# What 670nm Red Light Actually Does for Sleep My Weird Prompts — an AI-generated podcast. Daniel Rosehill sends in a prompt; AI hosts Corn and Herman discuss it. Verify important claims independently. - Episode: 5028 - Published: 2026-09-04 - Duration: 20:48 - Audio: https://episodes.myweirdprompts.com/audio/red-light-670nm-sleep-specs.m4a - Tags: circadian-rhythm, lighting-design, hardware-engineering - Web page: https://myweirdprompts.com/episode/red-light-670nm-sleep-specs/ - JSON: https://myweirdprompts.com/api/episodes/red-light-670nm-sleep-specs.json ## Summary You've seen "670nm" on every red light listing, but what does that number actually mean for sleep? We trace its real origin — not a sleep lab, but photobiomodulation research — and explain why the number matters less than you think. Then we get practical: the spectral chart, the dimming floor, and the driver specs that separate a real sleep light from marketing. If you've ever wondered whether you're buying a serious tool or an expensive red bulb, this episode is for you. ## The prompt > Someone is on Amazon trying to buy a good red light for sleep. They already use one, they find it genuinely helpful, and this is not a question about whether red light works — that is settled and the hosts should treat it as settled. The question is how to tell a good product from a bad one, and it starts from a specific point of confusion. > > **The confusion:** every serious sleep-lighting brand seems to print "670nm" on the spec sheet. So either it matters, or it doesn't. If it matters, what is it doing? If it doesn't, why do the good brands all list it — and is its presence a reason to avoid a product, a reason to trust one, or neither? > > **The answer that should organise the episode: 670nm is a badge, not a mechanism — and a badge can still be useful information.** > > Unpack that in three moves. > > First, what the number actually refers to and why it exists at all. It comes out of photobiomodulation research — longwave red absorbed by mitochondrial cytochrome c oxidase, and the retinal work in particular. Treat that even-handedly: a real mechanism, an active research area, genuinely interesting, and separate from the melatonin question. > > Second, why it is not the causal ingredient in a sleep light. Melanopsin peaks around 480nm, so what protects melatonin is the absence of short wavelengths plus low intensity — which means that once you are out at the long end of the spectrum, the precise figure is not what is doing the work for sleep. So no, it is not a reason to avoid a product, and it is not a spec to shop on. > > Third — and this is the part that resolves the caller's actual confusion — **why it nonetheless correlates with the good products.** A brand that prints a peak wavelength is a brand that knows what emitter it bought and has measured it. That implies binned LEDs, a real supply chain, and someone who could produce a spectral chart if asked. A white-label reseller who bought whatever red bulbs were available prints "red." So the number is a decent proxy for manufacturing seriousness while being nearly irrelevant as a mechanism. That distinction — a correlated signal versus a causal ingredient — is the transferable lesson, and it applies well beyond lighting. > > **Then the substance the caller actually asked for: what IS the spec sheet you should be reading?** Previous episodes established that intensity matters more than colour. Take that as given and go to the next axis, which is LED and product quality: > > - Whether a spectral power distribution chart is published at all, and what to look at on it. Peak versus full width at half maximum — spectral purity, and how much of the emission is sitting where you want it. > - Melanopic EDI and melanopic ratio as the metrics that actually answer the sleep question, and why almost nothing consumer-grade quotes them. > - **The dimming floor** — arguably the most under-discussed spec for a bedside light. How low will it actually go, does it drop out or step at the bottom of its range, and why the usable minimum matters more than the maximum for this application. > - **Driver quality and flicker** — PWM behaviour at deep dim levels, why cheap drivers get visibly or invisibly worse exactly where you want to use them, and how to check. > - **Binning** — what LED binning is, why unit-to-unit consistency separates a real manufacturer from an assembler, and what it means when two bulbs from the same box aren't the same colour. > - Thermal management, and its consequences: wavelength shift and lumen depreciation over the life of the product. > - Direct narrowband red versus phosphor-converted amber — the practical differences, including how they age. > > Close with the actual shopping heuristic: the three or four things worth checking on a listing, what a good product's documentation looks like versus a bad one's, and the honest note that above a certain point the returns flatten and a decent amber bulb used consistently beats an optimal one used occasionally. ## Show notes Every red light for sleep seems to print the same number: 670 nanometers. It looks like a spec, but it's actually a badge — one borrowed from a different field entirely. That number comes from photobiomodulation research, specifically Glen Jeffery's work at UCL on retinal health in older adults. The proposed mechanism involves cytochrome c oxidase absorbing longwave red light to support mitochondrial function. That's real biology, but it's about retinal cells, not sleep. The sleep story runs through a different pathway: melanopsin, which peaks around 480 nanometers in the blue range. What protects melatonin at night is the absence of short wavelengths and low intensity. Once you're at the long end of the spectrum, the precise figure barely matters. That doesn't make 670nm useless — it makes it a signal. A brand that prints a peak wavelength and publishes a spectral chart is a brand that bought real, binned LEDs. The question is whether they understood the mechanism or just copied the number. For actual shopping, skip the wavelength and look at three things. First, is there a spectral power distribution chart? Check the full width at half maximum — a narrow peak means intentional design. Second, does the brand quote melanopic EDI? Almost none do, but if they know what it is, that's a strong signal. Third, and most important, the dimming floor. A sleep light needs to drop to a faint ember, not click from off to twenty percent. Test the driver with your phone camera for flicker at low levels, and listen for whine. The best spec is the one nobody prints. ## Transcript Corn: Here's the thing about red light for sleep. The person asking Daniel's question already uses one, every night, and it works. That's settled. What's not settled is what happens when they go to buy a better one and every listing they open says the same thing. Six hundred seventy nanometers. Peak wavelength. As if that number is the reason the thing works. Corn: Daniel's got a whole buyer's confusion wrapped up in that number. Either it matters or it doesn't. If it matters, what is it actually doing for sleep? If it doesn't, why do the serious brands all print it, and is seeing it a reason to trust a product, avoid it, or neither? And underneath that, the real question: how do you tell a good red light from a bad one when every spec sheet looks the same? Corn: The answer that organises the whole thing is that six seventy is a badge, not a mechanism. And a badge can still be useful information. That's the episode. Herman: And the badge has a real history. It didn't come from a sleep lab, which is the first thing to get straight. Six hundred seventy nanometers is a photobiomodulation wavelength. It's the number Glen Jeffery's group at UCL made famous. Corn: The eye study. Herman: The eye study. Twenty twenty. Older adults, declining colour vision, and they had them sit in front of a deep red light for a few minutes a day. The wavelength was six seventy. And the results were interesting. Colour contrast sensitivity improved, in some cases by around twenty percent. The mechanism they proposed was mitochondrial. Cytochrome c oxidase, which is an enzyme in the electron transport chain, absorbs longwave red light. The idea is that it improves mitochondrial function in retinal cells, which decline with age because the retina is enormously metabolically active. Corn: So it's not a made-up number. It's a real research wavelength with a real proposed mechanism. Herman: Right. And I want to be careful here, because it's easy to sneer at photobiomodulation and I don't think that's warranted. It's an active research area. The retinal work is the strongest part of it, and Jeffery's group has published follow-ups. The mechanism is plausible. Cytochrome c oxidase does absorb in that range. Whether the effect sizes translate to clinical recommendations is a different question, but this is not crystals and magnets. It's biology. Corn: And that's exactly why the number acquired prestige. It borrowed credibility from a real scientific literature. A brand that prints six seventy is not inventing a wavelength. They're attaching themselves to a body of work that sounds serious because it is serious. Herman: But here's the problem. That pedigree is for a different question than the one you're asking when you buy a sleep light. The photobiomodulation story is about mitochondrial function in retinal cells. The sleep story is about melatonin. And those are two different mechanisms that both happen to involve red light. Conflating them is the root of the confusion. Corn: So walk me through the sleep side. Why is six seventy not doing the work there? Herman: Because the photopigment that drives the circadian system is melanopsin, and melanopsin peaks around four hundred eighty nanometers. That's blue. The intrinsically photosensitive retinal ganglion cells that tell your brain it's daytime are tuned to short wavelengths. So what protects melatonin at night is the absence of short wavelengths, plus keeping the intensity low. Once you're out at the long end of the spectrum, the precise figure barely matters. Six twenty, six seventy, seven hundred. They're all doing the same job for sleep, which is simply not being blue. Corn: So the number is nearly irrelevant as a mechanism. But not as a signal. Herman: And that's the turn. A brand that prints a peak wavelength is a brand that knows what emitter it bought. Somebody measured that. Somebody bought binned LEDs from a real supplier. Somebody could produce a spectral chart if you asked. That's a different operation from the white-label reseller who bought whatever red bulbs were available and printed the word red on the box. Corn: Or worse, copied six seventy from a competitor without knowing what it means. Herman: Yes. And that's the thing. The number is a decent proxy for manufacturing seriousness while being nearly irrelevant as a mechanism. It's a correlated signal, not a causal ingredient. Corn: That distinction is the transferable lesson. It applies everywhere. A spec sheet mixes genuine mechanisms with prestige markers, and the skill is telling which is which. Sometimes a badge is just a badge. But sometimes a badge is a signal that someone cared enough to measure. Herman: So then the actual question Daniel's caller has. If six seventy isn't the spec to shop on, what is? Corn: Previous episodes settled that intensity matters more than colour. So we take that as given. The next axis is LED and product quality. Let's go through the spec sheet you should actually be reading. Herman: First thing. Is there a spectral power distribution chart published at all? Corn: The squiggly line graph. Herman: The squiggly line graph. And most listings don't have one. They have a photo of a red bulb and the words six seventy nanometers. But a real manufacturer publishes the chart. And once you have it, you look at two things. Peak wavelength, which is where the curve tops out. And full width at half maximum. That's how wide the peak is at half its height. A narrow peak means the emitter is actually doing what it claims. A broad hump means you're getting a lot of light you didn't ask for. Corn: So a tight peak at six seventy with a narrow full width at half maximum is a very different product from a broad hump that spills down toward the shorter wavelengths. Herman: And that spill matters. If your red light has a broad spectral output that reaches toward amber and orange, you're getting closer to the wavelengths that start to matter for melanopsin. Not dramatically, but the point of a spectral chart is you can see what you're actually getting. Corn: Let me push on that, because I think there's a subtlety here. How broad is broad? If someone's looking at a chart and the peak is at six seventy but there's a shoulder that extends down to six hundred, is that actually a problem in practice? Or is this one of those things where the difference is measurable but not meaningful? Herman: That's a fair question, and I think the honest answer is that it's a gradient, not a cliff. The melanopsin sensitivity curve drops off steeply as you move away from four eighty, but it doesn't hit zero at five fifty. There's still a tail. So a broad emitter that spills toward six hundred is delivering slightly more circadian stimulus than a narrow one that stays tight at six seventy. Whether that slight difference matters for sleep depends on how sensitive you are, how bright the light is, and how close it is to your face. For most people, at the dim levels you'd actually use in a bedroom, the difference is probably small. But the spectral chart tells you whether the manufacturer even thought about this. A narrow peak is a sign of intentional design. A broad hump is a sign of "we bought whatever was cheap." Corn: So it's less about the exact width and more about what the width reveals. Herman: It's the same badge logic. The chart itself is the signal. The details matter, but the fact that someone published them matters more. Corn: What about melanopic EDI? You mentioned that earlier. Herman: That's the metric that actually answers the sleep question. Melanopic equivalent daylight illuminance. It weights light by how strongly it stimulates the melanopsin pathway. So instead of asking what colour the light looks, you're asking how much circadian disruption it's actually delivering. A true red light has a melanopic EDI near zero. A warm white bulb at the same visual brightness has a much higher one. Corn: And almost nothing consumer-grade quotes it. Herman: Almost nothing. Which is itself informative. If a brand quotes melanopic EDI, or even knows what it is when you email them, that's a strong signal. It means they've thought about the actual mechanism, not just the marketing. Corn: So that's two things. Spectral chart, melanopic EDI. What's the third? Herman: The dimming floor. This is the most under-discussed spec for a bedside light, and I think it matters more than anything else on the list. Corn: Because you're using this in a dark bedroom, not trying to light a warehouse. Herman: Right. The maximum brightness is almost irrelevant. What matters is how low it goes. Does it drop to a faint glow, or does it bottom out at something that still feels like a nightlight on steroids? And does it step at the bottom of the range, or fade smoothly? A light that clicks from off to twenty percent is useless for the middle of the night. Corn: The usable minimum. That's the spec. Herman: And it's the one nobody prints. You have to dig through reviews or test it yourself. But it's where the product either works for this application or doesn't. Corn: I want to dwell on this for a second, because I think the dimming floor is where the rubber meets the road for actual human use. You've got someone who wakes up at three in the morning. They need to get to the bathroom or check on a kid. Their eyes are fully dark-adapted. If the light clicks on at even ten percent, it's blinding. It's like someone shining a torch in your face. So the question isn't just "does it dim low" but "does it dim low enough that a dark-adapted eye can tolerate it?" Herman: And that's a much lower level than most people realise. True dark adaptation is incredibly sensitive. A single candle in a dark room is a lot of light when your eyes have adjusted. So the dimming floor needs to go down to what feels like almost nothing. A faint ember. And most consumer lights don't get anywhere near that. They bottom out at a level that's fine for watching TV but useless for the middle of the night. Corn: Which is why the smoothness of the dimming curve matters too. If it goes from off to a faint glow in one click, that's still a shock. You want it to fade up gradually, so your eye has time to adjust. Herman: And that's a driver sophistication question. Cheap drivers step. Good drivers fade. The difference is in the electronics, not the emitter. Corn: What about flicker? You've talked before about drivers. Herman: Driver quality. This is where cheap lights fall apart exactly where you want to use them. Pulse width modulation. The driver dims the LED by switching it on and off very fast. At high brightness the switching is fast enough you never notice. At deep dim levels, cheap drivers slow down or get sloppy, and you get flicker. Sometimes visible, sometimes not. Even invisible flicker can cause eye strain and headaches in some people. Corn: And there's a practical test. Herman: The phone camera test. Point your phone camera at the light, dim it to the bottom of its range, and look at the screen. If you see bands rolling or the light strobing on the screen, the driver is struggling. It's not a perfect test, but it catches the worst offenders. You can also listen for driver whine. A faint high-pitched buzz at low dim levels is a cheap driver complaining. Corn: The driver whine is one of those things you don't notice until someone points it out, and then you can't unhear it. Herman: It's like the hum from a cheap wall charger. Once you know it's there, it's all you hear in a quiet room. And in a bedroom at night, quiet matters. A light that buzzes at you while you're trying to sleep is failing at its job, even if the light itself is fine. Corn: Spectral chart, dimming floor, flicker. What about the LEDs themselves? Herman: Binning. This is the one that separates a real manufacturer from an assembler. When LEDs come off the production line, they're not all identical. Even from the same wafer, there's variation in colour and brightness. Binning is the process of sorting them. A manufacturer who bins their LEDs is one who has a relationship with an emitter supplier, who tests output, who cares that two bulbs from the same box look the same. Corn: The failure mode is two bulbs from the same box that aren't the same colour. Herman: Which you've seen. Everyone's seen it. You buy a pack of bulbs, screw one in, it's warm and lovely. Screw the second in, it's slightly pink. Or slightly green. That's the binning problem. The manufacturer didn't sort, or bought from a supplier who didn't sort, and the variation shows up in your living room. Corn: A brand that prints six seventy is more likely to bin, because they've already told you they measured the emitter. Herman: That's the correlation. The badge implies the process. Not perfectly, but it's a useful heuristic. Corn: Then there's thermal management. You said wavelength shift. Herman: LEDs shift wavelength as junction temperature rises. A poorly heatsunk six seventy light may drift off its claimed peak as it warms up. The wavelength moves, the output changes, and over time the whole thing degrades faster. Lumen depreciation. That's the gradual dimming of an LED over its lifetime. Good thermal design slows both. It's not something you can see on a spec sheet easily, but it's why a light with a chunky metal housing and a decent heatsink costs more than a plastic bulb that's all emitter and no cooling. Corn: The thermal story is one of those invisible quality differences. You can't see it on the shelf. You can't see it in the listing photo. But it's the difference between a light that's still working at full spec in five years and one that's drifted off and dimmed out. Herman: The drift is insidious because it's gradual. You don't notice it day to day. But if you compared a new unit to a two-year-old unit side by side, you'd see the difference. The wavelength has shifted, the output has dropped, and the whole thing is just slightly off from what you paid for. Corn: The last thing. Direct narrowband red versus phosphor-converted amber. Herman: Two different ways to make red light. Direct narrowband red LEDs are exactly what they sound like. The semiconductor emits at the target wavelength. They're more efficient, more spectrally pure, and more expensive. Phosphor-converted amber uses a blue LED with a phosphor coating that shifts the output. Cheaper, broader in spectrum, and it ages differently. The phosphor degrades over time, so the colour shifts as the product gets older. Corn: For a sleep light, the narrowband is the better choice. But it's also the pricier one. Herman: That's where the honest note comes in. Above a certain point, the returns flatten. A decent amber bulb used consistently beats an optimal one used occasionally. The best spec sheet in the world doesn't help if the light stays in the drawer. Corn: Walk me through the actual shopping heuristic. You're on Amazon. What do you check? Herman: Four things. One, is there a published spectral power distribution chart, and does it show a narrow peak? Two, how low does the dimming actually go, and does it step or fade at the bottom? Three, do the phone camera flicker test when it arrives. Four, does the brand name its emitter supplier, or does it just say LED? Corn: The documentation tells you which kind of company you're dealing with. A real manufacturer publishes a spectral chart, quotes melanopic EDI or at least knows what it is when you ask, and names its emitter. A bad one prints six seventy with no chart, no binning information, no thermal specs. Herman: The badge without the evidence. Corn: Right. The number with nothing behind it. Hilbert: I sold halogen bulbs in Birmingham in the mid nineties. Wholesale. Before LEDs were a consumer thing. And we had the same binning problem. Retailers would order warm white and get boxes that ranged from candlelight to hospital corridor. Same part number, same supplier, completely different colour. It drove them mad. Hilbert: The binning wars, we called them. Nobody printed a wavelength on a halogen box. You got a wattage and a vague colour word. Warm white. Soft white. Daylight. Meaningless. The variation was enormous and nobody could do anything about it because the bulbs were made the way they were made. Hilbert: When I see six seventy on a modern spec sheet, I don't read it as a wavelength. I read it as a price signal. The brand is telling you it knows its audience is informed enough to look for that number, which means it can charge more. The badge isn't just a proxy for manufacturing seriousness. It's a proxy for who the manufacturer thinks you are. Hilbert: The funny thing is, the binning problem I saw with halogens in ninety four is now solved by the same brands that print six seventy. Because binning is expensive. Only manufacturers who care about their reputation bother. So the badge correlates with a quality culture, not just a quality process. Corn: That's a different angle. The badge as pricing. Herman: It fits though. A brand that prints six seventy is signalling to a buyer who's done some research. That buyer is willing to pay more. So the badge does double duty. It signals quality, and it signals that the brand knows you're the kind of person who looks for quality. Corn: Which means the badge itself is part of the product. You're paying for the measurement. Literally. Herman: That's fine, as long as you know it. The measurement is real. The binning is real. The quality culture is real. You're just paying for it. Corn: I want to pull on that thread a little more, because I think there's a version of this that's less benign. The badge as a way to extract a premium without actually delivering the goods. Someone prints six seventy on the box, copies the language from a competitor, and charges fifty percent more. The number is doing all the work, and the product underneath is the same cheap LED with the same cheap driver. Herman: That's the dark pattern version, and it absolutely exists. The badge is easy to copy. The process isn't. So you get counterfeit badges. And the only defence is the documentation. The spectral chart. The emitter supplier. The melanopic EDI. If the badge is there but the documentation isn't, you're paying for a number on a box. Hilbert: That's why I always say the question to ask is "what's behind the number?" Not "what is the number?" Anyone can print six seventy. Not everyone can show you the chart that proves it. Corn: That's the whole episode in one question. What's behind the number? Hilbert: I've got one of those lights now. The narrowband one. Bought it two years ago. It dims down to almost nothing, no flicker, no whine. Cost more than I wanted to spend. But it's still on the bedside table and it works every night. The cheap one I bought first died in six months. Hilbert: The driver went. Started flickering, then just stopped. So I bought the expensive one. And the expensive one is still there. That's the whole story. Corn: The cheap one died at the driver. That's exactly the failure pattern we were describing. Herman: The expensive one is still there because the thermal design and the driver were the things the extra money bought. Not the wavelength. The wavelength was the same on both boxes. Hilbert: Six seventy on both. One's in a landfill and one's on my bedside table. Corn: That's the badge. Same number, different product. Herman: The shopping heuristic distills down to this. Look for a published spectral chart with a narrow peak. Check the dimming floor. Do the phone camera flicker test at the bottom of the range. And check whether the brand names its emitter supplier or just says LED. Corn: The honest note. Above a certain point the returns flatten. A decent amber bulb used consistently beats an optimal one used occasionally. The best spec sheet in the world doesn't help if the light stays in the drawer. Herman: Which leaves the open question. The correlated signal versus causal ingredient distinction. Where else are you shopping on a badge that's actually a proxy for something else? Corn: That's the one to sit with. Thanks to Hilbert Flumingtop for producing. This has been My Weird Prompts. Email us at show at my weird prompts dot com. We'll be back soon.