Negative film is stored in three emulsion layers (cyan, yellow, and magenta) on a base. Each layer has its own separate grain structure. What is Green on positive film is Magenta on negative film, and so on. Here are some graphics:



Most of the grain is in the colour layers, and each layer has its own grain structure hence "varying degree of granularity". Additionally some layers are physically thicker than others. Here's a good view of a photographic film:



Motion-picture film (negative):





Positive motion-picture film:





Dye layers are in a different order, due to using slower developing dyes/longer exposure.

Quote:Still don't get the negative/positive thing... don't we apply the grain plate (positive) to digital video (positive)? Why negative should be implied?

If we split the positive digital image to RGB we get the emulation layers that would exist on the film negative:

Digital Blue -> Yellow Emulsion Layer
Digital Green -> Magenta Emulsion Layer
Digital Red -> Cyan Emulsion Layer

Therefore applying grain to blue applies it to yellow and so on. As already mentioned positive film has a finer grain than negative film.

Well looks like there are no hide tags in this forum. I took apart a 4K scan supplied in Prores format. I did not resize the image, this is at the original 4K resolution with a small section of the frame cropped. This is what I found:









Manipulating it to negative first didn't make a difference. For positive film we do need to shift the hue, however it will clip the non-primary colours if done in RGB mode due to limitations of the rec709 colorspace. I guess the next best thing we can do in avisynth is apply the positive grain 1/2 per channel. We know it's the yellow channel that is the grainiest (which remains true if I were to shift the hue 60 and output the channels as above to represent positive film), and the other two are about the same. So I'm thinking use a grainplate with blotchier grain for the yellow channels (16mm or even 8mm) with a higher intensity. There are some free grainplates here. They're all at 4K though, we can downscale to 2K with ffmpeg like so:

ffmpeg -i "Fuji 16mm - 4K.mov" -c:v prores_ks -profile:v 3 -qscale:v 10 -vendor ap10 -pix_fmt yuv422p10le -s 2048x1080 "Fuji 16mm - 2K.mov"

Following your experiments...

This is fascinating and I'm really looking forward to poita explaining this properly, from a technical stand-point.

I see how you guys are dissecting it and trying to figure it out and I think it's awesome

Yes I'm looking forward to hearing from poita. Here's an example he might be familiar with seeing:






This time it's clear that Magenta is grainier than Cyan.

The most grain is always in the blue layer, when doing noise reduction you can often get away with just processing the blue layer by itself.
Our eyes are also least receptive to resolution of the blue layer, so you can soften it more than other layers before our eyes/brain will perceive a difference.

On old CRT projectors, they used to a a switch to blur the blue signal on purpose. You would focus each Red, Green and Blue CRT until it was really sharp, and then flick the switch to blur the blue channel for normal viewing.

I'm still doing a bit of a write up, but most of what has been written here is mostly correct. To do re-graining correctly you need to apply it by colour layer, but also separately for low, middle and high exposure areas as they grain is different depending on exposure levels. This is easy enough to do in something like Resolve, I'll see if I can make up a generic process tree for it.

(2017-03-12, 01:16 AM)poita Wrote: To do re-graining correctly you need to apply it by colour layer, but also separately for low, middle and high exposure areas as they grain is different depending on exposure levels. This is easy enough to do in something like Resolve, I'll see if I can make up a generic process tree for it.

That will be great! Thanks poita

I really want to try this when you get something written up.

Sweet, I'm going to install Resolve soon. I'd love also to know which film grains we should use? Or your opinions on them:

http://www.cineticstudios.com/blog/2015/...otage.html
https://tdcat.com/downloads/filmgrain/

etc. When you have the time of course!

I'll use Film Convert. It also have different grain size depending on the original film stock used by the production. You just pick the one you want. You can also just apply the grain instead of the whole film stock colors. https://www.filmconvert.com