For a while, 'veI used Neutral picture setting with contrast set to -4, saturation -2, but I'm
wondering how much flattening is too much, given the 8 bit codec. At what point does flattening
throw out too much picture information, that same way punching up contrast does?

Do I really want to flatten contrast because it gives me more control in post?
I'd rather adjust the contrast to "saturate the full range of grey levels". Bake it in.
Which in practice probably means not pushing contrast too far in camera.

Since we only get 0-255 from white to black, if you aggressively flatten the contrast to open up
both highlights and shadows, you have to be careful you're not baking in a range more like
35-220. Its a compressed the dynamic range, even if later in post it looks like I'm visually
dialling it back in.

Logically it doesn't seem like there should be any benefit from using "raw" shooting practices
on 8 bit video, and it best to stay at whatever gets you closest to the full 0-255 range you
can get in camera. Does anyone still think flattening is good even knowing mathematically
you're throwing out information? Whatever looks good?

It depends...

When the sun is out and the scene has high contrast, dial the contrast down so you can keep from crushing blacks and blowing out whites.

When the clouds roll in and the scene has low contrast, dial the contrast in the camera up so you can use the full 8-bit range.

One thing though - regardless of everything else, try to keep skin tones in the sweet spot. Even if the face is the brightest object in the scene, you don't want to push the face to the right of the histogram. Always try to keep faces in "zone 5".

Er, no.

Keep your contrast dialed down. Always.

You are not throwing out information. You are simply keeping your options open for later in the process.

Look at your histogram...

But if you shoot a scene where all the content is between levels 120 and 140, you only have 20 unique values. If you can increase the contrast and get the same image to span 100 to 160, you will have three times as much subtlety and you won't lose any info.

I agree that it's best to keep contrast low in general, but when the scene has very little contrast, a high contrast setting in the camera is an option.

One thing though - regardless of everything else, try to keep skin tones in the sweet spot. Even if the face is the brightest object in the scene, you don't want to push the face to the right of the histogram. Always try to keep faces in "zone 5".

I'm actually not sure about this... digital sensors aren't film, the lower the zone you get the fewer bits you get to fit things into. Max a face's exposure at zone 5 and you've essentially discarded 128 possible tonal variations, and pushed the skintones into the noisier half of the sensor's sensitivity range. You certainly don't want to clip a face, but as long as you don't you're far better off bringing down the brightness in post than bringing it up (and the noise along with it).

I was unclear. I wouldn't set the highlights of the face no higher than Zone 5. I would put the meat of the face in zone 5. Shadows will be darker; highlights lighter. A bad bounce off my thinning scalp might even clip.

The problem with ETTR on DvSLR video is that the S-curve kicks in as we get into the highlights. Exposing too high or two low will take the face out of the sweet center.

I'm a real fan of consistent skin tone exposure. The shadows and highlights might not be perfect, but if the skin tone is consistent from cut to cut, color correction can be a lot of fun. On the other hand, it can be a real slog if all the effort is on matching inconsistent clips.

Logically it doesn't seem like there should be any benefit from using "raw" shooting practices
on 8 bit video

No it doesn't. Some folks think, because their histogram shows a "flatter" image, that they are somehow magically getting "more" info to work with later. If you like grading from a flatter image, then flattening the image in post does not result in any more info lost that flattening in-camera, and in post you can monitor and measure more accurately, and can take the time to work with each scene individually.



Does anyone still think flattening is good even knowing mathematically
you're throwing out information?

No. There's a difference between dialing the contrast down, and then loading one of those profiles with a flattened curve, some of those super-duper flat curves can really throw a lot of info out. Each camera has a native response, and increasing or decreasing above or beyond that point is destructive. I'm not sure what that response is on the 7D, I'd guess that it is about -2 for "Neutral"

Optical methods for contrast control give one the best image, control the lighting, expose properly, filtration...

Jon, Liam fantastic comments.

The two bits of practical advice that I will use going forward:

-Be careful with flattening already low contrast scenes. It might
be wise to have different profiles to handle this.

-When in doubt, check the histogram.

I think the third bit of practical advice is to

- concentrate on consistent exposure of faces. The "meat" of the face should be around the 50% to 65%, depending on skin tones and your personal preference.

If you nail the skin tones, you minimize time spent in post matching cuts. Also, by keeping the skin tones in the linear part of the curve where you record the most detail. This avoids pushing faces into the s-curves where bits get scarce and the face turns plastic (too bright) or noisy (too dark).

The audience will forgive blown out whites and the lack of detail in the shadows, but they won't forgive skin tones that jump around.

50-65% of what Jon? Don't forget the 7D hasn't got zebras.

I agree completely about getting skin tones right in camera, but would add it is better to underexpose than over. If you overexpose skin, particularly caucasian skin it really looks terrible and is impossible to grade adequately.

Also, and I've mentioned this before get your white balance right. When it comes to skin tone It is so important not to blow the highlights, especially in your red channel.

True, the 7D doesn't have zebras (yet?), but some monitors have false colors.That would be 50-65% of full scale.

The Magic Lantern zebras (on the 5D2) have a 0xb000 (69%) level, which is on the high side. As I recall, Marshall's false color for faces is in the 55% range.

And, yes, WB is absolutely critical. A good exercise would be to do a test with a well controlled custom WB, and then to try some of the WB offsets to find your perfect settings for skin tones.

At NAB, one of the speakers at the Canon booth, who DPs romantic comedies, showed his "phases of the moon" test. He would shoot the leads with a single key light at various positions in order to study how their faces take light. This informs the lighting during production. A similar test could check various exposure and WB offsets that color their faces well.

At NAB, one of the speakers at the Canon booth, who DPs romantic comedies, showed his "phases of the moon" test. He would shoot the leads with a single key light at various positions in order to study how their faces take light. This informs the lighting during production. A similar test could check various exposure and WB offsets that color their faces well.

Pretty cool. Does he use stills for the moon testing? Or does he use the camera he'll shoot the film with?

As I recall, it was stills - which is perfect with a DvSLR.

I have not analyzed the Canon greyscale carefully on a waveform, but just using my eyes and looking at pictures it seems to me that you need to turn contrast all the way down to get something that would be considered a normal video contrast range. Even then it looks punchy to me, so I don't think with a Canon you can get anywhere near the low con of the film or cine gammas in Panny or Sony video cameras. So the question becomes kind of academic to me.

Leonard - you're right as far as in-camera adjustments are concerned. It's possible to go flatter though using a custom picture profile designed in Canon's picture style editor. It's not a very good app though, and the controls are fiddly making it almost impossible to get good results out of the gamma curve for anything other than minor adjustments. What these 'superflat' picture profiles end up doing is using a reverse-s curve on the gamma to stretch the highlights and shadows. The problem is that due to the nature of the application controls this can only be done by also flattening the midrange - which means things like faces end up getting compressed into a much narrower range of possible values.

Personally I think these profiles are a bad idea - to me the result is plasticky-looking skin which tends to posterize more easily once you start color correcting. I've found it best to stick to lowering the contrast in camera - but I don't always go all the way down depending on the subject. As others have mentioned, if the subject itself is low contrast you can actually store more information by raising the contrast in camera. It all comes down to the histogram - from left to right you have 256 possible values. If your histogram is bunched up such that there are flat, empty areas on either the right or left side - or both - there's room to store more information. You'll get the best image for post work by adjusting the contrast so that your highlights almost hit the right and your shadows almost hit the left but neither goes past into clipping.

I was unclear. I wouldn't set the highlights of the face no higher than Zone 5. I would put the meat of the face in zone 5. Shadows will be darker; highlights lighter. A bad bounce off my thinning scalp might even clip.

Got it, I misunderstood. I still tend to lean a little to the right with faces (depending on the contrast) because I like to avoid visible noise caused by bringing the shadows up in post - although if you're actually lighting it's obviously best to bring them up with light in the first place. I definitely agree with the skintone consistency thing - it's not just exposure but white balance that is important for keeping things consistent. I've found you get best results if you custom WB on the key from the face's position using an expodisc or similar.

One aspect of this conversation which hasn't been (explicitly) mentioned is the fact that whatever is being done in-camera is being done with more bits than you have in post (even if you move to a more robust intermediate for post work). So, there may in fact be some advantages to getting things "more right" in camera than in post. Of course, you'll always be doing both, but you get my point...

I have followed the general advice to use a flat response (actually the superflat preset) to give more options later in post. However, what I have found is that a number of clips have caused alot of grief, particularly skies where there is gradation between various shades of grey.

For example, I shot some clouds with the view to making them look stormy. So in post I increased the contrast. This left really unacceptable banding or posterization. Through trial and error I have found the best way to minimise this effect in post is to add the contrast at the point of conversion between h.264 and prores. Having said that I still need to process the clip with de-noise of smoothing plugins to make it acceptable.

I am now not entirely convinced that the picture style should be flat, but feel that the settings should be set as closely as possible to the desired final result, at least with regards to contrast. In hindsight, I should have adjusted the picture in camera to look how I wanted it.

I know the clips will still need to be graded in post, but is the best approach to avoid this type af artifact ?

The banding you are getting is as likely due to 8-bit processing as 8-bit content. I don't know if FCP has a 16-bit per color or 32-bit float mode, but if it does, give it a try. When I grade in Vegas in 8-bits, I get bands. When I grade in Vegas or AE in 32-bits, things are much smoother.

Another trick is to apply a light amount of noise reduction as a first step - in 32-bit 4:4:4 or 4:2:2, if possible. That will generate intermediate values. Do your grading next and only convert back to 10 or 8 bits at the very end - preferably with some dither or gentle noise added at the end.

I'm convinced that many people have condemned 8-bit capture over the years when the big problem is the combination of 8-bit capture and 8-bit processing. 8-bit capture isn't ideal, but when handled ideally, the results can be surprisingly good.

I have done some very very quick and dirty tests (not very scientific as I am not an expert on this field), but basically pointed the camera at some clouds under various settings / picture styles. Then I took this clips into FCP, added a brightness/contrast, whacked up the contrast to 70% and added a three way CC and boosted the midtones towards orange and the highlights towards blue (just to add some CC processing).
The clouds had a mass of graduated grey then a break underneath with bright sky.

The took clips of:

superflat - normal exposure - CRAP - loads of banding/posterization
superflat - under exposed - Better, but with slight posterization in the flat clouds
neutral - low contrast, normal exposure - More contrasty look, very slight posterization
neutral - low contrast, under exposed - Looks fine (but then it's a bits too dark)
neutral - max contrast normal exposure - Looks a bit over exposed in the highlights but clean
neutral - max contrast under exposed - Looks good and clean.

I then pushed the contrast up to 100 and observed the histogram in the scopes. The best setup (i.e. with virtually no gaps) was:

neutral - max contrast, underexposed

As I said, not exactly scientific, but I now would probably not use superflat at all now, and would change the picture style settings for neutral depending on what I was shooting and try and get the contrast as near to the final desired result, but allowing a little room in post.

In fact, I went over (at random) some clips I have that had this posterization / banding issue, and used exiftool to look at all the settings, and every clip I had an issue with used the superflat style.