ETTR: The Ultimate Exposure Technique?

[Guest Ebrahim Saadawi]

​Yes, it's the same skin color at first glance. Yet it looks as if the skin was softened in the ETTR image. You really can't judge that if you take just one shot. Shoot an entire sequence with two or more people hit by the light sources in different angles. Then try to match those in post. One looks perfect, the other 'posterized' (old Photoshop effect), and you can't do anything about it. 

Carpets and vases have no 'memory colors'. They can be reduced to, say, five nuances, and nobody will notice. A face will need - I don't know - fifteen? to look alive.

​ The skin isn't softer at all, not difference in detail at all even at 400% viewing. Perhaps you mean the back of his neck where I mistakengly clipped that part in the ETTR sample (so It was my fault, did'ny do correct ettr) more accurate tests coming on the skin issue. If ETTR doesn't have negative effect on skin then it will be the way to expose. We'll se. 

[Bioskop.Inc]

 

​Of course ETTR is right for raw, nobody disagreed. This is about whether it's also right for compressed codecs, which contradicts common beliefs.

​Must have missed the bit about ETTR working with compressed codecs - makes sense though. Will give it a whirl with my 60D just for fun.

The only question i have regarding this is technique is - does it increase the detail of the image in the same way that ETTR does with Raw?

[andrgl]

The only question i have regarding this is technique is - does it increase the detail of the image in the same way that ETTR does with Raw?

​Doesn't it? My eye can resolve greater detail in the ETTR stills.

[Guest Ebrahim Saadawi]

no you raw test was great, really helpful. thanks for sharing. 

 

now if you still have your 60d, do the same test in h.264, hopefullg in Cinestyle or Nuetralwith minimum contrast if you don't have cinestyle. (for some reason my tests showed ettr works better with log-like profiles vs normal contrasty ones (cinesty

e vs standard for example, ettr doesn't work att with these contrasty profiles, only with raw and with compressed LOG (Slog, CineD,Clog,Cinestyld etc)

 

so, choose Cinestyle, or neutral with contrast off, turn Highlight Tone Priority On (helps with ETTR as it reduces highlight clipping), shoot the same scene with a dark-ish normal exposure and one with a high exposure just before clipping (using your eyes and the histogram, and you can also hit the info button in the video preview and you'll get zebras flashing on the clipped parts if there's any, just to make sure), and extract TIFF. files and post both here for us to do the rest of the test. 

[Guest Ebrahim Saadawi]

The effect I am gaining is exactly this,

 

-Lower noise in the shadows,

-Higher DR, and

-Thicker colour if I may say, the colours look rich when you bring them down in post in ETTR while shooting the same scene in normal expoaure the colours are weaker desaturated and fall apart easier. 

-But no advantage in resolution/detail, perhaps in the extreme shadows due to lack of noise but in the nornal skin/objects detail us the same. Even in raw I don't see ETTR increasing detail at all, where did you guya get that?

[Bioskop.Inc]

These are some useful/interesting articles explaining ETTR - they are talking about RAW pictures, but the science seems to be the same:

http://schewephoto.com/ETTR/

http://www.luminous-landscape.com/tutorials/expose-right.shtml

http://www.luminous-landscape.com/tutorials/optimizing_exposure.shtml

 

[Deadcode]

I also tried to shoot ETTR with my LX100. I also had lower noise, and good colors, but the grading process is far longer, because i got different exposure on almost every scene. I dont think there is enough benefit to shoot ETTR with compressed codec, it's easy to clip highlights, and much more work for just a littlebit better results.

[Guest Ebrahim Saadawi]

Does the LX100 have a log-style mode? I as because as Samuel H pointed out to me on dvxuser, ETTR is far less useful with cameras that shoot a contrasty image, So I tested and found it true. For example it works with Canon Cinestyle but not with neutral, I found this particularly strange (didn't know cinestyle actually changed the image that much), it also works with SLOG2 on the a7s but not with normal picture profiles When using a rec.709 contrasty image there actually isn't that much room to increase exposure without clipping, so the noise benefit is so minuscule it's almost not worth bothering, but with flat Log images, it's complete magic (Slog,Clog,CineD,Cinestyle, which all are different but essentially a log curve, ETTR is tested and confirmed to make enormous difference with all of them).

 

[Guest Ebrahim Saadawi]

 but the grading process is far longer, because i got different exposure on almost every scene.

​You can get the benefits of ETTR by overexposing all shots a constant amount, say around a stop, and create a simple LUT to bring all the footage to normal range at once with one click. But yes this way you don't get the full 100% benefit of ETTR as some scenes will ETTR at +2 stops, and +0.5 stop, etc, so to get the lowest noise possible without clipping highlight, expose each scene to ETTR and match all shots in post using a Luma curve + waveform monitor. It's not that much of a nuisance for me, but I do understand it would be a problem for those who want fast direct deliverables, and I recommend against ETTR (or LOG) in these cases, shoot with a normal contrast and correct exposure -> deliver quick.

As I see it ​ETTR, like Log gammas, is for those who have time/experience to correct/grade the images, and seek the highest quality possible (film, music,)  over convenience (doc, news, family, etc,). So don't automatically jump on ETTR & S-LOG just because we say it's better without understanding the downsides. Rec.709 + Correct exposure still have a big place & more suitable than ETTR + LOG on most normal productions. 

[Deadcode]

You are right, LX100 dont have LOG profile, i tried with "standard" profile all dialed down + curves.

Something like this (https://vimeo.com/109709153) is the result, it's not log, but fairly "FLAT"

The codec on LX can usually handle +1 stop without clipping but not more. i will do some tests with fixed +1EV as you mentioned.

I think CineLike D on GH4 is still Rec709, with lower contrast and with a littlebit different color profile. 

SLOG2 on the A7s would be the only compressed codec which can have real benefit from ETTR, but the noise performance on that camera is not an issue

But A7 II users will love it

[captou]

Hi, 

I did some preliminary testing on this on my Canon 600D and it does seem to make a difference. I used the ML RGB histogram, Cinestyle picture style, Highlight Tone Priority On, Zebras (Luma) at 100%. The image is very flat but does get muddy a bit with higher ISO values (>3200) but 1600 seems ok. 

Could you just explain your steps a bit more in camera and in post? Why do recommend zebras at 100% and how about when you film a person, what zebra setting do you use for this? I read some keep it around 70-80 for Caucasian faces to expose correctly. But then I guess the whole point is to not expose correctly but overdue it a bit and then recover in post and that's only possible if you don't clip highlights in the face - so what would you recommend in an interview setup? 

Cheers!

[Axel]

Why do recommend zebras at 100% and how about when you film a person, what zebra setting do you use for this? I read some keep it around 70-80 for Caucasian faces to expose correctly. But then I guess the whole point is to not expose correctly but overdue it a bit and then recover in post and that's only possible if you don't clip highlights in the face - so what would you recommend in an interview setup? 

​For lack of a histogram on my BMPCC (it came with a FW update only a week later), I exposed a whole wedding (almost four hours of footage) to the right, with ultraflat ProRes 10-bit LOG, using only 95% zebra. Most shots were fine, as expected (did a lot of tests before), but some shots had weird plasticlike skintones that I couldn't fully rescue in post. I have been shooting weddings with various cameras before, of course I always EdTTC or used auto exposure (which does the same). That said, I do very often like skin colors of DSLR video, but not so much the skin's gradation. I propose you really perform many tests with ETTRd skin and match them before you announce the method to be superior.

[Guest Ebrahim Saadawi]

 I propose you really perform many tests with ETTRd skin and match them before you announce the method to be superior.

+1

​I also advice making your own tests before shooting on an actual production using my ettr findings. 
 

Hi, 

I did some preliminary testing on this on my Canon 600D and it does seem to make a difference. I used the ML RGB histogram, Cinestyle picture style, Highlight Tone Priority On, Zebras (Luma) at 100%. The image is very flat but does get muddy a bit with higher ISO values (>3200) but 1600 seems ok. 

Could you just explain your steps a bit more in camera and in post? Why do recommend zebras at 100% and how about when you film a person, what zebra setting do you use for this? I read some keep it around 70-80 for Caucasian faces to expose correctly. But then I guess the whole point is to not expose correctly but overdue it a bit and then recover in post and that's only possible if you don't clip highlights in the face - so what would you recommend in an interview setup? 

Cheers!

​Why I recommend zebras at 100% is that when it's set to that number, zebras will only start showing on parts that are clipped. For example, if you set it to 80%, zaebras will start showing on parts that are still not clipped, you can still go higher. So setting zebras to 100% and exposing exactly to the point before it stars showing, means your achieving perfect ETTR and exposing as high as possible without clipping. 

This is for all situations, either people or any other. The point of ettr is exposing higher than 70 IRE for skin and then going back to it in post. So for interviews or anything, expose as high as possible without clipping. Skin tests coming up but again do your own tests to verify before shooting something important.

About correcting ETTR in post, simply use a Luma curve. Here's a longer answer I sent to someone asking the same question about ETTR post correction: 

"are many ways to bring the exposure down in your NLE. You can bring down the entire waveform with a "brightness" plugin of some sort, of bring down the highs/mids/lows using levels, or a Luma curve.

My preferred method for correcting exposure in the Luma curve, I don't believe in LUTS and so forth, so I correct ad match each shot using curves, they give you precise control of which parts of the image you want to affect. 

Here's an example of how I correct an ETTR with a Luma Curve  (sorry for the random weird dental model subject!)

Here's before applying any correction, over exposed naturally



Here's after applying that luma curve shown on the left (it's showing in the first photo as well but disabled)



Here's the actual ETTR vs normal exposure for reference: 

ETTR

Normal

ETTR corrected with curve


So my advice, use Luma curve, it will be built in in any NLE I am sure. Simple and effective."

[Vesku]

I thought that when using 8bit codec and ETTR we must use darker profiles, not flat. There is a supertone method for GH4 which uses contrast +5 and highlight -5 and ETTR:

http://www.supertone.dk/#!GH4-Optimal-Film-Setting-works-in-stills-too/c24o4/8E18836A-F271-4A14-AF0D-C575B9D5F4B6

So when exposing to max and using darkest settings we get very clean image but less dynamic range. When using ETTR with flat 8bit profile we must darken it so much in post that it may introduce banding and other issues. We also darken floor noise when we darken image in post. In my tests with GH4 I prefer low contrast look with usually 1/3 underexposure. The noise is acceptable. I must also test more. The problem is that when I set contrast to normal or even + I loose dyn range.

[TSV]

Thanks guys, I learned something more thanks to you!  https://vimeo.com/74387127

[Guest Ebrahim Saadawi]

It's just a lot better to have a bright image to darken in post, with the most light information (brightest possible with no clipping, no luma or chroma clipping), than to have a dark image to manipulate. All the cameras I've ever tried behave like this. They're better to push down than up. Much better. With the former noise and codec breaking is immensely reduced in the shadows.

And I've never seen any codec break while pushing down even Canon's 8bit H.264 as theory suggests, huge pushing down, it just never happens, try it (unless you push down a clipped image which in this case ETTR is not achieved). But pushing up even a slightly miniscule amount, codec breaks into pieces, literally, with all non raw codecs that is. 

If need to directly deliver viewable footage, of course not this is not for you, expose for the desired look and desired contrast and colour in-camera, but you will be getting more noise and less DR and overally lower image quality than grading it finely in your NLE.

[Vesku]

It's just a lot better to have a bright image to darken in post, with the most light information (brightest possible with no clipping, no luma or chroma clipping), than to have a dark image to manipulate. All the cameras I've ever tried behave like this. They're better to push down than up. Much better. With the former noise and codec breaking is immensely reduced in the shadows.

And I've never seen any codec break while pushing down even Canon's 8bit H.264 as theory suggests, huge pushing down, it just never happens...

​If we shoot very bright images with 8 bit codec and adjust colors back to normal in post we may get banding even if we have perfect 8 bit source. Every light shade darkens to double so we get half the shades or steps in mid tones. Would it be better to have that correction in-camera (profile) when sensor makes the 8 bit image from 12 or 14 bit sensor. The problem is that when we select a darker profile (contrast +) the highlights clips more easily.

This thing is after all a compromise of better dynamic range/noisier (low iso underexposure) or less dyn range/cleaner (high iso ETTR). We must think what is the main subject or goal in image. 

One thing is that new televisions has very good NR and motion enhancements. Quite noisy videos looks very good with them and miles better than with normal computer monitor. So if I have a very good display  I can tolerate more noise.

[JuiH]

hi, i am fairly new to colour grading and to filmmaking. i used to use an old Panasonic MX and a Canon XL2 and did not do any adjustments to the video apart from editing, in-camera white balance and audio levels. i have more recently been learning colour grading with ColorFinale and Cinema Grade in Final Cut Pro X.

in terms of ETTR, would this be more suitable for normal picture profiles / picture styles than for Log profiles? i understand that Log profiles stretch out the data for highlights in order to increase the dynamic range that can be captured and in this case wonder whether there would be less data for ETTR in the highlight areas.

i currently use a Sony FS-5 which i usually shoot in Cinegamma (PP4) in 10-bit 4:2:2 Full HD, however i also use Canon DSLR's (5D MArk iv) and Mirrorless (EOS R) which i have been recently combining with a BlackMagic Video Assist 5" in order to obtain outboard 8-bit 4:2:2 Full HD. i recently upgraded my 5D MArk iv with C.Log but have yet to try C.Log with either camera.

[tupp]

ETTR is suitable for raw and uncompressed formats.

 

Just be careful going by a histogram alone.  If you have zebras, set them in the range of 95%-100%, and use them to determine your upper limit and to choose which parts of the image may or may not blow-out.  Waveform is also good for finding the upper limit.

[kye]

Also be sure to do tests to verify that you’re getting the kind of results that you want.   ETTR is more work to shoot with and process in post, so if you’re not getting the level of improvement to justify the extra effort then it’s good to know that early instead of doing it for months or years and realising later