I mentioned this in passing on a forum today, and wanted to elaborate a little.
If you choose to edit your images in 16-bit mode, then of course you need to convert to 8-bit for printing at the end.
Your final steps for printing probably include:
> Flatten layers
> Crop/resize
> Convert to 8-bit
> Sharpen
> Save
Now, the order of some of those steps is flexible. You could resize before flattening, for example. Some people say it's important to sharpen while still in 16-bit, and they're probably right (I haven't checked).
The big thing I want to discuss right now is this:
You must flatten your layers BEFORE converting to 8-bit.
I can't emphasise this enough. If you convert your image to 8-bit while adjustment layers are still present and active, then you might as well have been working in 8-bit from the very beginning. All benefit of 16-bit is discarded.
Adjustment layers
must be committed to the image (by merging with the background layer) before conversion.
I'll try to show you what I mean ...
Here's a simple gradient I created in a 16-bit document:
Then I made a Levels adjustment layer, and made a very aggressive contrast adjustment to it. If this file had been 8-bit, the banding would have been visible. But because it's 16-bit, it's fine:
I'll zoom in to 200% so we can see a bit clearer. Remember, this is a gradient with an aggressive adjustment, but because it's 16-bit, it's very clean:
Ok, for this one, I flattened the layers first, then converted to 8-bit. Still looks as clean as a whistle:
But for this one, I converted to 8-bit, THEN flattened. This time we can see the banding/graininess that can occur when editing 8-bit images:
If you found yourself peering closely to try to notice that banding, you're not alone. I did too. It's there, but it's not awful. That's a huge adjustment for not much banding.
The fact is, 8-bit is much safer than some people would have you believe. You can do pretty awful things to an 8-bit file without causing problems that anybody would notice. I'll say what I've always said - as long as you make your major tonal adjustments in Raw, and just tweaks in Photoshop, 8-bit is absolutely fine*.
However, if you
do work in 16-bit, please heed my advice, and don't undo all your good work by converting too early.
* If you work in ProPhoto RGB, or LAB mode, 16-bit is an absolute necessity. 8-bit wide-gamut data can band if you even breathe on it. Your output process will include conversion to sRGB (or a print profile), which should occur after flattening but before bit depth conversion.