Tales From The Grading Suite – Pains In The LUT

Perfect timing, Dan! haha. That TalkTalk HD color chart was very helpful in showing me how badly hue warping was going on. Looking forward to more “Tales from the Crypt”, I mean, “Grading Suite”.

“Sometimes I feel I need to just stop using LUTs” – my exact feeling everytime I get LUT burned. Trying to learn manual techniques to obtain more filmic results (soft shadoes, soft highlights roll off, subtle saturation on the darker side)

(off topic VFX rant) Maybe it might help if they just did a blank phone with reflections instead of paper taped to it, so you could actually comp back in the reflections.

The noise issue you highlighted is certainly food for thought. Also, the success of certain LUTs has made their look too easy to recognise and therefore somewhat defeating their purpose. Would it not be better to separate the transfer function of LUTs from the colour transform altogether, or is that just not feasible? That aside, I’ve found the trickiest aspect of mimicking a Film Emulation LUT is getting the saturation dependant hue convergence right, while still protecting skintones. It would be great if you had a crack at that.

I’d love to dive deep into this area! I’m not color scientist but I’m very interested in looking into the noise area the most.

I did another job today and tested the footage and found the same thing. Noise in the low midtones that shouldn’t be there.

Would love to speak more about this and learn as much as I can about the options out there!

You know it man! I want to try and keep them light hearted but also cover real world issues and goings on.

LUT burned is such an awesome phrase and I feel that pain oh so well!

Yea I always thought that was the best way to do it. I used to be a Flame op many years ago and loved doing that with tv’s etc..

Looking forward to your findings. That was a LOT of noise for a not so crazy hue adjustment.

I wonder if the differing luma for specific colors attribute to the noise more than the actual hue shift? Similar to if you went a lil too nuts w the hue v luma curve? Just a random thought

The Hue Vs Luma curve is still whack in Resolve 12, so you might be onto something there Jason.

I think a series of Insights about replacing LUTs with PowerGrades would be amazing Dan, especially if you push hard into the advanced stuff, and there’s probably no-one better equipped to do that than you. I’m pretty sure there’s plenty of interest out there for it.

+1 love this topic and agree with the other guys that it would be great to dive into more of this, always reading about how guys just match those LUTs using primary corrections only (curves, lgg, etc) but I find that can be really tricky at times. Thanks Dan!

Offset, contrast and pivot are the tools for filmic tone curves from log I’ve found, not LGG. Of course saturation goes all over the place so you have to deal with that.

I got got by a LUT just today, and all it was was a S-Gammut3.cine to Rec709 LUT from LUTCalc, no contrast. The talent’s shirt went noisy and the edges got blocky too. What a mess.

Nick Shaw of Antler Post is making an app that lets you put a high precision color matrix into a DRX, so I’ve started using a beta of that to use Sony matrices for my color space conversion.

That’s quite an alarming result, especially considering LUTCalc was primarily designed to service Sony footage. Did you send the results to Ben Turley? The high precision color matrix values that Nick’s DRX will use are the same ones used to make the 65x65x65 3D LUT in LUTCalc, so in theory there shouldn’t be a huge difference in results. Anyway, all the more reason to further investigate this noise issue.

I’m going to some serious research into this area. I think Nick from Antler Post is a leader in this area so would love to get him on board to speak more about it!

Hue vs Luma issues can usually be attributed to chroma subsampling and making changes requiring more precision than the color channels can handle.

The noise Dan is seeing is often the result of the low precision of the Cube sizes. A typical cube is 64×64’ish. A saved Power Grade is full 10 bit (floating point). In theory, Resolve Color Management will fix much of this, since it’s uses floating point math to make these transforms.

Also – floating points LUTs will probably solve this, but they seem to be rare in the real world (even though Resolve supports them). In fact, I thought I read that Resolve will save out a Floating Point LUT – but I’ve been unable to replicate that workflow (or find the original source of this information). Maybe worth an email to the Resolve Team?

RE: VFX Rant – I recently graded a bunch of dailies where they used displayed green on the phone rather than taping it on top of the screen. Man – I do not envy the VFX team pulling those keys! In reality, they’ll end up roto’ing out, but yes, adding the reflections back will be much more realistic… if only because they’ve got a real-world reference.

absolutely! Nick is awesome and has created some great “tech” LUTs that have been extremely useful in some of my VFX workflows supervision. +1 on this!

ouch…..I experimented using a similar strategy using a few of those apps that put green & “track points” dots and discovered it the hard way myself. Glad I dont do those too often & when I do I usually can supervise.

Feel free to use as you please :). Those LUTs can get a toasty haha.

I’m making an assumption of the footage that Dan was working with, but would you say this might not be as noisy with a higher quality source? Something RGB444 or RAW ?

A Floating Point LUT, VERY interesting and something I need to look into. I always use 64 x 64 sized cubes, which I assumed were precise enough (assuming no clipping were occurring), but I have no technical info to back that up. Several varieties of “Film Emulation LUTs” were all 64 x 64 when I checked, but still had some of the issues shown.

Seem worth an email to the Resolve team, although I’m not sure who Id reach out to besides Dwaine with this type of question. LUT workflows will still be used by some, even with RCM, so Id assume they would be open to reducing limitations brought on by using typical LUTs clipping at 0 to 1.

Jason – the problem, as its been told to me, is that larger than 64 x 64 LUTs become computationally expensive. IOW, they kill real-time.

But think about it, you’re asking 64 points to recreate (in 10-bit) the manipulation of 1024 data points. It does so by interpolating between those 64 points.

64×64 is the industry standard. You won’t find LUTs larger than that for sale. If you do, they’ll be a rare breed.

LUTs will still be used with RCM – but you’ll generally get more accurate results letting RCM handle the color space transforms and using LUTs that are designed for ‘normalized images’. Of course, the notion of a normalized image is almost as wide open as the notion of ‘camera originals’.

Plug and Play LUTs are BS. But well designed LUTs? They can be a ton of fun for the informed colorist. But an informed colorist is usually someone who could just as easily not use the LUT… and therein lies the rub.

Thanks for the info! It makes a lot of sense when put that way. Ill be experimenting more with RCM for transforms and simply using LUTS for color changes, if at all (depending on what they do).