Introduction

Yes, this is me again. Back with the same topic. Since my last post from September 2021, I have learned a few things that I would like to share. What is this about again ? Well, it comes with many names :

• Display Transform
• View Transform
• Output Transform
• Display Rendering Transform (DRT)
• Tone-Mapping
• Lookup table (LUT)
• Print Film Emulation (PFE)

In this article I will use a different name which I think is more appropriate : Picture Formation.

The scene-referred/display-referred workflow has proven to be successful for the past twenty years. But interestingly enough, this workflow seems to have overlooked an important step in our chain. Where does the picture get formed ?

The lack of proper terminology to describe images is one of the biggest issue in our industry. And this is exactly why we need to name this “mechanism” properly. Let’s have a look at a simple diagram :

The picture formation is NOT just “putting data on the screen” (like a display encoding), it is building up aesthetic decisions about creative image appearance. In my experience, it will have the biggest impact on the quality of your project (positively or negatively). Hence the obsession.

Anyone who cares for their art seeks the essence of their own technique.
— Dziga Vertov

Why does it matter ?

We spend more than 8 hours a day in front of our monitors “staring” at images but we barely ask ourselves how those work. Just think about it for a minute. Most artists (and even supervisors) believe that displaying their work in a viewer is “automatic”.

This could not be further from the truth. Let’s put it simply : no one on the planet has figured out this yet. There have been several attempts, more or less successful and that’s about it. And our winner so far, the “apex predator”, is the chemical film processing developed last century.

And just to be super clear, we do not paint pixels in CG workflows. This is NOT like Photoshop where you pick a color and it displays “automatically”. That would be true for textures because their range fit between 0 and 1. But not for CG renders where our lights and the Global Illumination (GI) allow for more complex scenarios and much higher values.

Before going any deeper into our analysis, I just want to emphasize that choosing the proper picture formation is going to be the most important decision that you can take on a show. Because every single artistic note that you will give will be dependent of it :

• Evaluate the roughness of a material.
• Evaluate the color of a fur shader.
• Evaluate the correct exposure of a shot.
• Evaluate the lens flare and noise of a delivery.

So, the better you take this decision, the easier your production will be. It is as simple as that.

This article is based on my experience with full CG shows. And its goal is to provide a guide to help supervisors choosing the “best” color management workflow for their project. It applies mostly to feature animation and video games, but the fundamentals will also be true for any VFX work.

Our monitors are cubes

Most websites (including mine) about color management start their explanation looking at “stimuli data” (inputs and/or renders). But this time, I would like to do it the other way around. We will start from the “end target” (our displays) and then go backwards.

Because we look at everything (shots, assets, renders, playblasts…) through a screen !

Basically a monitor can be represented as a cube:

• A cube has 6 faces, right ?
• So imagine 3 channels going from 0 to 100% emission for R, G and B.
• Then picture 3 more channels from 0 to 100% emission for C, M and Y.
• Finally visualize one corner being (0,0,0) and one corner being (1,1,1).

This eventually forms a cube. That is your monitor.

Now pay attention, this is important. No matter how good, expensive or fancy your monitor is, it only can display a range from 0 to 100%. That’s it ! A cube !

There is no magic : a monitor comes with its own physical limitations.

Let’s start simple

At this point, you may want to tell me : “Chris, you are making things complicated again. I just wanna display an exr render on my monitor. Nothing more.” Sure ! Let’s take an exr “HDR” render and display it.

Here comes the first question : why does it look wrong (or dark) ?

This is because our monitors apply an “EOTF”. The simplest explanation for an EOTF is that it applies like a “gamma down” (around 0.45~ish in most cases). This is part of the hardware and there is nothing we can do about it. So we need to compensate for that in our softwares. Here is the result :

And here comes the second question : why does it still look wrong ?

Well, we are displaying a “scene-linear” exr with a simple “Gamma 2.2 function” (EOTF-1), which means that any value above 1 will get just clipped. Because remember :

• Our exr files have values way above 1 (technically at 16-bit half float, it can go up to 65504~ish).
• Our monitors only display a range from 0 to 1. That’s it !

Hopefully, we all can see the issue we are trying to solve at this point. Somehow we need an operation that allows for a faithful representation of “high dynamic range data” onto a “standard dynamic range monitor”.

Maybe now you want to add : “Chris, I know about this. We need to use some kind of s-curve to display linear exrs properly on a monitor.” Well, this is partly true. Let me quote Cinematic Color here :

“Friends don’t let friends view scene-linear imagery without an “S-shaped” view transform.”
— Jeremy Selan

So let’s look at our image through the “spi_anim” config created by Sony Pictures Animation (2010) and used on Cloudy with a chance of meatballs, Surf’s Up, Arthur Christmas and My little Pony :

And here comes the third question : why does it still look wrong ?

Several years have passed since Cinematic Color (2012) was written. And we now know that the “S-curve” is just ONE of the different elements necessary for a proper picture formation. Our image has now more contrast but why doesn’t it feel… “filmic” somehow ?

Well how would our light sabers look with the “Filmic” config ? This picture formation has been developed by Troy Sobotka (2017) and was used on Next Gen and The witness.

And here comes the fourth question : why does it still look wrong ?

Well, we have made some progress. The light sabers are now white and the contrast seems consistent with our previous attempt. But somehow our characters do not have any shaping. It looks like all our values are collapsing somehow, like a complete loss of tonality.

Okay, why don’t we try one last time ? This time I will use a new picture formation released in 2025, let’s call it “Candidate A” :

And here comes our final question : what do YOU think of this image ?

Now, if you are a supervisor, can you imagine the amount of notes you would have given to the artist if the image would not have been formed properly ? This is what is at stake here.

For clarity, let’s make a summary :

• We have seen that our monitors are inherently limited devices.
• We have displayed THE SAME EXR five times : “None”, “Gamma 2.2″, “spi-anim”, “Filmic Very High Contrast” and “Candidate A”.
• We have seen that a simple gamma function is not enough to display properly an HDRI.
• We have observed that different picture formations will give different results. For better or for worse.

If you agree with the above points, let’s move forward.

If you are familiar with my previous post, you should not have learned anything new yet. So far this is mostly a recap.

What about ACES 1.X ?

Possibly after reading this, you may want to point out : “Chris, why do you re-invent the weel again ? ACES has been implemented in most DCC softwares providing the industry a standard for color management.“

First, just a reminder that I already wrote about ACES 1.X flaws and reported back to the Academy in 2021. And I am not the only one since an important feedback about ACES 1.0.3 called “ACES Retrospective and Enhancements” was published in 2017. 

Also, I agree that ACES has mainly brought three good things to the industry:

• It has brought attention to the importance of color management.
• It has been an entry point to learn about color management (via their forum mainly).
• It has given an agreed exchange color space (ACES2065-1) between studios.

Basically, I would not be writing those lines without ACES. So thank you.

So why don’t we look at our light sabers with an ACES 1.X config (created by the AMPAS in 2016 and used on Super Mario, Migration and The Wild Robot) ?

My main comment here is that our colors have shifted. The light sabers appear violet in the back, orange on Mery and yellow on our beloved zombie. We will see later why hue shifts are necessary, but the issue here is that they are a byproduct of the transform. You cannot escape them.

Maybe at this point, you want to reply : “Chris, this render is clearly using sRGB-BT.709 primaries. Why don’t use ACEScg primaries ?” Sure, let’s do that.

It is fascinating to me how this ACES image looks similar to the one formed with Filmic. Sure, they both used different stimuli data but this shows that we are back to square one. We haven’t solved anything (yet).

Please check carefully these two images above and then ask yourself : “Do these examples look like they are color managed ?” Or just put differently : “What does color management even mean at this point ?“

Anticipating some feedback

I can already anticipate three reactions to this article so far. Let’s see if we can address those concerns. The first one is : “Chris, your light sabers example is an edge case. Not all exr files have this kind of extreme data.“

I would argue the following :

• There are no edge cases. All we have are exr files with “light data” in them. And even on a project with a “natural” look, I have faced plenty of situations where the picture formation is being pushed to its limits. You don’t need light sabers nor fancy neon lights to see the possible limitations.
• Another way to put it is that “edge cases” are extremely important because they tell the full story. They just make obvious the potential flaws of the picture formation and help us to spot more easily the patterns.

Here is an interesting image formed on stimuli data made of “wavelengths”. I think this is the best illustration of what I am trying to express above :

The second argument I often hear is : “Chris, a colorist would fix those artifacts. That is part of their job.” For instance :

I think sometimes people are getting too fixated on every single image coming out beautifully when it’s untouched through the transform. Because, you know, we’ve got a colorist in the loop here and there are possibilities to do all sorts of other things to images that may be problematic.
— A color specialist

I would reply the following :

• Sometimes it looks like we have given up on forming proper pictures. Our goal should be to come up with a “good” picture formation that allows colorists to focus on their creative goals. Not duct taping.
• Also, can you imagine if in a review with a director I would say something like : “A colorist will fix it in DI” ? I would probably get fired. In full CG, we expect the images that we deliver to look “correct”.

For instance, I can replicate the exact same demo as above with an exr file from “The Grinch”. And we could ask ourselves : “Is the light bulb an edge case ? Should a colorist fix the light bulb ?“

The third argument would be : “Chris, I know you can show images where stuff looks clearly broken. BUT there are obviously many shows and projects being delivered under the ACES 1.X standard and I can’t say I am noticing some crazy problems anywhere. “

I would answer the following :

• I already explained in 2021 : most movies that claim to be ACES projects actually have their plates encoded in ACES 2065-1 and their own picture formation (not the ACES Output Transforms).
• And for the projects that use them, here are some of the most extreme examples I was able to find (from The Wild Robot) :

A feedback I am ready to accept at this point is the following : “Chris, you have mentioned a few times stuff looking good. But what does good even mean ? “

Well now we’re talking. This is exactly what I want to answer here, by proposing objective criteria rather than some subjective “golden eye” like it is too often the case. So let’s move forward.

My Picture formation requirements

First, it is important to acknowledge that there is no perfect picture formation. It is a game of compromises. If you optimize for one thing, you sacrifice other things. For instance, you may favor saturation over “smoothness” or an accurate inverse instead of a “pleasing” picture.

It took me several years to find these requirements. They are the result of comparing hundreds of images through different picture formations and discussing them online. Before diving in, I will share some tips to evaluate them properly :

• Test as much as possible on a great variety of footage. The broader are your samples, the better.
• Try increasing the exposure a lot. It is very important to push the limits.
• Try grading. All our work is being done under the picture formation, so we shall not fight it.
• I never test invertibility since we do not need it in full CG workflow. That makes things easier for us.
• It is critical to constantly compare with other LUTs. Because we have to fight visual adaptation.

Visual adaptation means that if you stare at an image for long enough, it will eventually start to look good. This is because of our visual system constantly adapting to what we are looking at.

It shall not break visual cognition

This my main requirement. On an animated feature, where hundreds of artists work, you cannot afford a picture formation that breaks visual cognition. Because you will be constantly fighting it.

Think of images that you like (from movies or TV shows). I bet that they all have one point in common : they do not break visual cognition.

Defining visual cognition is not an easy task : I already tried once and failed. So let me start with the simplest example I can think of : a gradient (or ramp).

If we can all agree that a gradient should look like a gradient, then maybe we can move forward. And yes I can picture you saying : “But Chris, you’ve gone mad. What the heck are you talking about ? Of course a ramp should look like a ramp !“

Let’s say I want to make a blue gradient for a sky in a shot. Something like that (this is “Candidate A”) :

This gradient goes from (0, 0, 0) to (0, 0, 20) using a BT.709 blue primary. Now let’s have a look at the same gradient with another picture formation released in 2025 (let’s call it “Candidate B” for now) :

If your display is correctly calibrated, you should see a line more or less in the middle of the frame. A disruption. Not ideal for a gradient though, right ? Now think of all the gradients we have to deal with in CG :

• Sky ? A gradient.
• Specular roughness ? A gradient.
• Glossy reflection ? A gradient.
• Glow ? A gradient.
• Light decay ? A gradient.
• Blocker falloff ? A gradient.
• Volumetric anisotropy ? A gradient.
• Spotlight cone ? A gradient.
• Subsurface ? A gradient.

You might think : “Why are we spending so much time on ramps ? All of this is kind of obvious…” Well, you cannot imagine the effort that it takes to get there. Because a color “specialist” might come with these different explanations in a debate :

• But a colorist can fix it.
• But the BT.709 blue primary is an edge case : a sky would never have that color.
• But does this picture formation invert perfectly ?

Similar to the gradient images, I will share another simple example : a sphere should look like a sphere. Hopefully we can agree on this as well !

In the examples above, the spheres are pure blue, red and yellow BT.709 chromaticities. The plane is a green BT.709 primary. And if you pay attention, the spheres have a better shaping with “Candidate A”.

I personally like to use very simple examples to get to the essence of the issue. They also make the facets we are trying to identify more obvious than with much more complex stimuli data. Visual cognition is so hard to understand and explain that I think the best approach is to keep it as simple as we can.

And if you like some mind-bending ideas, these examples of ramp and spheres are in the end the same. Isn’t a ramp just a cylinder seen from a top view ?

It shall preserve tonality

At this point, I can imagine some feedback : “But Chris, it is easy. You are trading chrominance for luminance. Candidate A looks less saturated and probably cannot reach the display primaries.“

And my answer would be : “Yes, you are totally right. Remember about how I talked about compromises earlier ? That is the main one in my opinion : in a picture formation, tonality cannot be compromised.” And if reaching the display primaries is an obstacle to it, well this is a trade-off I am willing to accept.

Do you remember the previous light sabers examples ? How some images had a complete loss of tonality ? Back in February 2021, Jed Smith shared a visualization in Nuke that makes this issue easy to understand. I think it is worth sharing again :

This example would probably trigger the following comment : “But Chris, if we cannot reach the display primaries, we are loosing saturation. Isn’t that an issue ?“

We already discussed what “tone” meant in my previous post and I won’t go over that again. But all I can say is that we do NOT need to reach the display gamut primaries (the corners of the cube we have seen earlier) to get a colorful image.

For example, the video game below uses AgX as a picture formation (which does not reach the display primaries) and it does look colorful :

And believe me, it has taken me so much time to take that leap. I come from an animation background where our images have like the most outrageous saturation. But that is not the right way to go : tonality is a key component to avoid cognitive dissonant images. Otherwise why would we call it “Tone Mapping” ?

I will finish this paragraph with one more example. I think it will allow to wrap nicely the two points I made above about visual cognition and tonality :

Here is a recap :

• Those spheres have ACEScg primaries and secondaries.
• “Candidate A” preserves much more tonality than “Candidate B”.
• The red and yellow spheres with “Candidate B” do not read as spheres.

And if you still have doubts, I have added a grayscale image because that is our “reference” for tonality. When you remove colors, 99% of our issues are basically gone. Our only ground truth are grayscale images. So compare carefully the two images and ask yourself : “Which one is closer to our ground truth ?“

Because our main issue here is that no one has defined exactly what tonality or “tone” means ? Is it luminance, brightness, lightness, brilliance or value ? What is the actual metric ?

An explanation that really helped me moving forward is that chrominance and luminance are on the same plane. So there must be some kind of trade-off. And trying to reach the corners of the display gamut should NOT be prioritized over forming “pleasing” images.

It shall be smooth

This requirement is very similar and related to the previous two, but I think it is worth spending a bit of time on it.

The footage above has massively helped me to evaluate different picture formations. It is a render of the Eisko Louise model. I tried to light this scene in the most holistic way possible with spot lights and volumetrics.

And just to be clear, yes, this is the same as the ramp test but in a CG render. Hopefully it shows even more clearly the possible issues if the picture formation is not solid enough.

The most obvious difference is of course the light sources themselves, where a disruption of tonality has broken the visual cognition. In the red example, we can even spot some Mach bands around the light source.

But I would suggest you also compare carefully the forehead, cheeks, lips and chin of Louise (anywhere where the light impacts directly) between both candidates. Stunning, right ?

It shall not pass g0 threshold

Before explaining this requirement, we should define what g0 threshold is. Interestingly enough, this requirement was mentioned in the ACES 2.0 document workspace :

highlights shall desaturate at a certain point for low-saturation things and less so for items that are bright and saturated (e.g. neons, car taillights, lightsabers, etc.) – (how do we determine the threshold? – is this purely subjective? can we make it objective?)
— Output Transforms Architecture VWG

Even if the wording is not super accurate, the idea is there. When you increase the exposure of an image, reflective objects will look emissive if the picture formation is not robust enough. This means that you have passed the g0 threshold.

This concept originally comes from Ralph M. Evans who was a physicist who worked at the Eastman Kodak Company.

As the luminance is increased from G0, lightness and hue continue to strengthen and a new perception appears in the central stimulus that […] can best be described […] as though it were fluorescent.”
— Ralph M. Evans quoted in this pdf from the Munsell Color Science Laboratory

An easy way to test this is to have a look at different stripes of exposure of an image. This is how I understood why g0 threshold is so important and that the “path to maximal brightness” is a key part of a picture formation.

If you look carefully at the legs of our Lego sailor, you will see that they look emissive with “Candidate B”. When this figurine clearly does not have any emission in its material ! And yes, there is a thin line between the two images but I have found this threshold to be a very good indicator of the quality of a picture formation.

It looks unnatural if something which is clearly a reflective object suddenly looks like as if it would emit color. This is a perception based threshold named g0 by Evans.
— Daniele Sigarusano in this video

Same observation as above. If you look at the bike and the side-car, they appear emissive with “Candidate B”. This is kind of an issue for reflective (or in this case metallic) objects.

Remember, there are NO edge cases. So take advantage of these exposure stripes and compare carefully your footage. Exposure stripes are incredibly helpful to evaluate a picture formation (as you can see in Liam Collod’s picture lab website).

The g0 threshold could possibly be re-stated as “a plausible purity/exposure ratio”, something that has been referred to as “pictorial exposure”.

It shall not break polarity

This requirement is a bit difficult to explain but it is related to g0. It can be defined like this : “at equal energy, a color cannot appear brighter than its achromatic corresponding value.” Basically, the chromatic strength cannot overcome the achromatic intensity. Because a picture is worth a thousand words :

If you really pay attention to the highlight area, you should see like a dark ring around the specular with “Candidate B”. Which is basically… another disruption of a gradient ! Just like if the red color of the sphere exceeded the white specular in intensity.

Maybe at this point you are thinking : “Wow, Chris really lost it. We are nitpicking the specular of a red sphere…” But I can promise you that all of these requirements are important and will help you to improve the quality of your work.

If you want to know more about the topic of polarity, you might want to dig in this thread.

It shall not be chromaticity linear

And since our last examples were some explosion and fire, this is the perfect transition to our next requirement.

I will go straight to the point : hue-path bendings are necessary (I used to call them “hue shifts”). Yes, I can hear you scream at me : “But Chris, what do you say ? I set a color in my scene and another one shows up on my screen ?” Short answer is : “Yes, you need this. You ABSOLUTELY need this”.

Let’s rollback a bit. Back in November 2020, I was pointing out an issue about the hue skews in ACES 1.X. And almost five years later, it seems I have completely changed my mind. I understand the confusion. Let me try to make a summary :

• We definitely need carefully engineered hue-path bendings in the picture formation.
• Issue with ACES 1.X is that they are just an accidental consequence of the transform.

Just like our previous images, you may notice that fires get a “salmon” color with “Candidate B” :

And the truth is that NO ONE has figured out the perceptual hue paths on the planet. So our best option for now is a system that allows some possible tweaks.

This diagram from Wikipedia shows the Purity-on-hue in CIE 1931 chromaticity diagram. As you can see, there is no consensus on the matter. It is also worth pointing out that these studies measure simple flat field stimuli and make generalized assumptions based on those measurements.

They say nothing about complex stimuli like images.

And as a side question, can you imagine the complexity of what we are trying to address ? This diagram is a top view of an actual 3d model where each color has to go at a certain rate down a certain path… Insane.

It shall respect the air material (e.g. atmosphere)

This is another requirement that is difficult to explain and it is also evolves around the idea of g0 and polarity. But basically something absolutely fascinating to observe is images with volumetrics such as as haze, smoke or mist.

If the picture formation has not been properly engineered, you may see objects punching through the atmosphere in unexpected ways.

Because volumetric effects are basically an addition (or offset), it becomes quite noticeable if the mechanics at play do not respect that. If you have trees lit by the sun with some kind of atmosphere on your project, I highly recommend to check them carefully. You might be surprised by the results…

It shall fit the cube

Remember how we discussed that our monitors are basically cubes at the beginning of this article ? Well, a good test is to check how your formed image fits them (or not). It allows you to visualize if there are any kinks or disruptions, if it reaches the corners, if the ramps are smooth…

I think this is an interesting and complementary way to check that things go as planned but it cannot substitute actual viewing of images.

This was the last requirement on my list. As you can see, those are pretty much oriented towards forming a pleasing image. This is actually my main goal. Of course, there are other factors at play such as the SDR/HDR behaviour and the support of wide gamut color spaces.

But I will say it one last time : none of them should go against our main requirement of crafting beautiful images because we are… image makers !

What are our options then ?

Since 2017, we have learned a lot about picture formations and new color management workflows have emerged. I have tried to list the main ones :

Name	Author	Date	Comments
ARRI K1S1	Harald Brendel	2011	THE most used LUT on the planet (ARRI Alexa workflow).
ACES 1.0	AMPAS	2016	Color Encoding System developed by the Academy.
Filmic	Troy Sobotka	2017	Original Blender Color Management used on this movie.
RED IPP2	Graeme Natress	2017	RED Image Processing Pipeline explained here.
Sony Venice	Sony / Picture Shop	2022	LUT files made in partnership with Picture Shop.
ARRI Reveal	Sean Cooper	2022	The new ARRI Alexa35 workflow described here.
Tony	Tomasz Stachowiak	2023	A cool-headed display transform.
AgX Blender	Eary Chow	2023	Blender Color Management used on this video game.
TCAMv3	Daniele Siragusano	2024	Baselight Color Management Workflow explained here.
AgX SB2383	Troy Sobotka	2024	Minimal AgX OCIO Config using Linear BT.709.
JP-2499	JP Zambrano	2024	A popular picture formation pipeline described here.
ACES 2.0	AMPAS	2025	Color Encoding System developed by the Academy.
Open DRT	Jed Smith	2025	State-of-the-art Color Management Workflow.

They are all available through OCIO and they all have their pros/cons. But hopefully this article has given you the keys to evaluate them properly.

I almost can hear you ask : “But hey, Chris, are you going to leave us in the dark ? What about Candidate A and Candidate B ?” Well, let’s put it that way : in 2025, two picture formations got released and given that I recommend openDRT… I will let you guess which is which.

But what about OpenColorIO ? OCIO is “just” a container. Basically OCIO is the piece of software that allows us to load LUTs easily in our DCC softwares. What really makes a difference is the “content” (what you put inside OCIO), not the container.

The ACES topic

I guess a valid question from you would be : “Hey Chris, what is your take on ACES ?” I have been thinking about it for quite a while and I will try to explain as best as I can :

• Having a color management standard implemented everywhere is great. It gives artists with little knowledge something to start with.
• ACES was actually my entry point into this wonderful world of color management and I have met the most amazing people on their forum and in the meetings.

But here is the issue :

• Because it is the only open-source standard out there and it comes from the “Academy Of Motion Pictures Arts and Sciences” (AMPAS), no one questions it.
• And because color management is still a black box to most of the VFX industry, artists just blindly accept that it’s great and use it without knowing the problems.

How many times have I heard : we want to use ACES because it is a standard.

But somehow, never ever anyone asks : “what does this standard give us ?” (Troy Sobotka actually tried but no one answered him). Just stop and think about if for a minute. Do you really think that one unique picture formation can be used in the whole industry ?

I will share again this quote that really broadened my perspective on this matter :

I cannot understand why anybody would like to limit all of their productions to use the same output transform. It would be the same as limiting productions to use a single camera. Documentaries, features, animation, hand drawn, all of them have their unique challenges. Do you think film would have flourished in the last 100 years if the Academy would have standardised the chemical receipt ? Instead the Academy standardised the transport mechanism. The 35mm perf. And this was exactly the right thing to do. People could innovate and interchange.
— Daniele Siragusano

There is a risk of leveling down in quality if the entire industry uses the same color management. And because ACES comes from the AMPAS, the appeal to authority is very real. How many artists, executives and supervisors out there think that ACES is the only valid and safe solution ?

And this image below (from a Renderman video) is the perfect illustration of the complete gap between the marketing and the reality. None of these titles are actually present in this image but in this post-truth era, who actually looks at this image and cares ?

There was a proposal for ACES 2.0 to open the system to other Output Transforms (called “Meta Framework“) and it got rejected by the Technical Advisory Council (TAC) and… Netflix. It was interesting to follow this topic because it raised all sorts of questions such as “if a project only uses one component of ACES, is it still considered an ACES project ?“

And why do you think Netflix pushed back on it ? They did a lot of advocacy about ACES because it is easier to deal with hundreds of vendors by using ONE standard. Because it is cheaper. Indirectly ACES has become an imposed norm to reduce costs. And because it only allows ACES Output Transforms, it limits creativity and innovation.

Finally, let me ask you this : why would you use ACES when there are better options out there ? YOU should aim at the BEST color management workflow for your project, not necessarily a “standard”.

You will never meet a DP, a colorist or a DIT that recommends ACES 1.X.

ACES 2.0 Opinion

I will just share some final thoughts about ACES 2.0. I have nothing but respect for the incredible team that met for more than four years (from December, 2nd 2020 to March, 12th 2025) during 184 meetings. Those are top-notch color specialists and experienced supervisors.

But they were given an impossible task : to design a fully invertible picture formation that should able to reach the display primaries and form “pleasing” images out-of-the box. Unfortunately those requirements were contradictory and they set the group down the wrong path.

We tried to warn them several times and share our concerns. But they were so many parties involved that it became impossible to make our voices hear. Somehow the group lost sight of what should have been our main goal : crafting beautiful pictures for the centuries to come.

I can even point out the three meeting that deviated the group from their noble quest :

• Meeting #26, September 15th 2021 : the meta-framework proposal was discussed.
• Meeting #27, September 29th 2021 : post-TAC meeting, Netflix pushes back on the meta-framework.
• Meeting #29, October 27th 2021 : CAMs are introduced to the group because “ACES is science“.

That was the turning point where the group got derailed. Because ACES tries to be everything to everyone, it just ends up being an over-complex and average picture formation (a “jack-of-all-trades“).

Conclusion

In this article we have seen that picture formations are clearly connected to visual cognition. I know that my examples might go against certain trends in the industry but I have tried to express them as clearly as I could. At this point I would like to share one final quote that helped me to get through this :

I know that most men, including those at ease with problems of the greatest complexity, can seldom accept even the simplest and most obvious truth if it be such as would oblige them to admit the falsity of conclusions which they have delighted in explaining to colleagues, which they have proudly taught to others, and which they have woven, thread by thread, into the fabric of their lives.
— Leo Tolstoy

In summary, we discussed about how we shape the image data and what cognitive weirdness results if the data is shaped wrongly :

• If data slams into the display cube, you get gradient disruptions.
• If data doesn’t trend downwards as purity increases, you get g0 threshold disruptions.
• If purity doesn’t get compressed, there are disruptions in “surface and atmosphere perception”.
• If data doesn’t bend as intensity increases, colors don’t look like the right colors anymore !

And if you’re still not convinced by the examples I shared because they are too simple or in full CG, I will share below some live-action footage where we can spot similar issues. Hopefully now you have the proper terminology and insight to spot and name these issues accordingly. Thanks for reading !

Acknowledgements

I would not have been able to write this article without the help of these truly one-in-a-kind amazing people :

• Troy Sobotka (aka “the idea factory”, who is behind 90% of the ideas presented here)
• Jed Smith (aka “the stubborn builder”, who built openDRT based on those principles)
• Zach Lewis (aka “swiss army man”, who patiently supported us during the whole process)

I would like also to mention the ACES 2.0 Virtual Working Group that gave us an unique opportunity to learn and share : Alex Fry, Nick Shaw and Kevin Wheatley.

Sources

• Exr Files footage (encoded in linear-egamut) / ACES 2.0 first meeting
• sRGB piece-wise EOTF vs pure gamma / Output Transform Tone Scale / List of different modules
• Gamut Mapping Approach / Getting to the Display / ZCAM for Nuke / ACES 2.0 ranting
• ACES 2.0 CAM DRT / ACES Issues and Terminology / ACES 2.0 HSV RRT / Look Transforms
• ACES 2.0 Meeting of August, 23rd 2023 / ACES 2.0 Chroma Compression / ACES 2.0 Transparency
• ACES 2.0 Output Rendering / ACES 2.0 grading colorful lights / ACES 2.0 DRT Testing
• ACES 2.0 Debatings CAMs / ACES 2.0 Polarity / ACES Critics / ACES Background Document
• ACES Additive Mixture Video / ACES 2.0 Reflections and Experiments / ACES 2.0 JzAzBz
• ACES short films / ACES Filmic Reproduction / ACES Federated Output Transforms
• ACES Gamut Mapping Approach / Why grading in log instead of linear ?
• Linear-sRGB / Tone response curve / footage darker / ACEScg workflow
• openDRT documentation / openDRT withdrawal / On Brightness
• Steve Yedlin’s website / Debunking HDR / Debate about Steve Yedlin’s demo
• GDC HDR Grading in Frostbite / ACES in RenderMan / What the fuck is a picture ?
• Spectral film simulations / spectral film lut / Color Mentor podcast with Troy and Part 2
• The Line of Apelles / AgX development thread / Alex Tardif’s about per-channel
• Liam Collod’s picture lab website / ACES 2.0 webinar by Edi Walger (in Spanish)
• What Netflix Has Done to Movies w/ Will Tavlin / Casual viewing
• Picture Formation Definition / Troy’s Mastodon about picture formation
• Why Color Studies Are So Powerful