You're at a dinner table, third photo in. The ones from outside look fine. These look like everyone has been slow-roasted over applewood. Warm, amber, vaguely honeyed. Not ugly, exactly. But that's not what the room looked like.
So what changed between the sidewalk and the bread basket?
The sensor is starving, and it's making choices
In bright daylight, a phone camera has options. The sensor gets flooded with photons, the shutter can stay fast, and the ISO (the sensor's sensitivity setting) stays low, maybe 50 or 100. Low ISO means low noise. Low noise means the software doesn't need to intervene much, and what you see is roughly what the sensor caught.
Dim a restaurant to pendant-light levels and everything shifts. The sensor is now starving. To compensate, the camera cranks ISO, sometimes to 1600, 3200, or higher on older hardware. Higher ISO amplifies the signal. It amplifies the noise right along with it.
That's where the warming starts.
Raw sensor noise isn't evenly distributed across color channels. Blue and green channels pick up more noise than red, partly because silicon is more sensitive to longer wavelengths (the warm end of the spectrum) and partly because noise amplification hits shorter wavelengths harder. When the processing pipeline tries to clean up that noise, it suppresses the noisy blue channel more aggressively than the red. The result skews warm, the way a cheap scanner skews yellow when the bulb starts to die.
This isn't a bug anyone forgot to fix. It's a side effect baked into the tradeoff.
White balance is guessing in the dark
There's a second mechanism running in parallel, and it's arguably more significant.
Auto white balance (AWB) is your camera's attempt to figure out what color "white" actually is in a given scene, then shift the whole image so whites look neutral. In daylight, AWB has an easy job: sunlight sits around 5500 to 6500 Kelvin, and phone cameras have been trained on millions of outdoor shots. The algorithm recognizes the pattern fast.
Artificial light is a mess. Incandescent bulbs burn around 2700K. Warm LEDs cluster around 2700 to 3000K. Candlelight dips below 2000K. These sources aren't just dimmer than the sun; they're a fundamentally different color, and AWB has to make a judgment call about what the scene's "true" white is. In mixed or dim artificial light, it frequently misjudges toward warm.
Here's the part phone manufacturers don't advertise: they've quietly tuned their AWB algorithms to err warm in low light on purpose. Cooler corrections in dim scenes look clinical. Slightly unflattering. Warm corrections feel intimate and soft, so that aesthetic preference got baked into the firmware. It's a design opinion, not a law of physics.
Consider two people with the same phone model: Maya shoots mostly under warm indoor LEDs and finds her portraits consistently golden. Her colleague Dan shoots under office fluorescents, around 4000K with a spiky uneven spectrum, and gets cooler, slightly greenish results. Same phone, same software version, completely different output. AWB is reading the room, and the room keeps changing.
The warming isn't coming from one place
The common assumption is that warm low-light photos are a post-processing artifact, something layered on in software to flatter skin. That's only partly true. The warming starts in the physics of the sensor, gets amplified by noise reduction, then gets shaped further by AWB. It's a chain reaction, not a filter someone dragged a slider on.
Also worth knowing: Night Mode doesn't always fix this. It takes multiple exposures and merges them to reduce noise, which does help the blue channel recover somewhat. But the AWB guess still happens, and many Night Mode shots still run warm because the scene itself is lit by warm sources.
Want to see it directly? Open your camera, point it at a neutral gray surface under a warm lamp, and switch between auto and a manually locked daylight white balance. The gray goes from amber to near blue-gray. Neither is technically wrong. One is what the scene actually looked like; the other is what the camera decided you'd prefer.
Which raises the only question that actually matters here: how many of your "choices" in a photo were already made before you tapped the shutter? The physics push warm, the software leans into it, and the result is usually a photo people like. But a camera that flatters you without telling you is still a camera making editorial calls on your behalf. Knowing which decisions are yours is different from just liking the picture. Both are fine. Only one of them is true.