White Balance: Fix a Colour Cast in a Photo
Two automatic methods and a manual one, with the estimate measured on mid-tones only so a bright sky cannot drag it, and the gains rescaled so correcting the colour does not quietly change the exposure.
Drop images here
or click to browse · or paste with Ctrl+V
JPG · PNG · WebP · GIF · BMP · AVIF
Method
Below 100 keeps some of the original cast, which often looks more natural than a fully neutral result.
Grey world assumes the scene averages out to neutral. On a photo that is genuinely mostly one colour, a forest or a red wall, that assumption is wrong and it will drain the colour. Use manual there.
Presets
Your photos are never uploaded. The estimate and the correction both run inside this browser tab.
People also use
Everything this tool does
Mid-tones only for grey world
Pixels below 30 or above 225 are excluded, because a white sky or a black shadow carries no information about the light. On a cast test image with a large white area that took the residual imbalance from 25 levels down to 1.
A percentile, not a maximum
Brightest neutral uses the 99th percentile of each channel. Injecting a handful of blown pixels into a test image shifted the maximum-based gains but left the percentile-based ones unchanged.
Exposure is not touched
All three gains are rescaled by one common factor so average luminance is preserved. On a cast image with headroom the mean luminance moved 0.042 levels out of 255; on a half-white image, where the correction has to clip, it gave up 4.6.
Manual temperature and tint
Both axes that matter, warm against cool and green against magenta, for the many photos where neither automatic assumption holds.
Strength blending
A fully neutral result is often not what you want on a sunset or candlelight. Sixty to eighty percent removes the wrongness and keeps the atmosphere.
Drag to compare
Before and after share one frame with a slider, which is the only honest way to judge a colour correction.
Six purpose-built presets
Auto, gentle auto, use the highlights, warm it up, cool it down and kill the green, covering the common lighting problems.
Batch with one ZIP
Correct a whole shoot with identical manual settings so the frames match each other, and take the set as one archive.
Transparency preserved
The alpha channel is untouched for PNG and WebP output.
Nothing is uploaded
No server, no queue, nothing to delete afterwards. Close the tab and every trace of the image goes with it.
Which method to use
The two automatic methods rest on different assumptions about your photograph, and each fails in a specific, predictable way. Knowing which assumption fits your picture is most of the skill.
| Method | Assumes | Works on | Fails on |
|---|---|---|---|
| Grey world | the scene averages to neutral | mixed scenes, people, streets | One dominant colour |
| Brightest neutral | the brightest area was white | photos with white or a highlight | No white anywhere |
| Manual | nothing | anything, including the two above | Needs your judgement |
| Strength 100 | full correction to neutral | obvious casts | Can look clinical |
| Strength 60-80 | keeps some of the original mood | sunsets, candlelight | Leaves a slight cast |
| Temp positive | warms toward orange | cold, blue photos | Overdone on skin |
| Tint positive | adds green | magenta casts, fluorescent light | Overdone on foliage |
3
correction methods
99th
percentile, not the max
0
files uploaded
0.04
levels of luminance drift
ZIP
batch download
Free
no account needed
FlipMyFormat vs other white balance tools
| Capability | FlipMyFormat | Typical free tool |
|---|---|---|
| Files stay in your browser | ||
| Estimate measured on mid-tones only | ||
| Brightest neutral uses a percentile, not one pixel | ||
| Average luminance preserved by the correction | ||
| Strength blending to keep some mood | ||
| Batch ZIP download without paying | ||
| No watermark on the output |
How to fix a colour cast without killing the mood
Automatic white balance is a guess dressed up as a calculation, and understanding the guess is what lets you tell when to trust it. The grey world method assumes that if you averaged every colour in a scene you would land on neutral grey, which is roughly true of a street, a room full of people or a mixed landscape. It is completely untrue of a forest, a photograph of a red brick wall, or a shot of the sea, and on those it will helpfully drain out the very colour that was the point of the picture. The brightest neutral method assumes instead that the brightest area was white, which is true whenever there is a white wall, a shirt, a cloud or a specular highlight, and false in a photo where the brightest thing is a yellow lamp. Neither assumption is better in general; they just fail differently, so look at your photo and pick the one whose assumption holds. The second thing worth knowing is that a fully neutral result is often not what you want. Golden hour is warm, candlelight is warm, an overcast beach is cool, and correcting those to neutral produces a technically accurate photograph that has lost the reason anyone took it. That is what the strength slider is for. Sixty to eighty percent removes the obvious wrongness and leaves the atmosphere, and it is where most good corrections land. Finally, if you shot a set under one light source, correct them as a batch with the same manual settings rather than letting each one estimate separately, or every frame will end up slightly different from its neighbours.
- 1
Add your photos
Drag them in, click to browse, or paste with Ctrl+V. Photos shot under the same light can be corrected together as a batch.
- 2
Try grey world first
It suits most mixed scenes. If your photo is dominated by one colour, switch to brightest neutral or set temperature and tint yourself.
- 3
Back the strength off, then export
A full correction can look clinical on a sunset or candlelight. Around 60 to 80 percent keeps the mood while removing the worst of the cast.
What the automatic modes actually measure
Mid-tones only, and a percentile not a maximum
Both automatic methods measure something and turn it into three channel gains, and both are easy to get subtly wrong. Grey world averages the picture, but a large white sky or a deep black shadow carries almost no information about the light and pulls that average toward neutral, which makes the tool under-correct exactly when the cast is most obvious. So the average here is taken over mid-tones only, excluding anything below 30 or above 225. On a test image with a blue cast and a large white area, including everything left an average channel imbalance of 25 levels behind, while measuring mid-tones only brought it down to 1. The brightest neutral method has a different trap: taking the literal maximum of each channel means one hot pixel or one specular glint sets your entire correction. Using the 99th percentile instead makes it stable, and we check that by injecting a handful of blown pixels into a test image, which shifted the maximum-based gains but left the percentile-based ones unchanged.
Correcting colour without changing exposure
Multiplying channels by gains has a side effect nobody asks for: it changes how bright the picture is. Correct a strong blue cast by pulling blue down and the whole image gets darker, which then looks like the tool has done something to your exposure as well as your colour. That is why all three gains are rescaled by a single common factor chosen so the average luminance of the result matches the original. Because it is one factor applied to all three channels, it changes nothing about the colour correction itself; it only removes the brightness side effect. On a cast test image with headroom in the highlights the mean luminance moved by 0.042 levels out of 255, which is as exact as eight bit arithmetic allows. The guarantee only weakens where a photo is already close to white, because there the correction has to push a channel past 255 and the excess is lost: on a test image that was half pure white, the same correction gave up 4.6 levels of average brightness. That is a limit of the output format rather than of the method, and it is worth knowing if you are correcting a high key photo.
What people use it for
Indoor photos under tungsten light
Household bulbs push everything orange, and this is the single most common cast people want removed.
Fluorescent and office lighting
Strip lights add a green cast that is very visible on skin. The tint slider is the direct fix.
Overcast and shade
Open shade is lit by blue sky, so photos taken in it come out cold. A warm correction restores skin tones.
Product photography
Accurate colour matters commercially, and a white background makes the brightest neutral method reliable.
Matching a set of photos
Correcting a shoot with the same manual settings keeps frames consistent, which automatic per-image estimates do not.
Scanned prints and negatives
Old prints shift colour as they age, and a mid-tone based estimate handles the resulting cast well.
Underwater and pool photos
Water absorbs red first, so these are heavily blue. A strong manual warm correction is usually needed.
Screenshots and documents
Photographs of paper under mixed light rarely look white until the cast is removed.
Six tips worth knowing
Grey world suits mixed scenes. If your photo is dominated by one colour, it will drain that colour, so use manual instead.
Brightest neutral works when there is genuinely something white in shot. If the brightest thing is a yellow lamp, it will fail.
A fully neutral result is often wrong. Sixty to eighty percent strength removes the cast and keeps the mood of golden hour or candlelight.
Correct a whole shoot with the same manual settings rather than letting each image estimate separately, or the frames will not match.
Temperature and tint are two different axes. Fluorescent light needs tint, not temperature, which is why warming it never quite works.
Fix white balance before adding any effects. Correcting colour after a filter has already shifted it is much harder.
Troubleshooting
Auto drained the colour from my photo
Grey world assumed the scene averages to neutral, and yours does not. A forest, a red wall or the sea breaks that assumption. Switch to manual.
The photo got darker after correcting
The gains are rescaled to preserve average luminance, and on most photos that holds to within a twentieth of a level. On a very bright, high key image the correction can push a channel past 255 and lose a little brightness, around 4.6 levels on a half-white test image. Lower the strength if it matters.
Brightest neutral made it worse
The brightest thing in your photo was probably not white. If it was a warm lamp or a coloured sign, that method has nothing correct to lock onto.
Skin looks lifeless now
Full neutral correction often does that. Drop the strength to around 70, or add a little warmth manually afterwards.
Frequently asked questions
What is white balance?
It is the correction that makes things that were white in real life look white in the photograph. Light sources have colour, so a white shirt under a household bulb records as orange until it is corrected.
What does grey world mean?
It assumes that if you averaged every colour in the scene you would get neutral grey, and works out the channel gains that make that true. It suits mixed scenes and fails on photos dominated by one colour.
Why does it only use mid-tones?
Because a large white sky or a deep black shadow carries almost no information about the light, yet drags the average toward neutral. On a cast test image with a large white area, using everything left 25 levels of imbalance; mid-tones only left 1.
What is the brightest neutral method?
It assumes the brightest area of the photo was white and scales the channels so it becomes white. It works whenever there is a white wall, a cloud or a highlight, and fails when the brightest thing is coloured.
Why a percentile rather than the brightest pixel?
Because the brightest pixel is one pixel. A single hot pixel or a specular glint would set your whole correction. The 99th percentile is stable, which we confirmed by injecting blown pixels that shifted maximum-based gains but not percentile-based ones.
Does correcting the colour change the brightness?
Not measurably, in most photos. All three gains are rescaled by one common factor so average luminance is preserved: on a cast test image with headroom it moved 0.042 levels out of 255. On a photo that is already close to white the correction has to push a channel past 255, and that excess is lost, which cost 4.6 levels on a half-white test image.
Which method should I use?
Look at your photo. If it has a good mix of colours, use grey world. If it has something genuinely white in it, use brightest neutral. If it is dominated by one colour, use manual.
Why did auto ruin my forest photo?
Because grey world assumed the average should be neutral, and a forest genuinely averages green. The tool removed the green to satisfy the assumption. That is the assumption failing, not a bug.
What is the difference between temperature and tint?
Temperature runs between blue and orange, which is the axis most light sources vary on. Tint runs between magenta and green, which is what fluorescent lighting shifts. They are independent.
Why would I use less than full strength?
Because a technically neutral photograph can lose its atmosphere. Golden hour is warm and candlelight is warm, and correcting that entirely away produces an accurate picture nobody wanted.
Can I correct a batch of photos together?
Yes, and you should if they were shot under the same light. Use manual settings so every frame gets the same correction, rather than letting each estimate separately and drift apart.
Will this recover a badly wrong colour photo?
It can fix a cast, but it cannot recover a channel that was clipped to nothing. An underwater photo with no red information recorded at all has limits no correction can pass.
Should I white balance before or after other edits?
Before. Correcting colour after a filter has already shifted it means you are fighting two casts at once.
Are my photos uploaded to a server?
No. The estimate and the correction both run in your browser using the canvas API. Nothing is sent anywhere.
Which format should I export?
Keep the original format for photographs. If the input was already JPEG, one more pass at quality 90 or above is fine.
Is transparency preserved?
Yes for PNG and WebP output. The alpha channel is untouched. JPG has no alpha, so transparent areas are filled with the background colour first.
Does this work like a camera raw white balance?
It works on the same principle but with less headroom. Raw files hold much more information than an eight bit JPEG, so a raw editor can push a correction further before colours start clipping.
Does it add a watermark or need an account?
No to both. No sign-up, no email wall, no daily limit and no watermark on the output.