Netflix ACES workflow schematic

[CharlesBoileau]

Hi to all,

Came across an interesting article this morning about how Netflix produces short movies to test its compressions. In the article there’s a schematic of their ACES workflow.

I have a question about what they call a “ACES GAM” (graded archival master) and the inv(RRT)/inv(Pulsar ODT) “step” they do before producing their master. (FYI this is done in #resolve)

Here is the article: https://medium.com/netflix-techblog/engineers-making-movies-aka-open-source-test-content-f21363ea3781

Would this step a simple deactivation of the ODT in Resolve? Or is it something more along the lines of what it describes literally?

I feel like it’s just simple step that brings back the grading into ACES-0.

Would anybody have insight on this?

I’m downloading the package of files as I write this.

Thanks!

[stobenkin (Steve Tobenkin)]

good find!

[sdyer (Scott Dyer)]

Interesting read, @CharlesBoileau. I too would be curious exactly what they are doing to get their ACES (GAM).

If they are grading on the ACES (NAM) data, then a simple deactivation of the RRT/ODT in Resolve would yield the ACES (GAM).

However, it is also possible that they are applying the “RRT + Pulsar ODT” step and then grading on the “Pulsar” code values rather than grading the ACES (NAM) values “under a LUT”. In other words, they might be grading output-referred data for a very high dynamic range output and then converting that back to ACES using inverse transforms to get their GAM. This would be instead of grading “under a LUT” as one would in a traditional ACES workflow, manipulating the ACES (NAM) values (or more likely, ACEScct) while viewing the content through the RRT+ODT.

Perhaps there are some Netflix engineers lurking here that could lend some insight?
It might also be possible to try to get a response through the comment section on the blog page.

[CharlesBoileau]

They probably have a SET of custom DCTLs to use “inverts” in their workflows. it’s just hard to “place” them properly in resolve unless you disable the ODT and use nodes (with DCTLs) to do the full workflow. It seems pretty hefty to use within Resolve. But hey this is Netflix and they have the ressources and knowledge to do so…

@sdyer: in what circumstance would someone want to grade the “ODT” code values. This would like grading “after a LUT” correct? I only to this in non aces workflows for very specific reasons (keys, saturation etc) but never as a full grading space.

[mdonato (Matthew Donato)]

Thanks for your feedback, @CharlesBoileau and @sdyer. For the color grading step, image manipulation was done on the ACES values, not values with the RRT and Pulsar ODT applied.

I’ve updated the workflow diagram on our TechBlog to address this confusion.

[sdyer (Scott Dyer)]

Hi @mdonato! Thank you for clarifying here and for the quick update to the image! Good to know we have ACES users at Netflix perusing the forum.

[sdyer (Scott Dyer)]

People should definitely click through the link to read the full context because it’s an interesting read.

But for convenience, here is the updated graphic. Dashed lines were added to clarify that the grading was done “under the RRT/ODT” for the Pulsar.

[CharlesBoileau]

Very cool! Looks like the 'typical" workflow!

@mdonato: do you ever export the “NAM” version of projects for archival purposes? Or would you go with the graded version?

Thanks!

[mdonato (Matthew Donato)]

@CharlesBoileau We archived the NAM, GAM and VDM for this test project. Every production is different.

[nick (Nick Shaw)]

@sdyer

But for convenience, here is the updated graphic.

Why is a dotted line shown returning from the Pulsar to Color Grading? Surely the Pulsar is the end of that path, which is monitoring only?

The only path back into the grading from the monitor is through the eyes and brain of the Colorist!

[CharlesBoileau]

@nick: My interpretation is that this just means the primary grade is done on the pulsar.

@mdonato: Thanks for the info. Good to know!