Jump to content

GX85 Alexa Conversion!! (and Colour Profile Investigations)


kye
 Share

Recommended Posts

Now I've "made friends" with the GX85, I'm now looking at its colour profiles.  I've installed the CineD / CineV hack from @BTM_Pix so are including these in my testing.

In these tests, I'll be:

  1. Looking at what the colour profiles actually do, with a colour checker and scopes
  2. Comparing these profiles to the colour profiles of the P2K and Alexa
  3. Grading selected GX85 colour profiles to match resemble the P2K and Alexa
  4. Working out what the best combinations might be and just generally working out what setup I will implement

First, what do the colour profiles actually do?

I took the GX85, 12-35mm f2.8 lens (at 35mm) in the 100Mbps 4K mode, and shot my colour checker in direct sunlight.  I then brought those into Resolve, cropped, and blurred a little for cleaner traces in the scopes.  
I actually clipped the white square on the checker a little, oops!, but no test is perfect and neither am I, however I think the results should still be useful.  Here goes.....

Natural - let's use this as the baseline.

For those unfamiliar with these kinds of tests, it is worth noting a few things though:

  • Firstly, this profile is NOT NEUTRAL.
    Anyone who's seen a colour checker in real life can tell you that the colours in the two middle rows (the two below the greyscale) are all about equally saturated - ie, very saturated.  These are the colours that are meant to go in the little reference boxes on the scope.  Notice how the red trace almost reaches the R box?  Notice how the pink and purple boxes look really pastel?
  • Overall the profile is much more saturated in the upper-left (orange) and lower-right (teal) directions.  This is a relatively typical colour palette that looks pleasing.
  • Also notice how the colours are rotated towards those two directions?  it's especially noticeable in the teal direction where the blue is rotated clockwise towards the Cyan reference box, and the cyan is rotated anti-clockwise towards the Blue reference box.

These characteristics give a nice starting point, and are very commonly seen in colour profiles in general, although as the name says, this is a relatively Neutral implementation of that look.

1276595927_ScreenShot2021-11-27at1_28_09pm.thumb.png.5e26de7822d149c007fb782db96d30c1.png

Scenery - it's basically the Natural profile with more saturation, except notice the green is pushed slightly towards blue to give a but more hue separation in the greens.  Oddly enough, that might help give visual interest with scenery shots!

1199946208_ScreenShot2021-11-27at1_28_21pm.thumb.png.255c6230545e4c85da6af8607b8d2477.png

Portrait: Similar to Natural but more saturation (less than scenery) but this one slightly compresses the skin-tone ranges, pulling the yellows towards orange and the reds slightly towards orange too.  This helps to mask blemishes and uneven skin.  Nice. 

100941515_ScreenShot2021-11-27at1_28_37pm.thumb.png.326e35a7019f11bacc00ae891904a016.png

Vivid is..... vivid.  Panasonic really deliver huh!  Like natural but with heaps of saturation.

2028712896_ScreenShot2021-11-27at1_28_48pm.thumb.png.205152410d84f8b63220f4457e0a4653.png

CineD.  The legend.  Lots going on here!  Here's what I see:

  • Slight compression of skin-tones and desaturating right on that hue
  • Higher saturation of all colours except yellow-greens (a hue I know I dislike) and mauve (a hue named after an old lady)
  • Also slightly compressing tones in the teal direction

322857773_ScreenShot2021-11-27at1_28_59pm.thumb.png.1c819c1485bcd4e87cf9f412b0fd4693.png

CineV which is obviously more contrasty with a much lower black level, and consequently is also more saturated than CineD, but is also slightly more "correct" with some hues being closer to their theoretically correct locations, for example, the Red is now almost perfect.

1620239791_ScreenShot2021-11-27at1_29_13pm.thumb.png.c743cb1718ee50f65ed98bed43f4f2d2.png

If you take CineD and pull the shadows down then you get something very similar to CineV, only with more saturation (contrast creates saturation).  If you then drop the saturation a touch to match, you notice that CineD creates more even saturation between the warm tones and cooler tones, whereas CineV has more saturation on the warmer tones and less on the cooler tones.  Note, this isn't a WB difference - this is with the WB set correctly on both profiles.

TLDR;  

Natural profile isn't natural, but is a moderate amount of a relatively common colour profile pushing things towards orange and teal, and compressing skintones for more flattering skin rendering.  Other profiles kind of build on this foundation and kind of do what they imply with their names.  

CineD and CineV are a bit more complex again, with the main difference being the amount of contrast.

Link to comment
Share on other sites

EOSHD Pro Color 5 for Sony cameras EOSHD Z LOG for Nikon CamerasEOSHD C-LOG and Film Profiles for All Canon DSLRs

Now for reference, and thanks to the beauty of standard colour checkers, let's review what the P2K and Alexa look like.

From this rather handy video:

I pulled various scopes.

It's likely to be relatively pointless (and depressing) to compare the GX85 Dynamic Range with these, so I'll skip that and focus on the hues.

P2K Primaries:

image.png

Alexa primaries:

image.png

Things to note here are:

  • Both these cameras have strong looks
  • Orange and Teal directions are much more saturated than the Green / Magenta directions
  • Hues in the Orange and Teal quadrants are rotated closer to the mid-point, with the Teal quadrant more compressed
  • Both the P2K and Alexa pull the Magenta towards Red and leave the Green mostly untouched
  • The P2K is stronger than the Alexa even, with its Magenta being almost on top of the Red, and with lowered Yellow saturation it's really skewing the colours to the Red / Cyan line, rather than the Orange / Teal line

Personally, I think the P2K profile is a little too strong a look for me (you might disagree which is fine) but it's useful to see that the P2K and Alexa are both in the same direction from a "correct" look (with the colours in the reference boxes).

As such, I'll proceed to compare the GX85 with the Alexa colours.

GX85 Natural:

1567458156_ScreenShot2021-11-27at2_51_27pm.thumb.png.ac2acd1d1c9abd2ad0545911540d8200.png

Alexa:

image.png

GX85 CineD

1275403491_ScreenShot2021-11-27at2_52_35pm.thumb.png.b70f4f071261fb95d8671e41b5a448a5.png

We can't compare absolute levels of saturation because I didn't perform the P2K test, which has also been put through YouTube compression, but we can look at the relative hue and sat of the hues.

What I'm seeing:

  • Alexa has similar saturation between Red and Yellow, Natural has Red>Yellow and CineD has Red<Yellow, but these are always adjustable
  • Natural has similar Magenta to Alexa, with CineD giving it more saturation and it being less rotated towards Red
  • Green in CineD seems to match better with Alexa
  • CineD representation of Blue and Cyan is too saturated (comparing to Red as reference) but Natural is also too saturated and with a less accurate match

In terms of which one to choose when grading to match the Alexa, I think you could use either profile.

The challenge is that you're capturing in 8-bit so you want to match in-camera as much as possible.  For this:

  • Both CineD and Natural have the highest shadows of any of the GX85 profiles I tested, so are more likely to contain the whole DR from the camera
  • Levels of saturation are a balancing act, too much sat in camera and you'll clip strong colours, but too little in-camera and you'll be amplifying noise when you boost saturation in post

Now to match the CineD to the Alexa.

Unmodified CineD

191679191_ScreenShot2021-11-27at3_23_16pm.thumb.png.fcd8e8fa5c4e8ffacf52aabe05a6e188.png

Alexa - with colour balance and sat

1131201183_ScreenShot2021-11-27at3_24_20pm.thumb.png.7647480bb10c99a1ead2c04fde36945c.png

GX85 - "Alexa conversion" (ha ha ha)

2024075989_ScreenShot2021-11-27at3_24_40pm.thumb.png.31fc9cf8f060e1accb34669a67dc817b.png

Just for fun I used the new spiderweb tool thingy in Resolve to match them:

1127680326_ScreenShot2021-11-27at3_27_43pm.png.6b02fe1143c102eb63a2ca837887a38e.png

Next is to test it on real footage.

Link to comment
Share on other sites

  • Super Members

If you want to drill down into the different permutations of the elements within the profiles then I'd humbly suggest that you have a look at my profile stepper application which will automate the process for you.

Here are examples of it being used on an FZ2500 (but obviously works on the GX85 and all other Panasonic cameras) to run through the contrast and saturation permutations of the profiles.

CINE LIKE D

 

SCENERY

 

VIVID

You can choose as many or as few parameters to change as you wish and the range of them so in these examples I'm only looking at Contrast from -5 to +2 and Saturation from -5 to +3 but you can go the whole hog if you wish to and do +5/-5 on every parameter if you are looking for a total in-camera look for example. I'd be tempted to include the NR one as it will have an impact on colour.

The interval for the parameter changes in those videos are set to 3 seconds but you can adjust it obviously in the app.

Its easy to generate the subtitle file to embed into the video like I've done here if you wish but it has a display mode whilst running so you can just use the screen of the phone/tablet to show the changes like below.

5ac5fdf8329c7_P1010547-Version2.thumb.jpg.0dd28ad0e1d3e2946bac3c5eb4fc6614.jpg.f1d3efbb6abe49437331ae44b97c3f1c.jpg

Unfortunately, the shadows and highlight controls are not accessible externally so can't be included in the automation so especially if you are looking for a total in-camera look (or just for completeness) then once you are in the ball park of the profile that you want then I would be inclined to re-run the stepper with smaller permutation changes of contrast and colour only with manually set shadows and highlights permutations. 

Link to comment
Share on other sites

I have the S1Alex Alexa conversion which I’m using with the S5 with great success. It’s for V-Log footage but includes conversions for HLG and CineD which I’m also using. So as much as I understand that it’s not the same camera, I’m currently experimenting with using the conversion on CineD footage from the GX85. I have not made many tests yet (let alone scientific ones!) but to my eyes the GX85 footage behaves in a similar way so the end result is pretty close. 

Link to comment
Share on other sites

One tasty read! Thank you very much. Quality and fun stuff. Very grateful, puts a warm feeling in my heart n mind!

I think Lumix cams have a great colour palette as a start for grading, that´s G6, G7, GX85 and GH5 from my experience. S1 is a bit trickier to grade, with superiour possiblities and wiggle room but also more chance of hitting the wrong note so to say. GX85 is a beauty. Too bad is starts producing mudd under lower light conditions. But lighting for 800iso and F2.8 still gives perfect results, 1600 is limit under sufficient light.

Link to comment
Share on other sites

Thanks for doing all the work, @kye

Probably the thing that is most surprising to me was looking at the scopes between the P2K and the Alexa. 

When I look at the skin tones, the P2K looks a lot more magenta and the Alexa a lot more green. So I thought the SATURATION levels would be much different than they are on the scopes (i.e., P2K have more saturated magenta, Alexa having more saturated greens).

But in the scopes, the main difference seems to be the P2K has magenta rotate more CCW and greens rotated more toward Cyan, while Alexa greens and magentas are more "true" and the yellow is really heavily saturated.

Unless i am missing something (I often do)...

Oh, and there is a luminance difference between the two captures, which leads into the other part of the equation, which is how bright or dark a profile captures a particular hue. If I understand correctly how the scopes work (and I probably don't), and how the human mind works (I definitely don't understand that) a color checker would show the same saturation levels even if the luminance levels varied somewhat, but the human eye might PERCEIVE more saturation in the darker level. 

I mention that because it is an old lightroom / photoshop trick that if you wanted to make a bright sky look more saturated, instead of going overboard saturating the blue channel, you would saturate some and then darken the blues to make it look like a deeper blue sky.

Link to comment
Share on other sites

3 hours ago, projectwoofer said:

I have the S1Alex Alexa conversion which I’m using with the S5 with great success. It’s for V-Log footage but includes conversions for HLG and CineD which I’m also using. So as much as I understand that it’s not the same camera, I’m currently experimenting with using the conversion on CineD footage from the GX85. I have not made many tests yet (let alone scientific ones!) but to my eyes the GX85 footage behaves in a similar way so the end result is pretty close. 

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?

Link to comment
Share on other sites

12 hours ago, BTM_Pix said:

If you want to drill down into the different permutations of the elements within the profiles then I'd humbly suggest that you have a look at my profile stepper application which will automate the process for you.

Here are examples of it being used on an FZ2500 (but obviously works on the GX85 and all other Panasonic cameras) to run through the contrast and saturation permutations of the profiles.

CINE LIKE D

 

SCENERY

 

VIVID

You can choose as many or as few parameters to change as you wish and the range of them so in these examples I'm only looking at Contrast from -5 to +2 and Saturation from -5 to +3 but you can go the whole hog if you wish to and do +5/-5 on every parameter if you are looking for a total in-camera look for example. I'd be tempted to include the NR one as it will have an impact on colour.

The interval for the parameter changes in those videos are set to 3 seconds but you can adjust it obviously in the app.

Its easy to generate the subtitle file to embed into the video like I've done here if you wish but it has a display mode whilst running so you can just use the screen of the phone/tablet to show the changes like below.

5ac5fdf8329c7_P1010547-Version2.thumb.jpg.0dd28ad0e1d3e2946bac3c5eb4fc6614.jpg.f1d3efbb6abe49437331ae44b97c3f1c.jpg

Unfortunately, the shadows and highlight controls are not accessible externally so can't be included in the automation so especially if you are looking for a total in-camera look (or just for completeness) then once you are in the ball park of the profile that you want then I would be inclined to re-run the stepper with smaller permutation changes of contrast and colour only with manually set shadows and highlights permutations. 

It's early days for this work for me, but I might want to dial something in further in future, in which case that setup looks spectacular!

TBH I've *almost* reached my limits (skill/patience/motivation..) with what I've done above, so I probably wouldn't go that much further in terms of matching them.  My goal, after all, is "good colour", not to mix the footage between cameras or anything.

One thing that I thought of afterwards was that when I said I applied a WB to the Alexa footage, I actually adjusted WB on the Gain and Lift controls, because what was neutral for the highlight square wasn't neutral for the shadow square.  I realised afterwards that step 1 for colour matching is to match the greyscales, which I forgot and skipped in the above.

My overall impression was that:

  • I was surprised how similar the Alexa, Natural, and CineD profiles are.  On the scopes at least - I haven't tried applying it to real-world footage yet.  
  • I would have thought that the Neutral would have been more "correct"
  • I would have thought that CineD would have been stronger of a look (which was my impression from looking at footage from it)
  • I was also surprised that the P2K look was so different to the Alexa, and also that it was stronger, but I'd already digested that surprise in the BMMCC thread, so wasn't new 🙂 
Link to comment
Share on other sites

9 hours ago, projectwoofer said:

I have the S1Alex Alexa conversion which I’m using with the S5 with great success. It’s for V-Log footage but includes conversions for HLG and CineD which I’m also using. So as much as I understand that it’s not the same camera, I’m currently experimenting with using the conversion on CineD footage from the GX85. I have not made many tests yet (let alone scientific ones!) but to my eyes the GX85 footage behaves in a similar way so the end result is pretty close. 

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.

Link to comment
Share on other sites

7 hours ago, PannySVHS said:

One tasty read! Thank you very much. Quality and fun stuff. Very grateful, puts a warm feeling in my heart n mind!

I think Lumix cams have a great colour palette as a start for grading, that´s G6, G7, GX85 and GH5 from my experience. S1 is a bit trickier to grade, with superiour possiblities and wiggle room but also more chance of hitting the wrong note so to say. GX85 is a beauty. Too bad is starts producing mudd under lower light conditions. But lighting for 800iso and F2.8 still gives perfect results, 1600 is limit under sufficient light.

Yes, I was surprised how similar the GX85 profiles were to the Alexa in terms of overall look.  Obviously the Alexa has hugely more dynamic range and codec quality, but colour profiles are something that a manufacturer has to implement into their cameras (unless they're RAW only) and because they're basically just a LUT you can put whatever sophistication into them that you like, and Panasonic seems to have done a great job in this aspect.

Link to comment
Share on other sites

5 hours ago, Mark Romero 2 said:

Thanks for doing all the work, @kye

Probably the thing that is most surprising to me was looking at the scopes between the P2K and the Alexa. 

When I look at the skin tones, the P2K looks a lot more magenta and the Alexa a lot more green. So I thought the SATURATION levels would be much different than they are on the scopes (i.e., P2K have more saturated magenta, Alexa having more saturated greens).

But in the scopes, the main difference seems to be the P2K has magenta rotate more CCW and greens rotated more toward Cyan, while Alexa greens and magentas are more "true" and the yellow is really heavily saturated.

Unless i am missing something (I often do)...

Oh, and there is a luminance difference between the two captures, which leads into the other part of the equation, which is how bright or dark a profile captures a particular hue. If I understand correctly how the scopes work (and I probably don't), and how the human mind works (I definitely don't understand that) a color checker would show the same saturation levels even if the luminance levels varied somewhat, but the human eye might PERCEIVE more saturation in the darker level. 

I mention that because it is an old lightroom / photoshop trick that if you wanted to make a bright sky look more saturated, instead of going overboard saturating the blue channel, you would saturate some and then darken the blues to make it look like a deeper blue sky.

You're not missing the point at all - the "conversion" I've done above is scratching the surface so little that my post is practically a stand-up comedy routine.

The vector scope does ignore luminance, so you're right there, but it's more complicated than that.  Here's a few thoughts:

  1. Firstly, it's not possible to do a "real" match between two cameras unless you know where middle grey is, as that's the luminance level that you will be using these cameras at for exposing skin-tones and so camera profiles typically leave this luma at the set WB but are free to push the balance of everything brighter and darker than this point (e.g., warmer highlights and cooler shadows are very common) and as you're meant to use a colour checker by having the brightest square just under clipping and the cameras have different dynamic ranges the middle grey will be different and might not even be on that scale
  2. LUTs are most often 3D LUTs because the transformation is in 3D, meaning across hue / saturation / luminance (HSL).  This means that while I've matched the hues at those levels of S and L, the same H could be different at literally every other combination of S and L.  Colour profiles and colour science are hugely subtle things, relying on very small tweaks all across the HSL 3D space.  To match cameras truly you need to point the cameras at all the combinations of H S and L and then make a transformation that pushes one 3D lattice to the values of the other one.  This is what @Sage did in his GHAlex work.  I have a theory that even this isn't enough because it sits on top of the cameras WB, but that's a different story.
  3. The practical concern of the above is that I have matched a few strongly saturated hues but haven't matched the less saturated ones, which is where the skintones sit.  The charts do have some of these, so there's another level I can go to with what I have.  As I mentioned above in my reply to @BTM_Pix I also forgot to match the greyscales, but that's subject to middle-grey considerations, so not sure how well I could do that.

Having said all that though, I've done simple hue matching (Hue vs Hue, Hue vs Sat, Hue vs Lum) to "match" radically lesser cameras to greater ones before (GF3 to P2K) and as crude as these adjustments are (and they are so tremendously crude as to basically be a joke) the resulting images certainly benefit considerably from that treatment.

The spirit of this is really that doing a "conversion" is impossible, my approach is laughably inaccurate, but there are still benefits to be had.  Both from learning along the way as well as getting something useful to use afterwards.

Link to comment
Share on other sites

4 minutes ago, Emanuel said:

Great thread as usual as your standard has already become a common place to all of us beyond any differences or distinct angles for the subject matter : )

Pleasant read on such marvelous GX85 piece, a dream made camera :- )

My GX85 cost me about 30% of what my GH5 cost, but the more I use it the more impressed I am by it.  If I didn't have a GH5 for some reason, I could get by with just this camera, and would enjoy the process too.

Link to comment
Share on other sites

15 minutes ago, kye said:

The spirit of this is really that doing a "conversion" is impossible, my approach is laughably inaccurate, but there are still benefits to be had.  Both from learning along the way as well as getting something useful to use afterwards.

Well... hardly laughable in my humble opinion. Certainly a noble effort. I think it does move general understanding about this forward, and the fact that it is pretty much demonstrable is great, too. I remember seeing the Juan Melara tutorial "Re-creating the Linny LUT" and I was like, "I never want to shoot video ever again." I was basically lost within the first 10 seconds of that video (since removed from youtube).

Link to comment
Share on other sites

1 minute ago, Mark Romero 2 said:

Well... hardly laughable in my humble opinion. Certainly a noble effort. I think it does move general understanding about this forward. I remember seeing the Juan Melara tutorial "Re-creating the linny Lut" and I was like, "I never want to shoot video ever again." I was basically lost within the first 10 seconds of that video (since removed from youtube).

It's laughable in terms of how crude it is when you look at colour science and colouring as a whole.

I have re-watched every Juan Melara video a dozen times or more and I've gotten to the point of understanding (almost?) everything that's going on in those videos.  I had to watch, rewatch, follow along and recreate them, and even research techniques and terminology to get to this point, so it was literally studying.

I say that not to brag, but as context for the following:

  1. I regularly see content on the colourist forums that is significantly more complex than anything Juan has included in his videos
  2. I see colour grading issues with a lot of online content, and am starting to see it in lower-budget TV shows now too
  3. I regularly see colourists talking about differences in looks that are so subtle I can't even tell the difference, and yet they are having a nuanced and productive conversation about it, so not only can they see it, but they're able to break it down and discuss it

However, here's the rub - getting great colour is 95% the basics and 5% the complicated stuff.  You can literally get 80% of the results of a high-end show by:

  • Use any RAW-shooting camera
  • Light well and within the limits of your camera
  • Expose it properly, and shoot a grey-card
  • Applying the manufacturers 709 conversion
  • Grade using only the RAW controls, the Contrast/Pivot/Offset and the LGG controls
  • That's it

And if you are willing to compromise even further, to get maybe 50% of the results of a high-end show, all you have to do is:

  • pick a camera released in the last decade
  • point it at something interesting
Link to comment
Share on other sites

Attempt #2.

This time I used a more sophisticated approach, and matched on all the patches in the colour checker.

Alexa:

2136560777_ScreenShot2021-11-28at11_33_56am.thumb.png.480861fc24add0aaa34cdda0eb27ba1a.png

CineD:

1686589758_ScreenShot2021-11-28at11_35_22am.thumb.png.976f0d4a57ef79c19dac2d4f6d85c658.png

This is almost a perfect match.......if all you ever film is colour charts!

To get this I started again, and used the Colour Warper but with the highest number of points it allows, and ended up with this:

1553767280_ScreenShot2021-11-28at11_39_59am.png.3ad111109ca0449675fc5c793fb6f472.png

Things to note:

  • The points with black outlines are points I've adjusted
  • The white patches below the spiderweb is the image data
  • There are entire radial arms with no image data below it to match - this is the challenge of having a colour checker with limited numbers of points...  if this were a real matching then I would need thousands of points to compare.  For the sake of this process, I simply took the areas that had no data and spaced them out relatively smoothly

This spider curve only adjusts hue and saturation, but ignores luminance, and was mis-matching the skin tone next to the brown patch.  For this, I did a qualifier and adjusted hue and saturation to match, making the qualifier as soft as possible while not messing up the other points.

This was the qualifier:

2089898486_ScreenShot2021-11-28at11_46_45am.png.a24944b88e449837e28485033a0b0993.png

In plain English, this takes lighter skin tone hues on the brown side and shifts them towards pink.

Without this adjustment, instead of getting this:

1834668376_ScreenShot2021-11-28at11_48_34am.png.cd738f1edb7b068de620df6f035f5324.png

You get this:

87749132_ScreenShot2021-11-28at11_48_55am.png.c4712d4a80d6521da57f117cb2d8280c.png

I still haven't matched the WB or Luma...

Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Restore formatting

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

 Share

  • EOSHD Pro Color 5 for All Sony cameras
    EOSHD C-LOG and Film Profiles for All Canon DSLRs
    EOSHD Dynamic Range Enhancer for H.264/H.265
×
×
  • Create New...