Björn Benckert
Well-known member
Alright, I went ahead tested my theories myself.
Here's some 2K versions of the test images. Open these up in multiple tabs and follow along so you can A/B them on top of one another: Shot A, Shot B, Shot C, Shot D.
TEST ONE - LENS VS LENS
In this test, everything is kept a constant except one thing, the lens. These are both 35mm focal lengths at f/8 on Epic-W captured at 8K, 6:1 compression and ISO 1000. Shot A is a Prime, shot B is a Zoom set to 35mm. The camera did not move between the two shots. Shot B was scaled up just slightly to compensate for a larger FoV (for whatever reason) and repositioned so that when A/B-ing between the two shots, we're only concerned with the apparent shift in perspective that happens. As we know, scaling and cropping an image up is not doing anything to the geometry of the scene, so this is fine for demonstration.
This is essentially a recreation of the other test posted before.
You'll also notice the computer tower on the floor tucks under the desk in B and the image is overall more "stretched" spatially - the back row of pawns get smaller as the front row grow larger.
So, this concludes that even two lenses of the same focal length on the same format size can show discrepancies just as large as the ones we're seeing with these multi-format comparisons. So which of these two is the definitive "look" of this format? One is going to make a face look flatter, and the other is going to make a face look more 3D. Those are both tools we can use in our work, but neither one is an indelible characteristic of the format.
TEST TWO - LENS VS LENS w/ REPOSITIONING
We're going to use Shot A as our baseline, so refer to it above.
Now we bring in Shot C.
Shot C's setup is identical in every way to Shot B, the only change is that the camera was pulled back exactly 1" on a set of rails. From my preliminary testing I did with these two 35mm lenses, it seemed that a 1" was the magic number to match the two. I tested many distances in half inch increments and found that pulling back 1" gave the closest match. Could I have gotten it closer? Yeah, probably. But I didn't test smaller increments, nor did I do what I should have actually done, which would have been to locate the no-parallax point directly for each lens and do the math between them. Either way, 1" is pretty close.
So, pulling the camera back on a set of rails ensured the only camera position change occurred along that Z-axis and not X or Y, which would have added lateral/vertical parallax shift. Like Shot B, Shot C was scaled up to compensate for FoV differences so we can look directly at perspective shift and not be concerned with irrelevant (for the sake of this argument) details.
Comparing Shot A to C, what you'll find is that there are still discrepancies that exist, but minimized quite a bit to what we saw in Shot B against A. The interesting thing to note here is that Shot B had the computer tower in the background tucked under the desk more than A. But now, the computer tower emerged from under the desk more so than A. So by pulling the camera back 1", we've gone the other direction with the perspective shifting. The real number must actually be slightly less than 1".
So what does mean? This means we can use this knowledge when we compare multiple formats, because inevitably we'll have to use two different lenses to keep an equal FoV between shots, because that's what matters to us as cinematographers. As we already know, if we kept the same lens on both formats, we'd have two images with wildly different FoV's and zero perspective shift. Which brings us to...
TEST THREE - FORMAT VS FORMAT
Here's the meat of this test. This is what we're all dying to know.
Again, we'll use Shot A as our baseline. Just to refresh your memory, it's an 8K shot on a 35mm lens at f/8 and ISO 1000.
Enter Shot D, which is the same as Shot C (including -1" camera position), but the lens was zoomed out to 17mm and set to f/4. Nodal points can change as lenses zoom out, but luckily for this test, the lens I'm using's nodal point doesn't shift substantially across the zoom range according to a database I found, so our previous 1" compensation still can apply. The camera was set to ISO 250, 4K and 2:1 compression. Since 4K is an exact 2X crop of 8K, this makes our math easy, and we're in the right ballpark for all these settings. As like the others, Shot D was scaled up and positioned to align with the image we got out of Shot A. No other distortion correction or spatial corrections were applied. The color was slightly different due to aperture changes, so that was compensated for (though it's still a bit off) and a touch of sharpening was added since this 4K image isn't seeing the huge benefits of 8K supersampling.
So comparing our baseline Shot A against Shot D, we can see that these two images are far closer in 3D representation than A versus B, minus some barrel distortion. And those were both the same, larger format shots. This is a 2x crop of that. Monstro to Helium is 1.37x, Alexa LF to Alexa is 1.3x, Alexa LF to Helium is 1.23x. And Sony Venice to Helium is 1.2x.
While we're at it, let's look at C versus D. This is also an 8K comparison against 4K. This time it's the same zoom lens against itself in the same physical position in 3D space. D is a bit bubblier looking due to some wavy distortion. Again, still discrepancies, but the front element was 12" away from the front row of chess pieces. We're so close we're amplifying our errors in testing at this point.
So that's it. If I can show two 35mm lenses on the same format looking so different, and then show S35+ against S16+ that look closer matched than the two S35+ shots, then I don't know what I'm supposed to be looking for.
All it took for my example was moving the camera 1" back in z-space to get the no-parallax points to line up. So if that's the difference between making or breaking the "look", then good luck getting repeatable takes within an 1" margin. Your actors will screw that up for you in no time, as will your Steadicam ops and dolly grips. Of course, I'm being facetious; we all know that filmmaking is a lot more of an organic process than that. So whatever.
Don't get me wrong, there's clearly other advantages to large format. I just can't swallow this pill about "natural" perspective, whatever that means. One thread it means things look more stretched out, "rounder" and 3D, another thread it means that things look naturally compressed and less distorted.
¯\_(ツ)_/¯
Two different lenses of the same focal length can have very different nodal points. Simply the nodal point can sit further away or closer to the sensor even if you got 2 50mm lenses. Especially if you bring zooms into the mix.
So before you can compare them you need to find the nodal point and make sure also they have the same FOV. Scaling the image around is not a way to compensate for the difference of the nodal points.
Prime lenses are easer to make distortion free than a zoom lens. But it is alos related to the image circle and speed. Making a fast lens with little distortion is whats difficult. So yes lenses of the same focal lengths can look very different even though the got the exact same FOV.
So first you need to find the clitoris, I mean nodal point it´s a bit tricky for those who dont know how to do it
