Fuji X-T3 reborn with HLG

[androidlad]

Since posting this a while ago:

..I've been exploring the power of HLG picture profile introduced two months after X-T3's release. I'm pleasantly surprised by the refreshingly new colorimetry and better (not higher) dynamic range distribution. To me, X-T3 now feels like a completely new camera shooting with HLG. So long F-log!

Here's a better comparison than the one from my previous post:

F-log with Alexa LUT generated by LUTCalc, no grading, WB 5300K, Shift 0:

HLG with Alexa LUT generated by LUTCalc, no grading, WB 5300K, Shift 0:

It's clear that HLG completely removes the magenta hue in F-log and renders shades of blue in a way that's reminiscent of Alexa. Reds are deeper and richer with very little orange shift seen in F-log images. Also, HLG makes full use of 0-1024 10bit code value, while F-log throws away 0-95 which could contribute to banding.

 

Here's a graded frame from a feature film I'm currently shooting, entirely using X-T3 outputting HLG to Ninja V.

Raw HLG frame:

HLG mapped to ARRI Log C Wide Gamut using LUTCalc:

 

Observations for working with HLG on X-T3:

1. Respect the ISO1000 rating and be extra careful, do not over/under-expose more than 0.5 stop, ideally should be spot-on. HLG wasn't intended to be an acquisition format so there's very little room for error.

2. In low key scenes (like the one above), shadow area will have visible grain, with ProRes HQ 10bit 422 recording, the noise pattern is pleasing and monochromatic with a fine 35mm texture, turn in-camera noise reduction off (-4) and "interframe NR" off to preserve the texture, otherwise you get blotchy patterns.

3. An external monitor/recorder with custom 3D LUT feature is a must. X-T3's HLG implementation follows ARIB STD-B67 NHK spec which places middle grey much much lower than BBC spec. You need custom LUTs to help you preview final exposure, otherwise it's hard to see what's going on in the scene and may lead you to overexpose.

4. HLG has a slightly cooler white balance compared to F-log and film sim, so be careful setting the kelvin value. Factor this in when lighting and developing looks for your film. Another reason why a colour accurate external monitor/recorder would be extremely helpful when shooting HLG.

5. Ideally you should shoot HLG with external ProRes HQ 10bit 422 codec due to the aggressive tone curve manipulation in post. Minimum is 400Mbps in-camera HEVC.

6. In post, use a BT.2100 HLG to Alexa 709 3D LUT for grading, HLG uses BT.2020 gamut so you should map it to a BT.709 gamut first, do not grade directly and simply add saturation, the colours will be wrong. Alternatively map it to ARRI Log C gamut and use popular film print LUTs.

[BrunoCH]

Thank you for sharing. Very interesting.

Not really sure that the HLG is better. For me, it depends on the parts of the picture. On the attached screenshot, I prefer the Flog. It's about saturation, colors, contrast. In particular, the dark parts are gray on the log. On the HLG the dark parts are blue.
And the fact that there's very little room for error about expo, makes me look rather for the use of the F-LOG. I will do my own tests to check.

 

[webrunner5]

Looks like it is pulling back contrast to be honest. Sort of the opposite of what I would expect in HDR.

[androidlad]

15 hours ago, BrunoCH said:

Thank you for sharing. Very interesting.

Not really sure that the HLG is better. For me, it depends on the parts of the picture. On the attached screenshot, I prefer the Flog. It's about saturation, colors, contrast. In particular, the dark parts are gray on the log. On the HLG the dark parts are blue.
And the fact that there's very little room for error about expo, makes me look rather for the use of the F-LOG. I will do my own tests to check.

HLG colour is more true to life, and the blue hue in the shadow is correct while F-log is more magenta, not grey.

The scene was shot 2 hours before sunset and there was very little direct sunlight (~5400K) at the time, it was all lit by the mostly clear blue sky (~10000K), which was reflected by the water.

In hindsight, I should have set the white balance a bit higher.

[DanielVranic]

I just have not had the same experience.

I took my rig out to CA for the NAMM show and walked around the beaches in Malibu with it, and every piece of FLog footage transformed to LogC has been gorgous ald lovely sky shots with no evidence of any banding at all, even zoomed in 300% in Davinci.

Shot a bit of stuff HLG and the exposure was at 0.0 the entire time, just walking around LAX and so far, I have not had any success. 

 

[Mako Sports]

Yeah HLG is super OP, love using it.

[thephoenix]

don't see a huge difference compared to flog.

will the audience see it ? i doubt it.

of course you maybe can see it in some projects but is it woth the hassle ?

if it is better, maybe you can use it when you have time and / or controlled environment but for everyday work i think i'll stick to flog that seems to be more flexible

[Grug]

Could you share your HLG to Arri LogC LUT? (or the steps to make it?)

I've been trying to figure out input settings in LUTcalc that would work for that, but I can't find HLG anywhere.

[androidlad]

1 hour ago, Grug said:

Could you share your HLG to Arri LogC LUT? (or the steps to make it?)

I've been trying to figure out input settings in LUTcalc that would work for that, but I can't find HLG anywhere.

[deezid]

2 hours ago, androidlad said:

A cleaner way - if using Resolve - is to use a CST node.

[androidlad]

5 minutes ago, deezid said:

A cleaner way - if using Resolve - is to use a CST node.

CST node is a pure technical transform and does not have any roll off.

[deezid]

6 hours ago, androidlad said:

CST node is a pure technical transform and does not have any roll off.

Luminance mapping helps.
Still cleaner than using a LUT. A 65 LUT shouldn't cause too many issues though to be fair.

[Kubrickian]

Could we see some examples with skin tones? 

[MeanRevert]

Hey guys,

Watching Mitch Lally's video on color grading HLG today makes me want to give it another try.  Thoughts on it?

 

[BrunoCH]

I did my own test HLG vs LOG. There is indeed a small problem with the magenta hue. But for me the problem is also present with the HLG. Download the attached video and play frame by frame and watch carefully the blue bag and sunhood on it .

Exactly the same shooting conditions, same settings, except 2 F-stop difference : f4 Log vs f8 HLG.

I find that the white balance is very difficult to do on the HLG. WB is correct on the gray card but not on the back white wall.

What do you think about that?

https://drive.google.com/file/d/1eCzHpgHrqgmM3ycrpvbauLu-p2vHd3Sd/view?usp=sharing

https://drive.google.com/file/d/17kyJwgaZdMR7q9fk7fotJofSPcepYxJx/view?usp=sharing

[androidlad]

10 hours ago, BrunoCH said:

I did my own test HLG vs LOG. There is indeed a small problem with the magenta hue. But for me the problem is also present with the HLG. Download the attached video and play frame by frame and watch carefully the blue bag and sunhood on it .

Exactly the same shooting conditions, same settings, except 2 F-stop difference : f4 Log vs f8 HLG.

I find that the white balance is very difficult to do on the HLG. WB is correct on the gray card but not on the back white wall.

What do you think about that?

https://drive.google.com/file/d/1eCzHpgHrqgmM3ycrpvbauLu-p2vHd3Sd/view?usp=sharing

https://drive.google.com/file/d/17kyJwgaZdMR7q9fk7fotJofSPcepYxJx/view?usp=sharing

What bitrate did you use? And LUTs? If you used Premiere/Resolve, did you apply the colour shift correction on F-log footage?

If you shot at base ISO 640/1000, the difference should be 2/3 stop, not 2 full stop.

HLG does have a cooler WB across the board.

[BrunoCH]

2 hours ago, androidlad said:

What bitrate did you use? And LUTs? If you used Premiere/Resolve, did you apply the colour shift correction on F-log footage?

If you shot at base ISO 640/1000, the difference should be 2/3 stop, not 2 full stop.

HLG does have a cooler WB across the board.

400Mbps, All I.

I used FCPX with fuji LUTs. WDRlut for Log and Gamut2020to709 for HLG.

Do you want to take a look at the original rushes (copy trimmed)?

https://we.tl/t-wqemKN1N2c

[MeanRevert]

What's the color shift correction? Are you talking about the levels thing? 

[androidlad]

12 minutes ago, BrunoCH said:

400Mbps, All I.

I used FCPX with fuji LUTs. WDRlut for Log and Gamut2020to709 for HLG.

Do you want to take a look at the original rushes (copy trimmed)?

https://we.tl/t-wqemKN1N2c

Thanks for the link.

I noticed that WB was set to auto for both shots, so we don't really know what kelvin and tint the camera decided to use.

Are you sure F-log was f/4 and HLG was f/8? Because when applying F-log/BT.2100 to Alexa LUT, HLG shot still looks very overexposed. It's really important to expose HLG correctly because it maps 18% grey at a low 22IRE.

F-Log to Alexa:

 

BT.2100 to Alexa:

[BrunoCH]

1 hour ago, androidlad said:

I noticed that WB was set to auto for both shots, so we don't really know what kelvin and tint the camera decided to use.

Yes you are right. I don't know exactly the WB. But I use the AWB lock. I did not unlock between the two takes. Normally it is exactly the same balance between the Log and the HLG.

 

1 hour ago, androidlad said:

Are you sure F-log was f/4 and HLG was f/8? Because when applying F-log/BT.2100 to Alexa LUT, HLG shot still looks very overexposed. It's really important to expose HLG correctly because it maps 18% grey at a low 22IRE.

I use a samyang with an aperture ring de-clicked. I'm sure it's around F4 for the log and around F8 for the HLG.
I exposed the log and the HLG with the waveform of my SWIT 55.

 

On the screen of my imac, the HLG seems pretty well exposed (on QT and FCPX with Lut) EDIT: Add iPhone photo on my XT3