GX85 Alexa Conversion!! (and Colour Profile Investigations)

[kye]

Attempt #3

Added in some Luma adjustments.  We're starting to get into pretty tricky territory here, as you'll see from the curves, so this is where I start saying near enough is good enough.  Plus, with a checker we're still firmly in theory-land and who knows what will happen with real footage.

Alexa:

CineD:

This was accomplished with this Hue vs Luma curve:

There are still some challenges though, like the olive green is way too dark, but bringing it up with a Hue curve will bring up the other greens, or if I make the curve specific to the olive patch then is likely to break real footage, especially 8-bit footage.  It's a similar story with the brown patch.

I could go on to soft qualifiers etc etc, but you literally have no idea what colours that aren't on the chart you're severely messing up, and this is where the brain starts having problems thinking about subtle shifts within a 3D space.

Plus, to get this match, I had to apply a luma curve to the CineD footage that matched the levels, which would mean that using this on real footage would darken everything, but that's not sensible in a real-world situation.

Time for real footage.

[projectwoofer]

14 hours ago, Mark Romero 2 said:

Would love to see some of your work using the S1Alex Alexa lut.

Also, are you grading in resolve? If so, what color management are you using?

Yes I started using Resolve Studio about one and a half year ago and still trying to learn it. My latest experiment with both the GX85 and S5 is firstly a tiny bit of halation and glow, then a bunch of S1Alex nodes (I’ve saved the whole thing so it’s just drag and drop) and then a CST node to Cineon so to apply Resolve’s Kodak 2383 film stock emulation. 

[projectwoofer]

9 hours ago, kye said:

I'd imagine they might be quite close.  After all, same profile from same manufacturer, and the GX85 however budget and physically small is still a high-performing camera - to put it in perspective 4K 8-bit 100Mbps is the same as the Sony A7SII.

If you have any tests comparing the cameras it would be interesting to see them.

They are, although I’m not sure what’s the difference between CineD on the GX85 and CineD 2 on the S5. I’ve done quite a bit of comparison shots between the two cameras in order to see if I can match them better using the S1Alex and they where almost indistinguishable from each other with a bit of WB and tint fiddling. I don’t have a color checker though, I’ve done mostly landscape type shots hence my tests are not scientific. Unfortunately I’ve deleted the files from my hard disks when I was done so I don’t have anything to show you but would be happy to try something if you have any ideas. Mind you I’m writing from Germany so it’s not exactly full of colors outside! Could still worth a try though. 

[kye]

and this is why matching only on charts is fraught with peril....

CineD:

Broken image:

I've played with the little tweak that makes the light yellow go light pink and there isn't a combination that smooths it out and doesn't impact the other hues, so that's not possible to make using this method.

Also, that little tweak isn't our only problem....

This is why there are 50,000 ways to do something in Resolve.  Some of them simply don't work!

The path forward from here is to go back to the drawing board and see if there are other ways to approach this that don't break the grade in this way.  Otherwise, we start by applying gentle adjustments and see how far we can get towards the "spirit" of the Alexas image before we start breaking things.

My prediction is that we'll end up with a relatively loose match that will be in the general direction and that'll be the best we can do (or I can do).

[kye]

"Version 3"

Started again, and used a few global tools to get closer, before employing the selective tools.  I also backed away from accuracy to get the "spirit" rather than an absolute match.

I smoothed out the grade, trying to ensure it didn't break the image, like the last one did.

Here's that grade applied to some real footage.

At first I thought that maybe there was some luma vs sat action going on, so I used a node in HSL mode and using the channel mixer I tried mixing in the luma channel to the sat, and vice-versa, but neither generated anything useful.

Then I tried using a YUV node, which made it into the final grade.  YUV is a strange colour space, where the U and V stretch the vectorscope in two directions, and the Alexa is skewed a little towards amplifying one of those directions.  This is a great transformation as it is applied to the whole colour space and therefore does not break the image.

Then I applied another colour warper:

This time paying careful attention to smooth out the transitions.

After the last one breaking the image, I also wondered if I needed to desaturate the transition into pure white, but it turned out that it wasn't necessary.

This is pretty subtle stuff, and I'm not really sure if anyone would bother with it TBH.  I'm happy to share it if anyone is interested, generating a LUT is very easy.

Really, it's a testament to how good the Panasonic colour profiles are to begin with, considering that in comparison to straight "correct" 709 they're both massively far away from that.

[kye]

More playing around....

CineD -> CST 709 to ARRI Alexa / LogC -> ARRI Alexa LUT (worse match)

CineD -> CST 709 to ARRI LogC -> Fujifilm 3513DI LUT (worse match)

CineD -> CST 709 to ARRI LogC -> Kodak 2383 LUT (worse match)

CineD -> CST 709 to ARRI LogC -> Kodak 2393 LUT (worse match)

I also tried a bunch of CST / LUT combinations, but none yielded any promise.

 

[PannySVHS]

We could do a GX85 challenge, including GH2, GH1, GH3, G7 and GH4 aaaand the LX15. The Seven Sisters 8bit Challenge! Let´s do it 🙂