A friend told me that one of the disadvantages of working with 8bits images within Photoshop is that every time I add an adjustment layer I lose some data. Is that true?

Is there a quantifiable amount of information per layer, or is there a way I can see this in practice? I've tried to add a bunch of layers to an image but don't know how to see if any information was lost.

  • Even with 16-bit images or any other common digital representation of image data, you will 'lose' information after most adjustments you make to the data. More precisely: After applying an adjustment to the original data, there may not exist a further adjustment which can be applied to the result of the first step to get you back to the original data. Your question should however rather be, wether you lose relevant information or not. – jarnbjo Jan 21 at 12:35

Adjustment LAYERS are non-destructive in Photoshop. You can delete or disable them at any time and get back to where you started w/o any losses. Adjustment layers only add data to the file; as shown by the file size (document size can be selected in lower left of the UI).

The problem with 8bit vs 16bit is the accuracy of the math (digital edits are all mathematics)... i.e. something like 8÷2 doesn't require a lot of precision and 8bit would be fine; but 10÷3 requires a lot more precision and 16bit would be much better. And yes, you can/should edit jpegs in 16bit.

This picture shows two copies of a jpeg opened in PS, and both have the same levels adjustment applied with quite significant tonal value shifts. The top image is edited in 16bit and the bottom image is still in 8bit; notice the difference in the color histograms...

enter image description here

| improve this answer | |

There can be less color in the result, yes. In a pixel, each color channel can take only one of 256 values (2⁸). Each adjustment transforms these 256 values to some other 256 values. But there is a constraint: for any pair of two values of input, the highest value of the two must remain the highest (otherwise you get an effect called solarization). If you use the Curves, this means that your curve cannot go downwards from left to right.

Now, consider a value somewhere in the middle, for instance 128. Your adjustment moves it to 140. The 128 values below 128 are mapped to 128 new positions between 0 and 140 (of course this means that some values below 140 won't be used. And the 128 remaining values (128 and above) are now mapped into 116 values. So you started with 256 values, and now have 128+116=244 values.

This can lead to what is called a haircomb histogram:

enter image description here

The histogram above is a linear black-white gradient, in which, using Curves, I moved 128 to 140. The gaps on the left are the 12 values that are left empty (128 into 140) while the teeth on the right are result values that correspond to two input values (12 of them, 128 into 116).

Each such adjustment eats a bit of color. This not too visible in the beginning because you have three color channels and the missing values don't apply to the three channels at the same time (sometimes you can even get more colors...), but in flat areas with uniform colors you will get visible bands.

| improve this answer | |
  • 1
    this doesn't go into the difference between adjustments and adjustment layers. Whether the errors are cumulative with the latter or not would be interesting. – ths Jan 21 at 10:08
  • The question is about 8-bit images, not about adjustment and adjustment layers. And you can infer from the answer that the problem would be the same in both. – xenoid Jan 21 at 11:05
  • 1
    is it really? can i? – ths Jan 21 at 11:45

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.