To be clear, I'm referring to images that have a color depth of 32-bits per channel.
Has anyone ever used images this large in a workflow – other than as RAW files directly from a camera?
If so, why did you need that much depth?
What benefits are there to using 32bit color depth?
32 bit per channel is most likely a floating point image. Tough it could also be 32 bit integers but i have never seen such images. Both TIF, EXR and some proprietary film production formats can use 32 bit float per channel inside them.
There are several reasons for using higher order bit depths. First of the image might not be a photograph. It might be some scientific or cartographic data. If it is a photograph its probably a HDRI image that has much wider color gamut than normal images Also HDRI is not 0-1 range but potentially 0-quite much more than 1000 in range).
Admittedly these data orders are not really for graphics designers except maybe HDRI conditioning. The reason for using these higher order data is that you might be processing the data further.
This often comes up in 3d graphics as you want to derive real lightning conditions form your images. Reflections depend on colors that over exceed the human gamut to look proper (like the sky reflection on cars). So what you really want your images to do is encode the energy levels. Which is not possible with the kind of standard images we commonly use as they are individually exposed. But what if you want to expose things after complex computation?
It also comes up in post processing of 3d graphics where you might be doing the final computation and you need the normal to be you know normal. So you might need to have negative colors at your disposal. Images are much more than what you think sometimes people use them to compute stuff like new lightning conditions on the fly.
There are also many engineering applications where it comes up, like game development, FEM calculation. It also comes up in measurement scenarios in engineering. You might also be dong some mathematical operations like FFT or Laplace transforms and then back and you dont want too loose all that much in the process so you'd up the sample levels. This is cool if you need to do very specific noise removal
Yes I have, I have also used 64 bit floating point images because i was getting nasty noise levels with 32 bit floats. Admittedly 32 bits is a bit overkill for most situations.
Not all output formats are 8 bit, printing in certain film substrate is often 12-16 bits also movies often use higher bit depths than 8 bit. But for traditional web/ print design its not much of thing. Then again you might get asked to design film titles (at least you wish this was true).
It's important to realize that the term "32bit" can mean different things to different people. Context plays a large role.
32 bit color depth is an aspect of photography. Photographers use 16 or 32 bit color depth to store as much possible color data per channel as they can. Doing this allows photos to be adjusted and retouched with much more precision, clarity, and prevents banding or "broken pixels" during the retouching stage.
Think of it like working on a logo in PNG form you snagged from a client's website versus using the vector version of the logo. 32bit color is to photography what vector is to design in that it's desired to achieve the best results possible.
In terms of design, 32bit color depth doesn't really play a large role. In order to reproduce any image it has to be an 8bit color depth image.
Designers aren't accustomed to "32bit" meaning color depth because, as far as I'm aware, nothing reproduces 32bit color depth images. For reproduction, everything needs to be down sampled to 8bit color depth - print or web. So, for designers 24bit often refers to a standard RGB image (8R 8G 8B) and "32bit" often refers to an RGBA image (8R 8G 8B 8A) or an 8bit CMYK image (8C 8M 8Y 8K). But that's not the same as 32bit color. This is where it starts to get confusing. :)
A "designer's" workflow doesn't often utilize 32bit color depth unless they are doing photo correcting and retouching. In which case, the folks over at https://photo.stackexchange.com/ will be able to assist far more with workflow suggestions and questions than many deisgners will be.
All that being posted, as a designer if you are creating raster images, you are absolutely free to create them as 32bit color depth images. As pointed out by @MichaelSchumacher in the comments below, using a higher color depth will assist in smoother gradients or blurs or anything where subtle variations are desired. In the end, you will still have to down sample to 8bit color depth to reproduce, but while creating you may find some advantages to working with higher color depth.
32 bits per channel are used to store HDR image information, and it is considered a floating point number.
I found the source: https://www.hdrsoft.com/resources/dri.html
The only time I use higher than 8 bits per pixel color (16/32) is when I want to do a gradient and I want to eliminate banding. For smooth grades, 16bpp has a lot of values to work with so you don't get a noticeable stepping pattern.
When downsampling to 8bpp, the values are dithered and the result is usually smoother than possible with 8bpp. The downside is you have to commit the gradient and redo it to make a change.
Side note: I wish TV/cable etc would increase bpp instead of adding more pixels (4K etc). Fades and wipes look like someone blew smoke on the screen with all the banding.
