White balance

White balance corrects a colour cast by scaling the red, green and blue channels against each other, so light that came out orange or blue comes back to neutral. Click any spot in the photo that should be grey and the tool averages a 5-by-5 patch of the original pixels there, works out the three gains that make it neutral, and applies them to every pixel. There is also a grey-world auto button, temperature and tint sliders running from -100 to 100, and a strength slider that blends the correction back toward the original.

How it works

  1. Drop an image Drag a photo in. It is decoded and drawn on the canvas, ready to sample from.
  2. Click a spot that should be grey Click anywhere neutral — a white wall, a grey card, a road. The click averages a 5-by-5 patch of the original pixels, so clicking the same spot twice gives the same answer instead of compounding.
  3. Or run grey world Auto assumes the whole scene averages to grey. That holds on a mixed scene and falls over on a photo that is mostly one colour, such as grass or a red wall, which is why it is a button rather than the default.
  4. Nudge temperature, tint and strength Temperature runs cool and blue to warm and orange. Tint runs green to magenta, which is the axis fluorescent light gets wrong. Strength blends the whole correction back toward the original; at 0 nothing changes.
  5. Apply and download Pick keep, PNG, WebP or JPEG, then apply. A single file downloads directly; a batch comes back as a ZIP.

Questions

How does clicking one grey spot fix the whole photo?
The sampled colour is what should have been neutral, so the three gains that pull it to its own average brightness are the gains that neutralise the cast everywhere. The correction is scale-invariant: a dark grey and a light grey of the same hue ask for the same gains, so it moves colour without moving exposure.
Are the temperature numbers in Kelvin?
No. Temperature here is a relative warmth from -100 to 100, not an absolute colour temperature, because the light a photograph was taken under is unknown to the tool and a Kelvin figure would be a guess dressed up as a measurement. Both axes are exponential, so -40 exactly undoes +40 and lands back on the pixels you started from.
What are the limitations?
One correction goes on every file in the batch, so photographs shot under different light need a pass each. Gains are capped at 8×, which stops a click on a near-black pixel from amplifying sensor noise into colour, and a reference that is clipped to pure white has already lost its colour, so the correction made from it is a guess.

Every tool