Bit depth in photoshop misbehaving...

[Pixelcon]

Heya!

So I've been told by IT that my files are way too big and I need to do something about it. I decided to do a quick test looking at the possibility of working only in 8-bit to cut down on files sizes and am a little disappointed with the performance of a 16-bit image in photoshop.

I decided to do a side by side test with the same image saved out of vray in 8-bit, 16-bit and 32-bit and try some drastic exposure changes on each. Down 4 EV then back up again to see how the 8 and 16-bit files performed against the 32.

As expected the 32-bit file performed perfectly while the 16-bit seemed to suffer from clamping just as badly as the 8-bit did.

My next test was to try opening my 16-bit file and changing the photoshop document to 32-bit mode and trying again. To my surprise both the 16-bit exr and tif performed as well as the 32bit exr did.

So the question is, why does photoshop handle a 16 bit file as well as a 32 bit file when in 32 bit mode yet handle the same 16 bit file as badly as an 8 bit when in 16-bit mode? Basically why does photoshop handle 16bit files badly?

A pretty confusing question but I'm sure you guys get what I'm trying to ask...

Cheers for any advice!

[Pixelcon]

An interesting note: If you open up a 32bit image, change the document mode to 8bit, SAVE as psd, close, re-open 8bit psd, change the mode to 32bit you STILL retain the information of the image! What the??

[joconnell]

I could be totally wrong, but 8 bit in your case means that each of your channels only has 256 levels of bright to dark so it's only got a small level of subtlety. 16 bit has 65336 (from memory) levels of subtlety between dark and bright so it means that if you've got really gentle gradients and shading, you're less likely to see stepping or banding than in your 8 bit files. 16 bit is basically a better quality version of 8 bit. 32 bit on the other hand changes how the maths of a program work in most cases where we change to float colour. With 8 and 16 bit in photoshop, you're working in a limited range of brightness, between black and white essentially. If you colour correct your image up, your bright data will get pushed up over white, and get clipped off to white. From that point onward, photoshop considers the data lost so if you then try to expose back down, you won't get the same detail back, what'll happen is your blocky area of white in the burnt out areas will simply colour correct back down to a grey.

With 32 bit colour, The mathematics of photoshop change to a different type called "floating point" colour. The major difference with this is that nothing gets clipped as such. If we take the same example as before, you colour correct your image and your bright grey values again get overexposed and they'll be displayed on your monitor as a block area of white, purely since your monitor can't display any brighter. The big difference this time though, is that photo still keeps the data for the colours once they go above pure white, rather than just setting them all to a brightness of 1.0, like 8 bit or 16 bit colour would. It's pretty simple maths really, for example if you had a light grey that had a brightness of 0.75 on a scale with black being 0 and white being 1.0 and if you used a brightness or exposure in photoshop to double the brightness of your image that 0.75 grey value gets doubled to 1.5. Your monitor can only display as far as 1.0 white so that value will look the same as a clipped white, but photoshop still has the 1.5 value saved in the file. If you were to then do another brightness or exposure except bringing every down by half this time it'd divide the 1.5 value back down to 0.75 and your original detail gets restored. Your files will be a tiny bit bigger since they have to save data for areas that would have previously been clipped to white and easy to compress, but you'll retain far greater flexibility.

So in your case, if you open an 8 bit file, double it's brightness so it clips, then half the brightness it'll have thrown away data that it can't get back and you'll end up with blocky areas. If you do the same with a 16 bit image it'll do the exact same thing since with photoshop both 8 bit and 16 bit colour mode operate between normal black and white only. As soon as you step into 32 bit colour, whether that's by opening a 32 bit file which automatically gets treated in 32 bit by ps, or by opening an 8 / 16 bit file and then switching it to 32 bit colour, you're pretty much just removing the regular black to white limit that photoshop was using previously. You're working in a much wider space of numbers (in theory infinitely wide) and thus you're not going to get the data loss. Using a 32 bit image over an 8-bit or 16-bit image doesn't give you better quality automatically, it just forces photoshop to deal with the file in a much more flexible range of colour.

[Pixelcon]

Thanks so much for the detailed reply. I think the last paragraph is exactly what I was looking for. I think issue I had was that I assumed that 16-bit didn't suffer from clipping, which is does.
I think for my purposes I could just correctly expose a render (no clipping) and get away with working in 8-bit for my inhouse "everyday" stuff and move up to 32-bit for the bigger stuff.

Thanks again!

Cheers

[flino2004]

When you finish with your images even in 16bit, and all the layers are merged, if you flat the image the file size will be considerably reduced.... So your image will be under "background" layer.

[dejaVu]

Interesting post.
A useful workflow I have found is to externally Link your render files in to your comp using a handy free little extension. This makes PS work very similarly to AfterFX procedural workflow. You can then work in 32 bit and hopefully smaller file sizes.