How To - Dolby Vision

JOHN DARO November 8, 2022

Dolby Vision - How To and Best Practices

What is Dolby Vision

Dolby Vision is a way to dynamically map HDR to different display targets. At its core, the system analyzes your media and transforms it to ranges less than your mastering target, typically SDR 100 nits.

Project Setup

The first step is to license your machine. Once that is in place you need to set up your project. Go into settings and set your CMU(Content Mapping Unit) version. Back in the day, we used an external box, but nowadays the software does it internally. You will also need to set which version. V4 is the current iteration whereas V2.9 is a legacy version that some older TVs use. Finally set your Mastering Display. That is a Sony x300 in my case which is set up for PQ P3 D65 1000 nits.

It’s not what you said, it’s your tone

The goal is to make our HDR master look good on an SDR display. To do this we need to tone map our HDR ranges to the corresponding SDR ranges. This is a nonlinear relationship and our shadows mid-tones and highlights will land in the wrong areas if we don’t tone map them first. See below for an example of an SDR image that has not been tone mapped correctly. You can see the highlights are way too hot. Now we could use a curve and shape our image to a discreet master for SDR, but most studios and streamers are requesting a Dolby delivery regardless if a separate SDR grade was made. Plus, Dolby does a pretty decent job of getting you there quickly since the v4 release.

The first step is to analyze your footage. This will result in three values that will set a tone curve, min, max, and average. These values inform the system how to shape the curve to get a reasonable rendering of your HDR master in SDR.

What we are trying to do here is fit 10 pounds of chicken into an 8-pound bag. Something has to give, usually the bones but the goal is to keep as much chicken as you can. Rather than toss data out, we instead compress it. The system calculates the min, max, and average light levels. The idea is to keep your average or the “meat and potatoes” of your shot intact while compressing the top and bottom ranges. The end result is an SDR image that resembles your HDR only flatter.

How a colorist goes about the analysis is just as important as the analysis itself. This is going to get into a religious debate more than a technical one and everything from this point on is my opinion based on my experiences with the tech. Probably not what Dolby would say.

The original design of the system wanted you to analyze every shot independently. The problem with this approach is it can take a consistent grade and make it inconsistent depending on the content. Say you had two shots from the same scene.

One side of the coverage was shooting the character with a blown-out window behind them. The other side shoots into the darker part of the house. Now even though you as a colorist have balanced them to taste, the Dolby analysis will have two very different values for these shots. To get around this, I find it is better to average the analysis for each scene vs doing independent shots. The first colorist I saw work this way was my good friend and mentor Walter Volpatto. He went toe to toe with Dolby because his work was getting QC rejections based on his method. He would analyze only a grey ramp with the d-min and d-max values representing his media and apply that to his entire timeline. His thought process was if it was one transform to HDR it should be one transform down.

Most studio QC operations now accept this approach as valid metadata (Thank you, Wally!) While I agree with his thought process, I tend to work based on one analysis per scene. Resolve has this functionality built in. When I’m working in Baselight I set it up this way and copy the scene averaged analysis to every shot in preparation for the trim.

Setting the tone

Now that your analysis is complete it’s time to trim. First, you need to set what display output your trim is targeting and the metadata flag for the intended distribution. You can also set any masking that was used so the analysis doesn’t calculate the black letterbox pixels. The most common targets are 709 100nits, P3 48nits, and PQ 108nits. The 709 trim is for SDR home distribution whereas the other two are for theatrical distribution. The reason we want to keep the home video and cinema trims separate is that displays that fall in between two trim targets will be interpolated. You can see that the theatrical 108nit trim is very close to the home video 100nit trim. These two trims will be represented very differently due to the theatrical grade being intended for a dark theater vs home viewing with dim surround lighting conditions. Luckily Dolby recognized this and that is why we have separation of church and state now. The process for completing these trims is the same though, only the target changes.

Trim the fat

Saturation plus lift gamma gain is the name of the game. You also have advanced tools for highlight clipping and mid-tone contrast. Additionally, you have very basic secondary controls to manipulate the hue and saturation of the six vectors.

These secondary controls are very useful when you have extremely saturated colors that are on the boundaries of your gamut. I hope Dolby releases a way to only target the very saturated color values instead of the whole range of a particular vector, but for now, these controls are all we have.

Another tool that affects the analysis data but could be considered a trim is the mid-tone offset. A good way to think about this tool is a manual shifting of what your average is. This slides the curve up or down from the midpoint.

I usually find the base analysis and subsequent standard conversion a little thick for my taste. I start by finding a pleasing trim value that works for a majority of shots. Then I ripple that as a starting place and trim from there until I’m happy with the system’s output. The below before and after shows the standard analysis output vs where I ended up with the trim values engaged.

anal_only_elf03_r01_ocn_W1863889_tdx_sv119_10k_141626.jpg

It’s time to export once you are happy with the trims for all of your needed outputs. This is done by exporting the XML recipes that when paired with your PQ master will create all the derivative versions.

XML

Here are two screenshots of where to find the XML export options in Baselight and Resolve.

Rightclick on your timeline -> timelines - >export -> Dolby XML

Shots View -> Gear Icon ->Export Dolby Vision Metadata… This will open a menu to let you choose your location and set primaries for the file.

The key here is to make sure that you are exporting an XML that reflects your deliverable, not your DSM. For example, I typically export PQ P3 D65 tiffs as the graded master files. These are then taken into Transkoder, placed into a rec 2020 container, and married with the XML to create an IMF. It’s important to export a rec2020 XML instead of a P3 one so that when it is applied to your deliverable it yields the intended results. You can always open your XML in a text editor if you are unsure of your declared primaries. I have included a screen grab of what the XML should look like for the Rec2020 primaries on the left and P3 primaries on right. Always go by the numbers because filenames can lie.

There is beauty in the simplicity of this system. Studios and streamers love the fact there is only one serviceable master. As a colorist, I love the fact that when there is a QC fix you only need to update one set of files and sometimes the XML. That’s a whole lot better than in the height of the 3D craze where you could have up to 12 different masters and that is not even counting the international versions. I remember finishing Katie Perry’s “Part of Me” in 36 different versions. So in retrospect, Dolby did us all a great service by transmuting all of those versions we used to painstakingly create into one manageable XML sidecar file.

Thanks for reading

I bet in the future these trim passes end up going the way of the 4x3 version. Especially with the fantastic HDR displays available from Samsung, Sony, and LG at continually lower price points. Remember the Dolby system only helps you at home if it is something other than what the media was mastered at. Until then, I hope this helps.

Learn more

Check out this Dolby PDF for more information and deeper dive into the definition of the various XML levels. As always thanks for reading.

-JD

How to - Convert 29.97i SD to 23.98p in Resolve

JOHN DARO October 6, 2021

29.97i to 23.98p

Have you ever needed to convert older standard-definition footage from 29.97i to 23.98p for use in a new project? Resolve is no Alchemist, but it can get the job done in a pinch. Read below for the best practices approach to converting your footage for use in new films or documentaries.

Setup Your Project

I would recommend creating a new project just for conversions. First up, we will set up the Mastering Settings. Set the timeline resolution to 720x486 NTSC. Are you working in PAL? If so set this to 720x576. All the other steps will be the same, but I will assume you are working with NTSC(North America and Japan) files going forward.

Usually, we work with square pixels with a pixel aspect of 1:1 for HD video and higher. Here we need to change the pixel aspect to conform to the older standard of 0.9:1. Remember when you were young and pushed your face right up against that old Zenith TV set. Perhaps you noticed the RGB rectangles, not squares, that made up the image. The 4:3 standard definition setting accounts for that.

Finally and most importantly, we set the frame rate to 23.976, which is what we want for our output.

At this point simply dragging a clip onto a timeline will result in it being converted to 23.98, but why does it look so steppy and bad? We need to tell Resolve to use high-quality motion estimation. This optical flow setting is the same engine that makes your speed effects look smooth. By setting it on the project page we declare the default method for all time re-mapping to use the highest quality frame interpolation. Including frame rate conversions.

Leveling the Playing Field: 29.97i to 29.97p

Technically 29.97i has a temporal sample rate of 59.94 half resolution(single field) images per second. Before we take 29.97 to 23.98 we need to take the interlaced half-frames and create whole frames. The setting we can engage is the Neural de-interlace. This setting can be found on the Image Scaling page. This will help with aliasing in your final output.

Now that all the project settings have been set, we are ready to create a timeline. A good double-check to make sure everything is behaving as expected is to do a little math.

First we take the frame rate or our source divided by our target frame rate.

29.97 ➗ 23.976 = 1.25

This result is our retime factor

Next we use that factor multiplied by the number of frames in our converted timeline.

1.25 * 1445 = 1818.75

That result will be the original # of frames from the 29.97 source. If everything checks out, then it’s time to render.

Rendering

I prefer to render at source resolution and run any upscaling steps downstream. You can totally skip this step by rendering to HD, 4k, or whatever you need.

I would recommend using Davinci’s Super Scale option if you are uprezing at the same time. This option can be accessed via the Clip Attributes... settings in the contextual menu that pops up when you right-click a source clip.

I hope this helps take your dusty old SD video and prepare it for some new life. This is by no means the “best” conversion out there. Alchemist is still my preferred standards conversion platform. Nvidia has also cooked up some amazing open-source tools for manipulating time via its machine vision toolset. All that said, Resolve does an amazing job, besting the highest quality, very expensive hardware from years past. The best part is it’s free.

Happy Grading,

Colorist PSA - Logarithmic isn't Before

JOHN DARO September 18, 2021

Camera Log is Not Your Before

This is a short service announcement aimed at colorists submitting before and after’s to showcase their work. This post will also be of interest to DPs, Directors, Producers, Facility Managers, or anyone else that is tasked with evaluating or choosing a color professional.

Over the last year, I have received many reels from colorists looking for work. One pet peeve of mine is seeing a clip where there is a “before” that is logarithmic and then “graded” to the display referred “after.”

LogC VS Graded Rec.709

Although this might seem like a drastic transformation, and a great grade, what this tells me instead, is the colorist has a poor understanding of display spaces. It’s pretty much an instant next for me.

Negative VS Camera Raw

Imagine if anyone did the same with film negative. I started with this and I colored it to look like this!

Negative VS Arri Rec.709

Did you, or was that 100+ years of Kodak color science? The log images we start with today are positve but the analogy stills holds. The log file could be considered your digital negative. It’s not ready to be shown until you “print” it. Now there are different ways to take it to display space and all of that is important in the development of the images, but every camera manufacturer has a standard display transform. Think of this as the base from which all other creative assessments should be measured.

LogC VS Standard Arri Rec.709

YouTube is Making Bad Colorists

I often see “tutorials” where an “expert” will start with a log image, then start adding a bunch of nodes, contrast, and curves, and then say ta-dah, here is before and here is after.

The reality is that colorists should be showing, “this is the manufacturer standard math to rec709” and, “here is what I did.” Regardless if that grade is display referred which is another topic, at least it is an apples-to-apples comparison.

Standard Arri Rec.709 VS Graded Rec.709

Your Before and Afters Shouldn’t be That Drastic

There is always a base look for a project. Whether this is the standard camera curves or something custom from a colorist. This should be considered the stock for the show. Colorists are there to help the DP and Director create a particular feeling or response to a set of images. The best grades are the simplest ones. Sure, there are times when shapes are necessary, but your DP knew this on the day, did the cost-benefit analysis, and knew what they could achieve in the DI bay. Most before and afters should show a subtle hand. One that is repeatable so the Director and DP can get consistent and expected results.

When is Log Viewing Appropriate

Filmmakers often want to see their un-LUT’d images. This is always in a technical capacity. It is not uncommon for a DP to want to see a log feed on set to confirm exposure or a director that wants to see their VFX with the LUT off to feel confident that the vfx work is clean and not hiding under the LUT. That said, there is rarely a time when the log image helps you evaluate the beauty or feeling of an image. This should always be done in a display referred space. There was a period, circa 2010 (right around when the first Alexas came out) where a flat washed-out picture was in vogue. I have a hunch this was caused by a creative seeing an un-LUTed monitor and saying “that’s perfect! exactly like that.” Now that’s fine to find inspiration wherever it comes from, but I think we could have avoided that milky period in grading history.

Know your target

Your before and afters should always be in the same target space. The difference should be the work that went into it. You wouldn’t show your PQ HDR “after” in 709 without a transform, so please don’t do it with the log “before.”

Standard Arri rec.709 VS P3 D65 PQ

I imagine your starting point was a well-exposed log picture. You would only need to show it to me if it wasn’t, but that would be doing your DP a disservice and He/She is your number 1,2 and 3 most important priority. In conclusion, a good colorist will never kiss and tell.

Thank You

Thanks for ensuring there is no more harm done to our scene-referred log images. Help me fight the fight to end misrepresented pixels. Thanks for reading and happy grading!

Color Grading & Finishing