Weird Depth of field glitch

I usually work in 8 bits, but render in 16. The way that 32 bit is composited can look fundamentally different.

If you work in 8 bits, you’ll often see low-color depth artefacts and 16 will get rid of those.

Here’s a test that demonstrates the kind of cumulative errors that can occur when every compositing operation is calculated in 8 bits.

Switch to 8 bits. Make a 500×500 yellow solid and put a circular mask on it with a 100 pixel feather. Set the opacity of the layer to 5% and duplicate it 20 times. Ugly, right? Now switch to 16 bit.

Because the circle is only 5% opaque, the alpha only ranges from alpha 0-12 out of 255. That’s 13 steps. Each subsequent duplicate that you make will also only have 13 steps, so when you’ve stacked enough to make them visible, those 13 steps will be quite apparent. Go ahead – count ‘em.

When you switch to 16 bit, instead of a potential 255 steps per channel, you have 65,536. For the same 5% opaque circle, those 13 steps become 3277. You can try it, but you won’t be able to count them!

So it does make a big difference, particularly if you’re working with a lot of translucent layers or smooth gradients.

I can’t say I’ve ever noticed any motion blur artefacting at 8 bit, although I guess it’s possible under the right circumstances.

InlifeThrill said
I assume working in 16 bits by default wont do you any harm. Anyone has a good reason why 16bit instead of 8bit or vice versa? Some features do act strange in 16bits, however it solves a few problems when working with color in higher bits (the ever popular ramp effects at least..though there are various ways of dealing with ‘em in 8bit).

8 bit = faster / worse
16 bit = slower / better
32 bit = much slower / even better & different

As I said above, work in 8, render in 16.

Only use 32 bit where it’s really necessary, for instance where you want to use linear blending to mimic how light really interacts in the world (in 8 and 16 bit, compositing is done with gamma corrected footage- in 32 you have the option to remove gamma / composite / reapply gamma – otherwise known as linear blending)

32 bit is also good if you want to retain all your out of range values. For instance, where a color correction action might clip in 8 or 16 bit, in 32 the values will be retained and can be brought back down within legal range as a final color correction pass on the finished article.

Very informative. Thanks so much.

yeap, 16 bits solved it Thanks Felt

Felt, have you written a book yet? I would like to pre-order. Your knowledge is music to my ears

andrenavarre said
Felt, have you written a book yet? I would like to pre-order. Your knowledge is music to my ears

Not yet. There’s some tutorials here though.

felt_tips said

andrenavarre said
Felt, have you written a book yet? I would like to pre-order. Your knowledge is music to my ears

Not yet. There’s some tutorials here though.

https://vimeo.com/channels/aequicktips

You should open a “Felt_Tips Hotline”, charging $5 a minute etc. Hire some goblins and elves to take calls and clone yourself to answer every answer in the world 24/7. I bet there is a hole in the market there.

Massfocus said

felt_tips said

andrenavarre said
Felt, have you written a book yet? I would like to pre-order. Your knowledge is music to my ears

Not yet. There’s some tutorials here though.

https://vimeo.com/channels/aequicktips

You should open a “Felt_Tips Hotline”, charging $5 a minute etc. Hire some goblins and elves to take calls and clone yourself to answer every answer in the world 24/7. I bet there is a hole in the market there.

What did you eat for lunch buddy ?

felt_tips said

andrenavarre said
Felt, have you written a book yet? I would like to pre-order. Your knowledge is music to my ears

Not yet. There’s some tutorials here though.

https://vimeo.com/channels/aequicktips

Oh that’s really cool. Thanks for sharing. I had no clue you did that, it’s really great!