Just because I ask my hot dog to be a cheeseburger doesn’t make it a cheeseburger.
Guess what? Not everybody knows what the fuck they are doing. Including me, sometimes. Okay, most of the time.
2 Likes
We deliver the exrs in AP0 because that is what you are supposed to do as that is the transport format for ACES. We do all our work in AP1 as that is the working colourspace format for ACES.
That’s why ACEScg, the cg rendering linear & compositing format is AP1. It is why ACEScc and ACEScct, the grading formats are AP1. There is no AP0 working format for a reason.
You are NOT supposed to work in AP0.
2 Likes
Generally the show LUTs will be made in the debayered colourspace of the main camera which will include a colourspace transform to Rec709. To apply the LUTs from 2065-1 their needs to be a colourspace transform to the camera native colourspace before applying the LUTSs.
If you were creating a show look in ACES colourspace, you wouldn’t be creating a LUT.
I explained above. Because you can’t monitor what is happening by outside the AP1 gamut. Anything beyond that is going to be clipped. You could have noise in keys and renders that you can’t see. You can create out of gamut values as well once you go back to grade in AP1 (as that’s what an ACES grade happens in) which can cause issues, especially in renders. We have had to reject other VFX vendors work due to their being technical issues and it has become apparent that they worked in AP0 which has introduced them (can look like coloured noise).
This plain and simply bad practice to work in AP0.
1 Like
I should add that In a full ACES colour pipe you would create the look with a LMT and not a LUT.
AP0 primaries are meant to enclose the entire CIE 1931 color set. Thus the largest possible color space in this set of color science.
ACES2065-1 uses AP0 primaries and linear gamma. It’s not a log color space.
One other point: 16bit vs. 32bit float has nothing to do with clipped whites. Both 16f and 32f can represent any type of white we should ever see in a camera file. The difference between 16f and 32f is precision. The exponent is at least 5 bit, but the mantissa is too small in 16f where you can experience banding in certain cirumstances.
So if grade sends it back because whites are clipped, it shouldn’t be on behalf of the encoding, but possibly on behalf of a processing step in the pipeline.
1 Like
Adam isn’t one who will just waffle on I suggest you conduct yourself in a respectful manner on this forum or perhaps use a real name instead of hiding behind mux
It’s your tone man, and u have been setting all of this up for a response. We can all discuss stuff respectfully man
1 Like
I should explain myself better. Colours will be clipped in the end monitor if they are unable to display what the numbers are. So the numbers aren’t clipped but what you are seeing is. A bit like if you have numbers above 1000 nits in HDR on a display like a domestic OLED. Anything above around 700 nits bar the newest models would be clipped at 700 nits.
You can call bullshit on what I said earlier if you want but on shots we have rejected and they changed their pipe to ACEScg then fixed the issue. I don’t understand technically why that could happen (likely some kind of rounding error somewhere) but it fixed it. Call it waffle if you will but I don’t really care
I’m sorry but it is you that likely needs to do more research. We have answered your questions but if you don’t like the answers then so be it.
No worries, that was in response to a different statement. I should have quoted to avoid misunderstanding.
I get it man sometimes the tone can be misinterpreted, but also takes 2 to tango and you are trolling Randy now
I called your saying that studios wanted you to work in AP0 as bullshit. I was not saying you are bullshit. And I have taken no offence to the waffling statement either so no need to apologise.
They are NOT asking for it. I look at the actual spec sheets from Netflix, Disney, Amazon, Warners as a small example and not once has any of them said to work in AP0. Any tech documents says quite the opposite. You said you do as you’re told but who is telling you because it isn’t the studios.
And I’m done
1 Like
So I think we can agree that AP0 is a much wider color spectrum, because of the different color primaries.
Secondly, we work in 16bit float or 32bit float, which means each color channel will be divided between the white point coordinate and the channel primary. As a result, the further the distance, the same amount precision, makes for a coarser numeric representation.
Given that, any processing algorithm may produce different results in AP0 (whether visually perceptible by eye or not), than will impact the end result.
Thus switching from ACES2065-1 to ACEScg can result in fixing problems, by applying the smaller primaries, therefore allocating the precision of the data the algorithm allocates to the area that has most impact on the end result.
Goes back to that spec maximalism doesn’t always yield superior results, unless you really understand the tradeoffs.
It’s reasonable to store files in AP0 to preserve the original information. It’s not necessarily reasonable to do all your processing in AP0 unless you have verified that all your tools are capable of doing their job without degradation.
That test should be done as a null test, not a visual inspection.
As an illustration - we had a conversation two days ago on the Discord whether it’s possible to use ACEScct with ProRes instead of ACEScg in EXR, to avoid to have to deal with large image sequences. The argument was that it may be plausible. But a quick null test showed that there was significant quality loss even without modifying the viewer parameters.
Not opening this ACEScct argument, but showcasing how we should approach this question based on data, and not believes.
As my old finance department said: In God We Trust, Everyone Else Bring Data.
Research? I’m providing links from the maker of ACES itself and the maker of the tools, Autodesk.
And don’t you dare bring waffles into this. Make it about pancakes or else.
But is that the streaming service, or someone in the chain?
I think the question is whether Netflix itself is asking for this, and do so in their delivery specs. Or are there other people in the chain that may just be too eager to overdeliver on specs with haste leading to this confusion.
1 Like
they were very strict on security so I don’t want to name names. it was a few projects all for one company. I guess given where this has gone I should clarify that my interpretation of it was that any rendered out file should be AP0 if it needed to be passed from one tool to the other, not that every process inside the box needed to be run in AP0, and most of the time I did color manage to cct because the nature of the work was better done in log. should have been more clear!
3 Likes
I don’t have detailed insight into negative values to speak with authority on that aspect.
Logically, to have a negative value, you need a color that is outside of your primaries. That is possible in AP1 since it’s not all of CIE 1931. It’s not possible in AP0 because that’s all the color there is in this way of representing colors.
But getting negative colors at the end of your pipeline would be a sign of bad processing. Should be clamped somewhere, or shouldn’t be allowed in intermediate steps unless there is value.
But working AP0 because you’re worried about negative values, is like throwing a boulder at a bird, and the baby out with the bath water. You’re accepting loss of precision to avoid a poorly formed process that doesn’t protect against invalid values. Not a successful strategy.
Ya know your last paragraph contradicts your claim, don’t ya?
2 Likes