rush – good to know, I wasn’t aware of that. It’s not too far off though.

Ahmad Nassar Reply
Ahmad Nassar September 13, 2013

No one can stand against 5D M3.

Archie September 13, 2013

Thank U for a test on Croatian coast(me half Croate)
I think people are making mistake by matching the cameras by equal settings. It should be tested by trying to get equal(perfect)exposed image, specially because of different sensors. Then U can see how the sensor react, how much ISO, how many you can compare the highlights and shadows between cameras more objektiv. I see at KineRaw a stop or half underexposed to Canon, that´s probably why the highlights are better..but I didn´t had Cameras so this test can be also conclusion from many days of shooting. Greetings from rainy Innsbruck, Archie

Jason Brooks September 19, 2013


Of course you’re completely right. Whilst this test is very interesting (thanks for posting :-), it’s not necessarily definitive or even representative of the DR performance of the two cameras.

Especially given the whole ETTR strategy/philosophy, the ISO-matching approach doesn’t quite make sense. Sensors are different, and so need to be handled differently.

Very thought provoking though – thanks again.


Morten September 13, 2013

Great test. I have a question though. You said that you could bake in the picture style in the RAW footage? How does that work? Isn’t the whole idea of RAW that you don’t have to use any of the picture styles? Selecting Cinestyle doesn’t change anything from when i’m using any of the other picture styles. I only use it as a flat reference in my EVF.

Silton Buendia September 14, 2013

Cinestyles don’t work for Raw. It would have to be kinda like a LUT but for raw and so far especially with the canons I don’t know of any software that will use the style. Lightroom which is one of the most current raw photo editing softwares (rich would kinda apply to canon raw video) has camera calibrations that you can apply to match the camera styles but even then they aren’t exact so I would doubt that you could apply a cinestyle even more.

Nino Leitner Reply
Nino Leitner September 17, 2013

Thanks guys, you are of course absolutely right. I was fooled by the option in the Magic Lantern menu. But it doesn’t result the outcome of the test. I have amended the text and video.

Andy September 15, 2013

Regular 5DMKIII footage certainly doesn’t have the dynamic range of RAW, but the skin tones always look great and overall the scene would look more beautiful despite the overblown highlights.

Skin tones in both RAW examples look sick. I realize that RAW footage can be color corrected, however, for whatever reason, I have yet to see RAW 5DMKIII footage that looks “as beautiful” as RAW BMCC footage.

Also, what’s up with the serious purple/blue fringing on the tree from 1:56-2:39 on the KineRaw Mini footage?

Victor Martins September 16, 2013

So you’re looking for skin tones? How about this?

Nick October 14, 2013

This makes no sense. Why have the sun in the frame of the first camera? And why have it behind a tree branch on the second? Why even have the sun? The sun is not useful in testing latitude, it’s beyond the capabilities of any sensor or even most optics. A static shot showing sunlight in one part of the image, and shade in the other would show everything about latitude.

Dan Hudgins October 25, 2013

To shoot tests to compare the sensors in various cameras you need to set the 18% gray level to the same code value, in the case of Canon vs. KineRAW-S35 Mini ™ you need to use 4x the value in the Canon data since its converted to 14 bit linear. The 12 vs. 14 is another issue since the noise level is not the same code value in the lower bits.

The shot with the sun on the water has lens flare not showing in the KineRAW-S35 Mini ™ shot.

To get maximum dynamic range you need to use ND filters and set the KineRAW-S35 Mini ™ into “expert mode” that lets you select ISO curve and analog gain separately, in that case EI ISO 2560! (green gain 1x) will give maximum highlight detail, and EI ISO 80! will give maximum shadow range. The noise level at 18% gray will vary depending on the EI ISO curve selected.

For best shooting the EI ISO curve is selected to match the subject matter, and then the shutter angle and ND filter + IR cut filter are used along with the lens iris to get 18% gray aligned onto the 18% gray tick mark on the histogram in Post or RAW mode (in post mode 18% gray card is about 46% on the waveform monitor).

Using KineLOG is different from using LOG90 from the Cineform ™ recording as KineLOG is normalized for the EI ISO selected, in the case of KineLOG you do not get full super white range used unless you set at least EI ISO 1280, so if you set a lower EI ISO you do not get full super white range and there will be no data set in the upper parts of the code range after going through the 3D-LUT saved for each shot, the reason for that is that at EI ISO 80! there is not highlight range, sensor clip is at code value 685/1023 in the Film Log conversion. At EI ISO 1280! there is enough data to get to code value 1023/1023 filling the full Cineon ™ data range like a film scan. At EI ISO 2560! the upper stop is rounded in a shoulder curve like film would have so no highlight data is clipped.

Its best to use the zebras and raw RGB waveform when setting the exposure, then grade the results later from the 12 bit linear data in the DNG files, that way you can ignore the 3D-LUT and extract the maximum amount of shadow detail. In this case the shadows could have been graded up more in my free DI system to then be de-noised to get more shadow detail while holding the highlight detail captured.

Dan Hudgins October 25, 2013

If the exposure of two cameras is set so that the highlight range is the same, that is 18% gray subject is the same number of bits under sensor clip, in this case you should have used a Cir.Pola filter to cut the glare on the water by one stop, then with the correct grading, such as can be done with my free programs, the range between sensor clip and 18% gray would look about the same, under 18% gray the number of bits would vary, but noise bias can level how that looks after temporal noise reduction is applied. I’ve seen results from both the KineRAW-S35 ™ prototype and some Canon data DNG files someone made, and in spite of the Canon data being saved in a 14 bit packed DNG, I don’t see a huge difference, there is still quite a bit of noise in the lower 6 bits of the Canon data, and the Canon OLPF looks like it passes more aliasing and chroma moire on the test images I looked at. (sm) made some changes to their cameras AFTER I did the color calibrations for them using the S35 prototype #3, so the current LK5 and LT5 files may be somewhat invalid. I told them if they wanted to send me a Mini I could try to fix what got messed up by comparing the raw data from S35 prototype #3 and the current production, so far Jihua has not given me any idea of if or when that may happen, so any color errors you may see are not my fault to at least some extent as I told them the color calibrations should ONLY be done after the camera design was fixed and would NOT change any more, otherwise they could invalidate the calibrations which seems to have happened. How bad things are is hard to tell since the way people are processing the camera data puts the results so far out of whack that there is not white reference that matches the cameras K and light type calibrating as far as I can tell, otherwise the results should be neutral, even so the manual analog gain override menu should be able to to be used to zero the white balance, as I was able to do with some of the current S35 they had with them at NAB 2013.

Dan Hudgins October 25, 2013

Jihua does not seem to understand the correct use of “expert mode” where the [F2] button selects the EI ISO curve in 1/3 stop increments as he requested me to calibrate, nor does high engineer Cheng really seem to have a full grasp of how 3D-LUT are used in post production and how such a 3D-LUT must be calibrated RELATIVE to some kind of raw or normalized data.

The use of DNG processing programs upsets the normalized data levels, as the 3D-LUT are made relative to the Cineform ™ LOG90 data levels, to use the DNG with the 3D-LUT requires using the extra 3D-LUT supplied that converts the 12 bit packed linear data in the DNG files into “LOG90″ data equivalent to the Cineform ™ log encoded data (not to be confused with FILM LOG or KineLOG, as the Cineform ™ log data has no matrix and is not normalized for any 18% gray signal level) so that the 3D-LUT can operate on the log-raw signal levels in a way that is equal.

Not all DNG processing software supports the linear to log conversion 3D-LUT so that the monitoring 3D-LUT can be applied, you can ask them which programs they now work that way and how to set them up to use 2x 3D-LUT at the same time as would be required (linear to log, then log to monitoring match).

Dan Hudgins October 25, 2013

I was told that they have the cameras default to “non expert more” because their customers in Asia were “confused” by setting the sensor gain and EI ISO curve manually (as far as I could make out), so when in default mode the analog gain for the green may be set to more than 1x when you select the ISO button, and so reduce the dynamic range causing more clipped highlights, to avoid that you should disable the “simple” mode and set “expert mode” so you can have manual control over the EI ISO curve selected and the amount of analog gain applied to the green channel. In white balanced operation (EQUAL.LT5) the red and blue analog gain are always above 1x to get matching signal levels for red and blue with the green signal level, that was done to improve the results overall, as well as for Cineform ™ compressed raw recording and to avoid having the DNG matrix settings change for each K and light type selection.

Because the data values for RGB are to always be equal, you can go into the manual analog gain override menu and set the RGB gains to all be the same, in this case 1x for best results, then use an optical filter like 81EF and or CC30M to CC60M to set the light going into the sensor to have a warm pink color to counter the native greenish sensor balance, there by getting the “best” results possible, better than in default operation since there would then be less noise in the red and blue channels, but that only works when you have enough extra light that you can loose a stop or two, but better than using an ND filter since using ND with analog gain is not as good as using no analog gain and a pinkish filter.

I suggested to Jihua that he have Tiffen make a set of optical correction filters for his cameras that are correct balanced for Daylight and 3200K etc. but he ignored my suggestion, so you will have to stack some CC filters to get close and then apply a little red and blue analog gain in the manual override menu to trim out the white balance to get it exact.

Dan Hudgins October 25, 2013

The only thing that matters when shooting to the DNG files is that you have the sensor raw signal levels not clipped on important highlights and that you have the actors faces at high enough signal level that they are not pushed into the bottom four bits too much.

If the highlights are important, use a Cir.Pola filter to rotate until the highlights are reduced in their waveform value, that lets you open the iris a bit to increase the midtone code values. If the mid-tone code values are still too low on the raw RGB waveform display, you can then use a reflector or fill lights to bring them up to be high enough that post production noise reduction will give a satisfactory result.

The selection of KineLOG, KineCOLOR, or Kine709 has no impact on the saved data values in the DNG files, only the raw RGB waveform exposure levels matter. You should set the zebras to 90% raw signal and the waveform and histogram to raw RGB and keep the actors faces between the 18% gray and 90% white tick mark lines on the histogram, if you cannot hold enough highlights then you need more fill light or you have to cut the highlights with a Cir.Pola filter or both.

Dan Hudgins October 25, 2013

Violet fringe can in part be due to the lens used, the thickness of the OLPF filter can impact different lenses differently, and the way the image is graded and the original exposure level on the sensor can impact how visible that is later.

The white balance the sensor is set to can also impact the violet halo, using a UV cut filter may help as some lenses do not focus well in the UV. The f/ stop used can also impact this issue as some lenses have a bad blueish halo at more open stops.

The de-Bayer software can also impact purple halo as chroma moire can convert into purple halo. Even focus settings can make this more or less obvious as can over exposure of the sensor or excess blue gain.

Dan Hudgins October 25, 2013

If someone is confused about the operation of the KineRAW ™ cameras they can email me at: tempnulbox (at) yahoo (dot) com.

Unless one is trying to make the cameras look bad in comparison to some other camera, knowing how to set the exposure level correctly and what methods are used to process the raw data that show it to best advantage should be applied, otherwise all you end up with is badly exposed and processed camera footage looking bad, so how could it look otherwise?

Digital Cinema Cameras are not film cameras, you do not set an ISO and hope for the best, once something important goes over the sensor clip or down in to the FPN and noise you cannot save it, that is why the KineRAW ™ cameras have a live raw and post monitoring waveform monitor built in, so there is no guess work with regard to the relationship of the subject light levels and the code values being saved in the DNG frames.

You should learn and understand how to use the Waveform monitor and zebras, and purchase Cir.Pola, ND and hot mirror filters, as well as UV cut skylight filters, no camera and totally compensate for the user, and in the case of a totally manually operated Digital Cinema Camera, the DP and camera operate need to know what they are doing, experience with video and film cameras is no substitute for understanding the operation of the camera being used.

If you want something explained ask me, as talking to Jihua or Cheng may not get you to where you need to be as they are not filmmakers nor do they have experience as DP.

Sometimes you have to have a generator for fill lights if you want good results, or you need to re-frame the shot so that it turns out well for the audience. That was certainly true for 35mm movie film, and it is no less true for Digital Cinema Cameras, the idea of a camera you can point at anything any time of day and expect cinematic results is not in alignment with the history of the work of the better cinematographers and their best films which used lighting and framing selected to look cinema-graphic within the limits of the media they ware using. Just setting a camera to HDR mode does not give you cinematic results as those are obtained with lighting and art direction as well as camera work.

Dan Hudgins October 26, 2013

There is no need to “test” the dynamic range of the KineRAW ™ cameras, with EQUAL.LT5 active the data range is 12 bits for red, green, and blue, unlike say the BM camera shooting native sensor balance were the top one or two bits is cut off so you only use maybe 12 of their 14 anyway.

The dynamic range should be at least 11.5 stops in the green, and may show more in photographic tests due to lens flare and noise bias when temporal noise reduction is applied.

The sensor native speed is about EI ISO 80 when green gain is 1x, that is based on having 90% white subject show about 90% on the raw RGB waveform display, so in effect there is no highlight head room at the native speed, to get highlights above 90% white subject, specular highlights, you need to underexpose the sensor by selecting EI ISO curves that lift midtones and crush the highlights, the raw recording will be underexposed for all speeds selected above 80 when analog gain is not applied to the green channel.

So when green gain is 1x you get this highlight range,

80 = 0 stops
160 = 1 stop
320 = 2 stops
640 = 3 stops
1280 = 4 stops
2560 = 5 stops

Its not useful to use an EI ISO curve faster than 2560 since there are not enough bits under 80% gray past that to avoid serious histogram gaps. To get higher EI ISO analog gain is applied, it can range between 1x and 64x for all the EI ISO curves, so,

80! = 80 at 1x to 5120 at 64x
160! = 160 at 1x to 10240 at 64x
320! = 320 at 1x to 20480 at 64x
640! = 640 at 1x to 40960 at 64x
1280 = 1280 at 1x to 81920 at 64x
2560 = 2560 at 1x to 163840 at 64x

Speeds above 20480 are not useful for cinema use at 2048×1080 resolution, but could be used perhaps for SD documentary when needed if heavy nr is applied in the right ways.

In general its best to apply analog gain to increase the EI ISO when you have enough highlight range, that is 640 based on 160×4 may look better than 640×1.

For low light shooting try 1280! x2 for 2560 and x4 for 5120.

Because the camera controls are not labeled in a way that is clear, [F2] on the S35 controls the EI ISO curve, 80! to 2560!, and the ISO button controls the analog gain (1x to 64x by way of the ISO value based on the [F2] curve active, which get confusing as it does not start at gain 1x by default when you change the EI ISO curve with [F2] so you need to check what the gain is or you may not be shooting at the speed you think you selected), it would have been more clear what was going on if they had a Curve and Gain button, but they did not listen to my suggestions in that regard.

Once you figure out the odd menu and button options it should be straight forward as to what the exact highlight latitude you have selected, unlike other cameras that do not publish clear descriptions of their internal operations, perhaps.

How you process the raw data is important as some programs can clip the highlights as part of their internal operations (such as XNVIEW sometimes and maybe DCRAW as well). My free de-Bayer program is under full manual control and ignores the DNG header tags to let you select the clip points manually, so the green clip point is normally left at 100% signal all the time to avoid highlight losses. Not so in other software, so frequently you are not comparing the camera data only your workflow, unless you make linear tif from the DNG using a program, like mine or from Adobe ™.

Dan Hudgins October 26, 2013

If you want to apply some stock “3D-LUT” to apply a “look” you would need a 3D-LUT that is referenced to the camera data you want to transform, in this case the KienRAW ™ camera data.

But if you don’t have 3D-LUT made for one of the normalized results, Kine709, KineCOLOR, KineLOG, then you will not get an exact match to some other camera, KineLOG may be the closest though since it conforms well to 3D-LUT made for correctly exposed Cineon ™ film scans.

Keep in mind that you CANNOT process the DNG files in any-old-software and get the correct output for KineLOG, LineCOLOR or Kine709 as the results will probably look the same no mater which look group you selected in the camera, that is because you have not applied the linear to log and then log to monitoring 3D-LUT that is saved in each shot folder, those were intended for some DI systems that allow for such DNG processing, but mostly for the Cineform ™ workflows. If you process the DNG frames in some other way you will get results that are in no way calibrated and do not represent the Kine709, KineCOLOR or KineLOG target result code values, that is because the program you are using is not employing the supplied 3D-LUT to transform the data.

In that case you may do better to convert the DNG to Cineform ™ using their conversion program then apply the 3D-LUT from each shot folder to the processing of each shot in the Cineform ™ codec by draging the shot’s 3D-LUT into the Cineform ™ 3D-LUT tool program so that the Cineform ™ codec can know where to find the right 3D-LUT for each shot. After you de-Bayer to DPX files using the Cineform ™ coded, you can then apply additional 3D-LUT look files as you like provided that they are based on similar normalized data code values for midtone etc. The conversion of the DPX files can be to other DPX files, of you could convert Cineform “raw” to Cineform RGB or YUV then apply the 3D-LUT for the look in the Cienform software from go-pro, they provide many stock looks in their software that work with any data once its in their Cineform YUV format etc.

So the workflow is,

DNG through linear to log, then log to monitoring 3D-LUT to make RGB or YUV


RGB or YUV through “3D-LUT look file” to get result format RGB or YUV etc.

or you can go,

DNG to Cineform ™ “raw”

Cineform ™ “raw” to Cineform ™ RGB or YUV or DPX through each shots’s 3D-LUT

Cineform ™ RGB or YUV or DPX through “3D-LUT look” to end use format.

Other workflows based on DNG to something without going through the 3D-LUT supplied will not represent the KineRAW ™ supplied looks because the software knows nothing about using their 3D-LUT from the camera’s shot folders.

Leave a Comment

Connect with:

Your email address will not be published. Required fields are marked *