Nikon Z8 Video exposure

Horshack said:

PhotonDoe said:

Horshack said:

PhotonDoe said:

Horshack said:

PhotonDoe said:

Horshack said:

PhotonDoe said:

Horshack said:

Here are the input -> output values for Rec.709 and N-Log. You can use these to set the Zebras to the value you want anywhere on the curve. I've included data callouts for 18% grey.

Click to expand...

Hi,

Thank you. How did you do it? By using the nlog specifications document ? What about REC 709, I found nothing in that document.

It doesn't seem linear, so I'm quite curious now why my two tests gave a very different curve shape.

Click to expand...

Yep, I converted the formulas in the N-Log and Rec.709 specifications to Excel formulas then graphed them. Here are the Excel formulas I created if you'd like to graph it yourself:

Rec.709, Input -> Output
=IF(B2<0.018,4.5 * B2, B2^0.45*1.09929682680944 - 0.099)*255

N-Log, Input -> Output
=IF(B2<0.328,650*(B2+0.0075)^(1/3), 150*LN(B2)+619)/1023*255

Put your input values of 0.0 to 1.0 in B2, B3, B4, ...., copying the above formula into C2, C3, C4...

Rec.709 isn't linear, although many mistakenly believe it to be. It's actually logarithmic, just using a different curve than the log formats like N-Log.

Click to expand...

Thanks, a lot of math there! Excellent work!

So middle gray in nlog should be at 93 zebras or 105 for rec 709 (quite far from my 180 conclusion, if one wants to expose skin same as middle gray).

To recap, to see if I understand correctly, and for others that might read:

Luminance from 0 to 1 translates to IREs from 0 to 1023 (or 0% to 100%).

Click to expand...

The x-axis luminance is the scene-referred luminance. The IRE is actually the same as the RGB zebra values on the y-axis, just with a different scale. You can convert the y-axis RGB value to IRE via RGB/256. For example, 93/256 = IRE 36% (which is the N-Log 18% grey IRE) - so N-Log middle grey is 93 on the Zebra, which corresponds to IRE 36% on the waveform.

Horshack said:

This applies regardless of 8-bit or 10-bit or 12-bit (not that I find a good idea to shoot log in 8bit), or is it just for 10 bit ?

Click to expand...

Yes, the bit-depth on the scale (8/10/12 bit) doesn't matter - the values scale accordingly.

Horshack said:

On another note, if Nikon says 32% IRE for Gray 18 in LOG, things I've read say to expose skin at 60% IRE (Rec 709), but considering many say skin should be in middle gray area, it sounds like a huge difference in exposure, even looking at your graph, unless i'm looking at it wrong (skin 60% IRE Rec709 vs 32% IRE nLog)

Click to expand...

Nikon says 35% IRE for N-Log 18% grey (page 12 of Nikon Z7/Z6 II Professional Technical Guide N-Log Recording)

Click to expand...

Yes, thank you, got it after I asked. IRE0-100 = Zebras0-255 on camera screen. I put your formulas in google sheets, replicated your graph, thank you so much easier to understand now.

But why does nLog curve ends at 150? There is definitely space above if I set the zebras to values higher than 150.

Click to expand...

N-Log continues to 255 for input luminances from 1.0 to 8.0. Those represent the 3EV of extra highlight headroom created by N-Log when its "fake" ISO 800 induces exposures that are -3EV "underexposed".

Click to expand...

Thank you, all clear.

I measured on the Calibrite, it seems the light skin swatches in log should be 105/115 on zebras which corresponds to 41 / 44 IRE.

Also tested, setting zebras to 115 and getting them on skin, then importing to DaVinci and CST to Rec709 gives correct exposure, with IRE for skin highlights at 70.

Click to expand...

Great, glad to hear. Like I said earlier, I have a tutorial video planned that I think will bring a lot of people to N-Log and other log formats like S-Log and V-Log. I just have to find the time to work on it

Click to expand...

Andre Yew said:

I do an even simpler version of this (which is what Paul Leeming suggests for his N-log LUTs): I set peaking to 100%, and make sure no highlight detail I want preserved has zebra stripes. I increase exposure up til I just see zebra stripes, and then back down 1 click which is usually 1/3 stop to make them disappear, and that's my ETTR exposure. I've confirmed this by taking it through my Resolve workflow and looking at where the highlights end up on Resolve's scopes, and they're just right below clipping. I have View Assist on in the camera as well.

For your skin reference, you can run tests with bracketed exposures with a chart and a representative person, and then see/measure in your editor which exposure places the skin level you want in your final output space and gamma (say rec.709A for Mac people), and determine from that what your camera metering has to look like on its waveforms to replicate it. It shouldn't take very long to run this test.

In either case, be sure your WB is set to what you want already as that can affect the metering. Usually I WB off the WB panel in my test chart for my tests, but in the field, your WB choices can be pretty different so YMMV.

Click to expand...

PhotonDoe said:

You are absolutely right, simplicity is good, which is one of the reasons I am happy not to need the Ninja V anymore.

What I meant is that I will use zebras to know ar what power to set the light after already exposing for the ambient on the waveform. So measuring zebras on skin tone (Ive found the values in another post) seems the simple way to do this for me

Click to expand...

Horshack said:

The waveform represents IRE, with 0% at the bottom line and 100% at the top line. The three lines in between are 25%, 50%, and 75%.

IRE is a measure of the encoded video signal, not a measure of the light from the scene. Obviously there is a correlation between the two, but that correlation changes depending on how the camera is configured to encode light. With N-Log, the encoding is logarithmic, using the curve described in Nikon's N-Log specification. The N-Log specification says middle gray scene reflectance is encoded as 35% IRE in N-Log. Regardless of the encoding, 0% IRE always refers to clipped blacks and 100% IRE refers to clipped whites.

I'm working on a video tentatively entitled "Log demystified". It will explain how log formats like N-Log work in a level of detail not available anywhere else. My goal is to make log as relatable to video shooters as raw images are to photographers. It's an ambitious project

Here's a quick tip whenever you're at a loss about how to expose a scene in N-Log to the waveform. There are exceptions but this will get you going:

1. While in video mode, switch from N-Log to SDR
2. Set the ISO to 3 stops below your N-Log ISO. For example, for N-Log ISO of 800, set the SDR ISO to 100. The reason for this is explained here.
3. Expose your scene to the SDR waveform however you're accustomed.
4. Switch back to N-Log and restore the ISO back to your N-Log value (the camera will do this automatically if your non N-Log ISO was < 800).
5. Look at how the waveform changed when you switched back to N-Log. This is your future waveform target for that same exposure. In other words, use this technique to train yourself on what the waveforms should look like in N-Log for the exposures you want.

If you prefer to expose in photographic terms (for example, to the exposure meter), follow the same procedure as above but using the stills mode of the camera instead of video mode. Use the "Neutral" picture control when doing so, as that's the closest to Nikon's N-Log intent (for the LUT anyway)This

Click to expand...