A CRI-95 LED Can Still Flatten Skin at 3200K. Read the Spec Sheet Like a Colourist
Why LED colour scores drop in tungsten mode, what CRI, R9, TLCI, TM-30 and SSI really measure, and a checklist to test lights on your own camera.
Cover illustration generated with AI for Fix It In Post.
A CRI of 95 on a spec sheet tells you almost nothing about how a light will treat a face at 3200K on your camera. The number that matters is rarely the headline one, it is usually measured at daylight, and it is often a claim nobody has checked. Read LED spec sheets the way a colourist would: tungsten-mode figures first, red rendering second, camera-specific tests last and most important.
The measurement that should worry you
In April 2026 Newsshooter put the Aputure STORM 700x through a Sekonic C-800 spectrometer. At 5600K with the 35-degree reflector it measured a CRI (Ra) of 97.6, an extended CRI of 95.6, TLCI 95 and R9 of 76.3. Set to 3200K, Ra fell to 94.9, extended CRI to 93.4, TLCI to 91 and R9 to 67.4. The reviewer's summary: the fixture is optimised for daylight, slightly sacrificing tungsten rendering.
That is a flagship light, and these are still good numbers. But they show the pattern that matters for anyone who lights interiors warm: the same fixture gets measurably worse when you dial it to tungsten, and the headline CRI barely moves while R9, the saturated red, drops by nine points.
Two other details in that review are worth knowing. First, R13, the CRI sample meant to approximate light skin, stayed at about 99 in both modes. So "muddy skin" is not something the skin sample predicts. The risk sits in the deep reds: lips, flushed cheeks, darker skin with red undertones, wood, wardrobe. Second, at 5600K the same reading gave R9 of 76.3 against the CIE daylight reference and 92.5 against a blackbody reference. Change the reference illuminant and R9 moves by sixteen points without the light changing at all. A spec sheet that does not say which reference it used is not comparable with one that does.
What each number actually measures
- CRI (Ra) averages eight pastel test colours viewed by a model of the human eye. As LED professional notes, those eight are medium-chroma samples that exclude saturated reds. A light can post a 95 Ra and still be weak in the reds.
- R9 is the extra CRI sample for saturated red. It is the quickest single check for the problem in the headline, and it is the number manufacturers most often leave off.
- TLCI is the EBU's index for television. It models a standard camera and display rather than a human eye, on a 0-100 scale, and is specified in EBU Tech 3355. Better than CRI for us, but the camera is a generic model, not yours.
- TM-30 uses 99 colour samples, including skin tones, textiles and inks. Rf is fidelity (100 is perfect); Rg is gamut, where 100 means the same saturation as the reference, above 100 means more saturated and below means duller. An Rg slightly below 100 with weak reds is a likely recipe for skin that looks grey-brown.
- SSI, from the Academy, compares the light's spectrum directly with a reference spectrum, such as studio tungsten or daylight, without assuming any eye or camera. The Academy's own guidance: high values mean predictable rendering on most cameras; low values may look fine on one camera and wrong on another. For mixed-camera shoots this is the most useful number on the sheet.
Claims versus measurements in 2026
This year's launches lean hard on emitter counts and high indices. What has actually been published:
- Godox PaletteLab (announced April 2026): nine diodes, and Godox claims R9 of about 99, TM-30 Rf of about 97 and Rg of 101, SSI figures of 92 and 87 without stating which reference each refers to, and 92% of Rec.2020. The SL300RF lists at $719. We have not found an independent measurement yet.
- amaran Ray (announced November 2025): an "OmniColor" engine with an added indigo emitter. amaran claims SSI 87 at tungsten and 80 at D56, from $189 for the 60c to $699 for the 660c. Note that the claimed tungsten figure is higher than the daylight one.
- Nanlux Evoke 5C: an eight-emitter "Nebula C8" engine including deep red and indigo, rated CRI/TLCI 98/98 and TM-30 Rf/Rg 96/100 on average, at $219.
Notice the words "average" and "about". An averaged figure across the CCT range can hide a weak tungsten end, and that is exactly what the STORM measurement found. More emitters, deep red ones especially, should help R9. Whether they do at 3200K on a given unit is an empirical question, and all three sets of figures above are manufacturer claims as of launch.
Why tungsten mode is a colourist's problem
On set, a light at 3200K with an R9 in the 60s does not look wrong to the eye, and the monitor may look acceptable after a white balance. The damage shows up in the grade. Weak saturated red means less separation in the reds and oranges of skin, so faces flatten. The colourist then pushes saturation or hue-versus-hue curves to bring life back and drags everything else in that range with it. If the key is a narrow-spectrum LED and the fill or practicals are tungsten, you also get two different skin renders in one shot, which is far harder to fix than a global cast.
Multiple cameras make it worse. A low-SSI light can render well on one sensor and badly on another, so the B-camera match you budget for in the grade becomes a shot-by-shot job.
The spec-sheet and test checklist
- Ask for the figures at 3200K, not just 5600K. If the sheet gives a single "average", treat the tungsten end as unknown.
- Look for R9 and TM-30 Rg. Our own rule of thumb, not a published threshold: be wary of an R9 below about 80 at the CCT you will actually use, or an Rg below 98.
- Check the reference. A figure with no stated reference (CIE daylight, blackbody, studio tungsten) cannot be compared across brands.
- Prefer independent measurements over claims, and note the date. Firmware and production runs change.
- Test on your own camera. Light a colour chart and a real face with the LED at 3200K, then with a tungsten halogen source. Balance each on a grey card, shoot both in your usual log and compare them on the vectorscope and with a skin-tone qualifier. Include something saturated red in frame.
- Keep the key and fill on the same engine. If you must mix, put the better-measured unit on the face.
A high CRI is the minimum price of entry in 2026, not proof of quality. Light for the CCT you shoot at, check the reds, and let your own camera have the final word.
