V-LOG L: A Bit Bucket Half Full
One guy’s take on Panasonic’s V-LOG L profile after beating on it for a few months.
- So many opinions and tests are already out there. Do I really have anything to add?
- I wanted to make sure that I knew what the hell I was talking about before I did so.
I finally feel comfortable that I understand what’s what, and that I have something worth saying. For the purposes of this post, I’m going to posit that there are basically three good reasons to shoot with a LOG profile:
- Capture more dynamic range than camera profiles pre-compressed for REC709 display
- Better color reproduction than “stylized” camera profiles
- Allow for more flexibility to manipulate image in post
How well does V-LOG L achieve those goals? It turns out there are two different answers depending on how you record it.
WHAT IS V-LOG L
I’m not going to try to explain what a logarithmic profile is, or exactly how it works. To do so correctly would take thousands of words, and if I simplify the explanation then people will complain that I’m not being accurate. Besides, there’s already a lot of published information about logarithmic image curves…and some of it is even good!
It does bear repeating, though, that LOG profiles are as much art as science. This is evidenced by the fact that there is no one LOG profile. Canon C-LOG, Sony S-LOG and Panasonic V-LOG are all different animals. Even a single manufacturer (Sony especially comes to mind) may have several different LOG profiles. I mention this because Panasonic decided to make the V-LOG L curve compatible with the V-LOG curve of the GH4’s big brothers in the Varicam line. While this decision may make it easier for a tiny few people to match the two cameras, unfortunately it exacerbates an already challenging situation. There’s really no denying it…V-LOG L is a significantly compromised LOG profile that brings with it a number of challenges that other, arguably better designed LOG profiles (such as C-LOG) don’t suffer.
BY THE NUMBERS
Take a minute to familiarize yourself with the chart below.
- The horizontal axis represents 16 stops of dynamic range. The midpoint, zero stop of under or over exposure, equals 50% grey exposed at 50%.
- The vertical axis represents the available data range. Recorded internally, the GH4 has a bit-depth of 8 (or 256 steps between black and white). Recorded externally, the GH4 can utilize a bit-depth of 10 (or 1024 steps between black and white.)
What can we learn from this chart? I think a few points are worthy of highlighting:
- Note how the differences in the gamma curve between the standard profile and the Cine-D profile are relatively small, yet anyone who has shot with the two knows they yield very different images.
- Now, note just how different the V-LOG L curve is from those two, and how close it is to a linear response. If you don’t know what this means: you have to drastically manipulate a V-LOG L image to make it look natural.
- Note how much of the available bitspace V-LOG L wastes. More on this later.
- Note how V-LOG L protects about half a stop of extra dynamic range in the shadow area, and about a half a stop of extra dynamic range in the highlights (compared to the Cine-D profile)
- Note how V-LOG L protect about a whole stop of highlight data more than the standard profile.
- Note how Cine-D actually protects about a half stop of highlight data more than the standard profile.
But what does this mean? Specifically, how does V-LOG L’s response impact our three stated goals for using a LOG profile?
Let’s start with the good news…extra dynamic range. Based solely on the chart, you might think that half a stop of highlight protection isn’t a big deal. But remember, f-stops are measured on a logarithmic scale themselves, so each stop is twice the amount of added brightness. A real world image demonstrates the benefit:
Note the view out the back door, and the corner of the pool table behind the bananas. At the exact same exposure, Cine-D blows the highlights, but V-LOG retains a bit more usable picture information. Is it enough to justify the hassle of shooting with a LOG profile? That’s for you to decide.
V-LOG L wastes a lot of the available bandwidth. It wastes even more than would otherwise be necessary because it uses the same curve as V-LOG (which was designed for two extra stops of dynamic range.) In fact, it wastes almost exactly 30% of the available bit depth. If you’re using an external 10-bit recorder, this is not too big of a deal because you have 1024 steps to work with. Even with the 30% loss, you still have almost 3 times as much data as a perfect 8-bit image. Really? Yup. 10 versus 8 may suggest a small difference, but actually 10-bit images have 4X the amount of color resolution. (See my past post on the three different types of resolution.)
There’s very little visual impact when reducing 1024 steps to 716, but when reducing 256 steps to 179, the loss is much more noticeable:
- Top ramp is 10-bit.
- Middle ramp is 8-bit.
- Bottom ramp is 8-bit V-LOG L.
And this is a best case scenario. Assuming you’re shooting to protect shadows or highlights, you’ll have even less data to work with.
Charts are great and all, but nobody uses their GH4 to shoot greyscale ramps. What’s the impact on a real world image? Paul Leeming (a talented DOP and author of the great Leeming One LUT) graciously donated two images that illustrate exactly what this means:
Natural gradeV-LOG L, REC709 LUT
Saturated gradeV-LOG L, REC709 LUT, saturation pushed
Now, what happens if we take that already starved 8-bit image and stomp it with heavy compression, such as the 100 Mbps H264 codec the GH4 uses for internal recording?
This is a real word example. It’s the wall in the back of the image with the pool table.
- The right column was recorded at 10-bits using the Prores HQ 422 codec. The left column was recorded internally.
- The top row is straight V-LOG L. The middle row was processed with a REC709 LUT. The bottom row has had 25% saturation added.
Ouch. What the hell happened?
Simply put, the codec is doing what it was designed to do. It is using much more compression on areas of low “frequency” (detail). It does so because it knows that our eyes aren’t particularly sensitive to this. Compression introduces errors, and in the case of H264, these manifest as large “macroblocks” being averaged to the same color. In areas that are also not very saturated, slight color hue errors are introduced from block to block.
The problem is that in a LOG image, everything is desaturated, so you’re going to end up with a lot more errors. And then, you guessed it, you have to heavily process a LOG image to make it look natural, so, WHAM, you can’t help but exaggerate the defects. Double ouch.
If you’re using an external monitor/recorder like those from Atomos, Convergent Designs or VideoDevices to capture the GH4’s 10-bit signal with a good codec, V-LOG L can be useful tool for saving dynamic range. Not only do those products protect the image, but they allow you to preview what the image will really look like (with a LUT) and offer focus and exposure tools that are really helpful when trying to work with a flat LOG image.
Recording V-LOG L internally is risky. You may get lucky and not have any glaring problems, but you should expect trouble with common scenarios. The only reason to risk it is if you need that extra bit of highlight protection, as the net is that the extremely fragile 8-bit image is actually less gradable than one of the more natural profiles. (At NAB, even Panasonic admitted as much.) Add to this the fact that the GH4’s built in focus and exposure tools are compromised when shooting V-LOG L, and shooting V-LOG L internally is just a big bag of hurt.
- Wait, what about using a LOG profile to get better color reproduction? V-LOG L offers 0% more accurate “color science” when compared to the Cine_D profile. None. Nada. Zip. Love it or hate it, it’s the same. (Reference the first set of images.)
- Why didn’t you mention the extra shadow detail? Because it’s not usable. The GH4’s shadows are very noisy. What little you gain, you’re gonna want to crush. (This is why some people think V-LOG L is noisier than other profiles. It’s not, the curve just makes the noise easier to see prior to correcting.)