Photograph the same white t-shirt under a kitchen bulb, then under overcast daylight, and run both photos through a palette extractor. You won't get the same palette. One will lean warm and slightly yellow, the other cooler and closer to true white. The shirt didn't change — the light did, and that's the whole story behind a color cast.
What a color cast is
Every light source has a color temperature, measured in kelvin. A candle sits around 1,900K and reads warm and orange. Midday sun sits closer to 5,500K and reads close to neutral. An overcast sky can push past 7,000K and read cool and blue. A camera has to guess which of these it's looking at and correct for it — that guess is called white balance, and when it's wrong (or simply left on "auto" in mixed lighting), the whole photo picks up a tint. That tint is the color cast, and it doesn't just affect the whites — it shifts every color in the frame a little toward the same hue.
Why this matters for an extracted palette
A palette extractor doesn't know the difference between "this product is genuinely cream-colored" and "this product is white under a warm bulb." It reads pixels, not intent. If a product photo has even a mild cast, every extracted color inherits a fraction of that tint — a true navy blazer can come back reading closer to slate, a white wall can extract as a pale beige. This is the single most common reason someone extracts a palette from a product photo and the resulting HEX codes don't match the product's actual, physical color.
It's rarely a dramatic shift. A cast is usually subtle enough that a person looking at the photo wouldn't immediately call it "wrong" — but subtle is exactly what throws off an automated extraction, because the tool has no reference point to correct against.
Spotting a cast before you extract
A few checks catch most cases. Look at anything in the photo that should be neutral gray or white — a wall, a piece of paper, a plain background — and ask whether it looks perfectly neutral or faintly tinted toward orange, blue, or green. Compare a product photo against the same item under a different light source, if you have one; a consistent shift across both tells you it's the light, not the object. And if a photo was shot indoors under a single bulb with no daylight mixed in, assume some warm cast by default — incandescent and most LED bulbs skew warm unless the camera corrected hard for it.
What to do about it
The cleanest fix happens before extraction: shoot (or re-shoot) reference photos in diffused daylight or under a light source close to 5,000–5,500K, which reads as closest to neutral for most cameras. If a re-shoot isn't possible, a basic white-balance correction in any photo editor — using a known-neutral point in the image as the reference — will pull most of the tint out before you run the extraction.
If neither is an option and you're extracting from a photo you didn't take and can't correct, treat the resulting palette as directionally right rather than exact. Use it to confirm a color family — "yes, this is in the navy range" — rather than to lock in a precise brand HEX code. For anything that needs to be pixel-accurate, like a genuine brand color match, go back to a physical reference (a swatch, a manufacturer's spec sheet) instead of trusting a single photo.
None of this makes photo-based extraction less useful — it just means the palette is only as neutral as the light it was taken in. Knowing that is usually enough to catch a bad extraction before it ends up in a stylesheet.