16 bit float and 32 bit float doubts

Question

I'm starting in digital paint and I cant figure if it would be better to work in a 16 or 32 floating point bit rate depth, or even neither?

I've noticed that working in a floating point precision is important for HDR paint, but apart from it I'm still not sure if there exist any advantage. And most importantly, any advantage in sRGB color space work.

conclusion:

This discussion was somewhat vague as I myself had near zero experience with digital painting.

Particularly by being "transparent" about how things are done Krita works in a way completely different from Photoshop (and it is good, you get to learn a lot! Maybe it will cut a lot of work needed to get familiar with weird color mixing and "magic").

Particularly I answered a question about painting here that can answer a lot of doubts:

https://graphicdesign.stackexchange.com/a/107632/42515

Going to the point... 16 bits may matter (and some editors may hide it from you), linear color spaces can be really great and they need 16 bit, also it produces much less banding due to math on smudging paint methods to mimic wet techniques. But clearly it mostly depends on the result you expect while painting.

Answer 1

Notations of color depth

There are 2 ways to describe color depth. An older total bits per pixel, and a newer colors per channel.

The older model describes total memory usage per pixel. The only time this model is generally used is when talking of indexed color formats or monitor color mode. This should be considered legacy as its nomenclature clashes with the newer one. tough be honest only 8 bit and 32 bit color is confusable with he per pixel variant. the 32 bit color is the reason why this model has been deprecated, because 32 bit color can not actually be displayed as one of the color channels is alpha which isn't a color at all.

Now for composing of images its a bit misleading to talk of total numbers of bits per pixel (its ok for traditional image formats like jpeg and png tough). Since you have possibly many layers and many channels. So saying the total bit amount says little if you dont know how many channels your using. So instead its been decided to talk how many bits each channel eats. This causes a bit of confusion as the most common formats are (in increasing rarity):

• 8-bit integer per channel (often just known as 8-bit)
• 16-bit integer per channel (often just known as 16-bit)
• half float (or 16 bit float)
• float (or a 32 bit float)

do you need higher bitness?

Depends on what you do. In general if you need to ask you do not need this info (see What benefits are there to workflows using Images at 32-bits per channel? for a alternate answer).

For photography, 3d arts, animation, and science it can be useful. However the floating point format brings in its own convolutions. For painting the more than full color values bring minimal additional info. Your also going to limit your options quite much if you do.

So i would say its not useful for somebody designing for print or web today wont see much of benefit over a 16 bit integer image, even so its a bit overkill and 8 bits is enough to almost all work. With one exception you need to integrate photographic elements that may need some serious (and i do mean serious) color correction afterward.

Answer 2

For expanded color and dynamic ranges, start with 32 bit color. This will produce the highest quality rendering - 4,294,967,296 possibly colors including alpha channels for 32 bit, compared to 65,536 with 16 bit.

Depending on where the final asset will end up, I recommend down sampling it to 16 bit to save file space. 32 bit files are huge in comparison and cannot even be exported in many common formats.

sRGB was created for digital screens, so if that is where it will end up, that is the best color space and most common period.

Also, googling these topics will yield and encyclopedia's amount of info. If you have a very targeted question, please comment below.

Hope that helps!