We tested how two of the most popular cameras match up on exposure, noise, color fidelity and more.
[Editor's Note: Contributors to this piece include Charlie Anderson, Alex Chinnici, Justin Derry, and Oren Soffer.]
Given all the buzz around RED’s new Helium sensor, we decided to put it to the test against one of its biggest competitors, the ARRI Alexa, in a variety of real-world lighting set-ups in which we could push both sensors to their limits and compare them at different extremes.
Note that this is a test specifically comparing the RED Epic-W camera with the new Helium sensor to the ARRI Alexa Mini with the Alev sensor. This is not meant to be a general “RED vs. ARRI” test, as there are other configurations of RED’s sensors that have been demonstrating very impressive dynamic range and image quality results, including the Weapon Helium, and the Vista Vision Dragon, which we did not look at in this specific test.
Before we dive in, some technical info:
The ARRI Alexa Mini was shot in 3.4k Open Gate ARRIRaw mode recording to C-fast cards, while the RED Epic-W Redcode Raw footage was shot at 8KHD with a 6:1 compression. This was selected because it was the lowest possible compression we could attain with the 240GB RED Mags we were recording to. It’s worth mentioning that with the 1TB high speed RED Mags, it is possible to shoot 8KFF at 6:1 compression on the Epic-W. The Epic-W was also equipped with the Standard OLPF for the purposes of this test.
Unless noted on the clips, both cameras were set to a color temperature of 4000K as a mid-point between their ideal color temperatures, and were shot at ISO 800. In our testing, we found that the Helium’s native ISO was probably closer to ISO 1280, but decided to match both cameras to 800 for the ease of comparison. All footage was shot on 25mm, 35mm and 50mm Leica Summicron-C lenses, at varying T-stops. We utilized a variety of units for lighting, including a pair of Digital Sputnik DS-1 heads as our main units, but also incandescent bulbs, practical lamps, candles, and natural daylight.
For post, we performed a one-light grade on the footage with Alan Gordon from Post Pro Gumbo; no secondary grades were performed. The Alexa footage was debayered into a Log-C gamma space and then graded based on a Rec709 LUT, with slight modifications. Alan noted the IPP2 and the Helium post-pipeline is still a work-in-progress, which makes it a bit difficult to compare to the Alexa, whose post pipeline has been tried and true and unchanged for many years now–but we still thought it would be pertinent to compare, as both cameras are currently on the market.
For the Helium footage, we used RED’s new image processing pipeline, IPP2, and debayered the footage into a Log3G10 gamma space, which was then graded based on the IPP Preview 2 LUT. During testing, we discovered that the previous standard Log space for RED footage, RedLogFilm, was a much closer match to Log-C in terms of brightness and exposure. From our observations, Log3G10 didn’t actually demonstrate a particular advantage over RedLogFilm in terms of highlight retention, despite it being designed for that purpose. Instead, Log3G10 seems to just pull the overall exposure of the image down by about a half stop, and creates slightly smoother rolloff in the shoulder.
The Alexa has at least a stop and half more dynamic range in the highlights when compared to the Helium.
Despite this, we decided to stick to RED’s recommended workflow for Helium for the purposes of this test, while acknowledging that it is a work in progress. We should mention that we edited and colored this footage a month after the first version of IPP2 was released and a few weeks before the latest iteration of IPP2 came out.
Test 1: Overexposure and dynamic range in highlights
In our first test, we wanted to push the sensors to the upper limit of their dynamic ranges, and see how much information could be retained in the highlights and brought back down to “neutral” exposure in the grade. We determined our “neutral” exposure of the subject’s face with our key light measured at an F/2.8 to be at 60IRE, and graded the overexposed shots back down to match that value. In each subsequent shot, we increased the brightness of the key light by one stop, until we capped out at +5 stops overexposure from our base exposure (T/2 on the lens).
In the test results, it is evident that the Alexa has at least a stop and half more dynamic range in the highlights when compared to the Helium. At four stops overexposure, the Alexa footage could be graded back down to 60IRE without losing any information in the highlight at all. At the same stop, the Helium footage begins to clip and lose data that cannot be retrieved or graded back to match. The cutoff is pretty sudden—at three stops overexposure, the Helium footage could still be perfectly matched to the base exposure, but at four, the information is already lost. Notice the information loss in the bare bulb as well in the Helium footage compared the Alexa, which retains more detail in the filament.
At five stops overexposure, the Alexa footage begins to clip in the most overexposed regions of the subject’s face. The Helium footage has total information loss in, and sharp falloff from, overexposed regions. The Alexa also seems to retain more color detail in the overexposed regions, so when they are graded down, the skin tone still maintains its color. In the overexposed regions of the Helium image, color information is lost, and the result when graded down is a very washed-out look.
Test 2: Underexposure and noise
In the second test, we set out to examine the cameras' performances at the opposite extreme: low light scenarios. Our subject was lit entirely with candlelight, his face reading at 40 IRE at our “base” exposure of T/2 at ISO 400. The rest of the image was left to fall into darkness, except for a string of background lights. In each subsequent shot, we stopped down on the lens and increased the ISO to compensate for the exposure loss. Our goal was to track the noise level at increased ISOs as well as color and image fidelity in underexposure.
In this comparison, it is clear that the Helium is much cleaner at higher ISOs than the Alexa, but we found that the Alexa holds up surprisingly well and performed better than we expected. At 1600 ISO, the Alexa is still relatively clean; at 3200, the noise floor is plainly evident, especially when compared to the cleaner Helium image.
Interestingly, the RED image loses color detail when it is underexposed and re-balanced in the grade, whereas the Alexa footage retains all of its color detail in the lower-mid tone range of the image, even at higher ISO’s. You can see in the comparison above that the color fidelity in the subject’s face and the background of the scene changes from ISO 400 to ISO 3200 on the Helium sensor.
The visual and color details in the shadows on the Helium sensor seem to lose information sooner.
The rolloff into the shadows on the Alexa is more gradual as well; the visual and color details in the shadows on the Helium sensor seem to lose information sooner. From what we can tell, the Log3G10 gamma space clips information in the shadows. Therefore, when attempting to lift information out of the shadows in Log3G10, the information remains clipped, whereas Log-C on the Alexa re-interprets the information and is able to pull more visual and color detail out of the shadows when underexposure is “lifted” in the grade.
Test 3: Color fidelity
In this test, we set out to specifically check the color fidelity of the cameras and how closely they are able to match to colors in reality. We lit a brightly saturated scene using our Digital Sputnik DS-1 heads and set our initial exposure at T/2 and ISO 800, and stopped down incrementally to test color detail retention in underexposure. In the color grade, we matched the green and magenta tones in the background across both cameras, and let the other colors fall the way the cameras interpreted them.
The first and most obvious difference between the images is how the two cameras interpret the teal key light on the subject’s face. The Alexa leans towards the green end of the spectrum, which results in a much greener shade of teal. The RED, on the other hand, has a magenta lean, which results in a much bluer-looking key light.
A more subtle difference in interpretation between the cameras can be found in the color red: the Alexa has a lot more pink in its rendition of red, while the same color on Helium leans a bit more towards orange. Throughout our testing, we found that in general the Alexa Log-C to Rec709 LUT is far more aggressive with saturated colors than the IPP2 workflow, which favors more muted colors.
Test 4: "Natural" interiors
We wanted to set up a real-world lighting scenario that DPs find ourselves in often: a naturalistic and contrasty interior with a bright, over-exposed window, deep shadows, and mixed color temperatures. We shot the test on an overcast day and so had to place a 1x1 LED panel just outside the window in order to create a blown-out hotspot on the curtains in the background. This light also flared the lens a little bit, which we decided to embrace, as it fit with our goal of creating a “real world” testing scenario, warts and all.
In the shadow regions, the Alexa manages to retain more information and visual detail.
The rest of the scene was lit solely with the practical lamp on the coffee table, and with no fill light. Each subsequent shot was underexposed by a stop (our starting stop was a T/2) and the ISO lifted in the grade to compensate. We attempted to retain detail in the highlights as much as possible while lifting the mid tones in the image to see if we could find a spot in which the window and the foreground were both properly exposed.
This test served as a great summary of all the differences between these cameras for us. First, in the highlights, we found that the Alexa is able to retain more detail in the overexposed curtains than the Helium, and the rolloff is more gradual. At two stops underexposure, the curtains (which metered at a 32, +8 stops overexposure from middle gray) retain full detail on the Alexa, while they still show some clipping on the RED. At three stops underexposure, both cameras are able to retain full detail in the curtains. The lamp on the coffee table is also clipping in our base exposure (T/2) shot in the Helium footage, whereas the Alexa shot retains all highlight detail in the lamp.
In the shadow regions, the Alexa also manages to retain more information and visual detail, whereas the Helium clips information in the shadows. This is best demonstrated at the farthest extreme we tested—4 stops underexposure—where you can see in the comparison above that there is information loss in the cheeks and shadow regions of our subject’s face in the Helium footage. This can also be seen the darkest regions of the leather couch the subject is sitting on.
The Alexa footage has a green tint that differs greatly from the RED’s interpretation of daylight, which is a lot cooler and more neutral.
In the overall image, the Alexa footage has a green tint that differs greatly from the RED’s interpretation of daylight, which is a lot cooler and more neutral. The light emulating from the lamp is also a lot more saturated in the Alexa footage and results in more warmth in the subject’s skin tone as compared to the more muted Helium footage. Finally, worthy of note is how much cleaner the shadow regions are in the Helium footage compared to the Alexa footage.
Test 5: Available light at night
In our final test, we took the cameras out at night to shoot some uncontrolled available light and see how they interpreted some odd color temperatures that can be found on the street. Specifically, we wanted to see how the cameras handled sodium and mercury vapor lamps differently. Similar to our colored lights test, we noticed that the sodium vapor spectrum leaned more towards yellow in the RED footage, whereas on Alexa, it was a deeper orange. Additionally, the mercury vapor lighting looks greener on the Alexa, while the RED interprets it as a little more blue. Ultimately, we found both interpretations of mercury and sodium vapor interesting in their own way and both cameras performed well in uncontrolled lighting environments at night, with relatively clean images even at ISO 1600.
All in all, we found this test quite illuminating. It reiterated things we knew from using these cameras in the field but also demonstrated many things we were surprised to learn about each of them. Namely, the Alexa has more latitude in the highlights (about 1.5 stops more dynamic range than the RED) and is able to retain more color detail through various degrees of over and under exposure. The RED has less noise in higher ISOs than the Alexa, and by our estimation has a significantly improved color fidelity when compared to previous RED sensors.
That being said, both of these cameras create viable images and are at the top of their technical game, and a preference of one over the other is purely subjective. Ultimately, they are just tools, and no one camera can substitute a DP’s artistic eye and taste for lighting and composition when it comes to creating a compelling image.
Feel free to draw your own conclusions by downloading our test footage from both cameras, as well as the stills from this article, at this link to our full resolution RAW files. Please note that in order to use the IPP2 Preview LUT, you need to select "Show Advanced Controls" in Red Cine X. The LUT can be downloaded here.