2024 CML Camera/Lighting Tests – Setup & Processing

Part I: 9 Cameras and 23 Lights Combined

As in our NFA 2020 tests we captured the same scene with 9 different cameras — the GH7 captured in both raw and ARRI LogC3 modes, so it counts twice — under 23 different lights, at both 3200 K and 5600 K settings.

In addition we captured 3 lights at 4400 K following a discussion on CML about using that color temperature: the suggestion was that 4400 K optimizes the output from bi-color fixtures since both the “warm” and “cool” white emitters are driven equally, and the warmer-than-daylight CCT should render better skintones than red-poor 5600 K. You can compare the 4400 K clips against those from the same lights at 3200 K and 5600 K.

Two models and several test charts were illuminated with lights filtered through a vertical 4x8 foot panel of Lee 216 diffusion 4 feet from the subjects. The 216 evened out different lights’ beam patterns (except for the DMG Mini, undiffused due to its low output). A 4x8 bounce card on the opposite side provided fill. Tube lights (Aputure Infinibars) provided color-temperature-matched rimlight and served as a subtle, consistent reference.

(This setup matched Geoff Boyle’s specifications, unlike the NFA 2020 tests where we mistakenly set up the 216 horizontally, 8.5 feet fom the subjects. As it turns out, the NFA setup gave us more even illumination across the frame; I'd recommend using that setup the next time we do this test.)

All lights were set to their manufacturer’s preset values for 3200K and 5600K and, where the option was available, they were set to the appropriate camera (for example an “Alexa mode”). Light level was set to 25 fc measured in the middle of the Grey Scale card using a Sekonic C-800 pointing directly at the camera.

All cameras were set to ISO 800 and T2.8. This results in all cameras being exposed one stop more than the “correct” exposure, as this is often the best balance between dynamic range and noise. Cameras were set to manufacturer’s presets for tungsten and daylight white balance. 4400 K was dialed in directly; on the Canon 4440 K was the closest available setting.

Most cameras used a Cooke S8 65mm lens and captured a Super35mm-ish image, even if the camera has a full-frame or larger sensor. The GH7 with its 4/3 sensor used a 32mm Ultra Prime for a similar angle of view.

All clips aside from the GH7 in ARRI LogC3 mode were captured raw; the GH7’s LogC3 was recorded as 10-bit H.265.

Two Blackmagic cameras were tested: the URSA Mini Pro 4.6K G2 and the URSA Mini Pro 12K. The 4.6K uses a “traditional” RGB Bayer-pattern sensor with red, green, and blue photosites. The 12K uses an RGBW color filter array; in addition to red, green, and blue the CFA has broadband, “white”-sensitive photosites. Thus we had two Blackmagic cameras with different sensor CFAs but a common “color science”.

The GH7 was captured as ProResRAW and decoded in Final Cut Pro, resulting in an image that looks very much like V-Log. The GH7 also captured using its optional ARRI LogC3 profile, recorded as 10-bit HEVC. Thus we had one camera capturing with two different “color sciences”.

All clips aside from the GH7’s were loaded in Resolve and the default raw processing applied on an ACEScct timeline.

The GH7’s ProResRAW was loaded in Final Cut Pro and output as ProRes 4444 V-Log, then taken into Resolve. The GH7’s ARRI LogC3 HEVC clips were taken directly into Resolve. ACES input transforms for Panasonic V35 and ARRI LogC3 respectively were used to convert the GH7’s clips to ACEScct.

All clips were sized and repositioned to align the ColorChecker and CML Special charts. Level on the gray card (the gray patch of a DSC Labs OneShot) was set to 48% with a Rec.709 output transform (48% = Kodak's recommended value for Rec.709: 122/255) using the offset control’s “master wheel” only.

(As the DSC ChromaMatch GEO chart was bumped between setups, its illumination wasn't consistent between lights. A power window was placed over the ChromaMatch chart and a lateral gradient applied to even its exposure side-to-side, and to bring the level of its grayscale crossing up to 50%. This was done to compensate for variations in chart illumination from test to test. Unfortunately the AcuFlect surface suffers from substantial flare, raising black levels and desaturating colors, often affecting the left side of the chart more than the right (visible in the WFM). ChromaMatch images in these tests are not indicative of a light’s and/or camera’s color rendering ability and are only useful for comparing the relative performance of different cameras under a given light. Compare ChromaMatch images and ’scope patterns only between tests with the same light.)

Some clips had incorrect camera names on their slates; these were fixed in post. The 5600 K GH7 LogC3 / DMG Mini clip was captured with the wrong white balance, so it was color-corrected using the offset control’s “color ring” (trackball), and that fact is noted on the slate.

These clips were then output as “leveled” — their exposures have been matched but their colors are the result of default settings on both light and camera (that one GH7 clip aside).

The clips were then white-balanced on the gray card, using only the offset control’s “color ring” (trackball), and output as “leveled and balanced”.

All outputs are 10-bit H.265 Rec.709, uploaded to Vimeo for viewing.

We chose Rec.709, not Rec.2020 ST-2084 (1000 nits, P3 D65 limited) as used in the NFA 2020 tests. Back then we let Vimeo convert the Rec.2020 to rec.709, but Vimeo’s conversions reduce apparent exposure and contrast, as Vimeo compresses the 1000 nit range of the original to fit the 100 nit range of Rec.709. If you downloaded the Rec.2020 originals, you saw one thing; if you viewed or downloaded Vimeo's 709 conversions, you saw something rather different (see a comparison of outputs here. If you have an HDR capable display and a browser that understands it, the Rec.2020 clips will play in HDR/WCG, otherwise in a darker, low-con SDR. The Rec.709 clips will appear the same regardless of whether your display chain is HDR-capable or not). Our tests do not stress either dynamic range or color gamut, and we saw no significant comparative differences between clips when they were output as 2020 or as 709 — so we opted for the consistency of Rec.709.

Part II: 33 Lights Measured

We set up 33 lights and measured their CCT, ∆uv, and color-quality metrics at 3200 K, 4400 K, 5600 K, and 6500 K, all at 100% power. We also measured color and output at 50%, 25%, and 10% power at 5600 K, and looked at the effect of modifiers (diffusers, fresnels, etc.) on both color and brightness.

Lights were rigged 9 feet 11.5 inches from a wall — 3 meters, plus the thickness of a C-800 spectrometer — and aimed at a target, in this case the middle of three mysterious wall-mounted pipes. Tape marks on the wall indicate degrees of beam spread at 15º intervals for beam-pattern photos. C-800 readings were taken by holding the meter up against the wall at the aim point of the light, so all readings are from the center of a light’s beam (or, in this case, 1.5 inches / 3.5 cm off-center, as the target pipe was in the way).

Beam-pattern photos were taken from a fixed position beside the light with a 7mm lens on a Lumix GH5 set to 5600 K, “standard” photo style. No editing in post other than resizing for the web.