Site icon ProVideo Coalition

Review: Panasonic AU-EVA1 4K Cine Camera, part 3

This is part 3 of a three-part review; there’s more in part 1 and part  2

[Updated 8 Jan 2018 with image of Xyla at ISO 2500.]

[Updated 9 Jan 2018 with links to Xyla source clips, however these links are not guaranteed to work forever!]

[Updated 11 Jan 2018 with clarification of HLG appearance on the camera’s display vs. its output.]

Performance

Dual Native ISOs, Dynamic Range, and Noise

Like VariCams, the EVA1 offers dual native ISOs, either 800 or 2500 (in 4/3 CROP&MIX 2.2K mode, these values are halved). These are “base” sensitivities with no highlight clipping and no added electronic gain.

You can set the camera to allow working with just the two native ISOs, or to start with either one and the variable ISOs based on it. In 800BASE you can dial in values from ISO 200 to 2000, while in 2500BASE you can go from ISO 1000 to 25600.

If you dial in a value below the base ISO, you’ll lose highlight headroom. At 800BASE, ISO 400 highlights clip a stop sooner than ISO 800 highlights, and ISO 200 highlights clip two stops sooner.

If you dial in a higher value, you don’t lose any headroom, but noise gets boosted, as is always the case: ye cannae change the laws o’ physics, Captain! As you boost gain, noise builds proportionally and naturally, with a tight grain pattern evenly balanced between chroma and luma noise.

How far you go before you declare the noise unacceptable is a matter of taste. I found settings of ISO 5000 or 6400 to be clean enough that noise didn’t call attention to itself. Above that, noise was noticeable enough that I probably wouldn’t go there unless I was shooting low-light documentary or I wanted the noise as an effect.

At gains above ISO 12800 I saw a slight shift towards green in the midtones and shadows, but it wasn’t anything that couldn’t be fixed in post (and if you’re shooting at those gains, you’ll probably want a noise-reduction step, too). Running auto black balance didn’t remove that color cast.

Native 2500 is noisier than native 800 is, but the advantage of dual native ISOs is that the higher native ISO is less noisy than it would be were it derived from the lower base ISO. ISO 2000 in the 800BASE range is noisier than ISO 2000 in 2500BASE range:

Sample UHD frame, with detail area highlighted.

Native ISO settings, 1:1 details:

ISO 2000 derived from both native settings:

In case that’s too hard to see, I did a second pass with an extreme grade, as if trying to pull things up about two or three stops (with no regard to highlights as all I care about is our detail area):

Sample frame with extreme boost.

Native ISO settings, boosted:

ISO 2000 derived from both native settings, boosted:

Panasonic claims 14 stops of dynamic range for the EVA1, and I found that to be accurate whether shooting at 800BASE or 2500BASE:

Xyla in V-Log at ISO 800. Count visible steps after the first two, which are clipped.
Xyla in V-Log at ISO 2500: Count visible steps after the first three, which are clipped.

(The Xyla 21 is a 1-stop-per-step transilluminated test chart designed for precisely this kind of test. I won’t bore you with all six scene file settings shot at both base ISOs. Let it suffice that I spend a lot of time in dark rooms squinting at pictures, so you don’t have to—or at least, so you don’t have to squint at Xylas.)

(You can download the 17 MB source clips for these two tests: ISO 800 clip, ISO 2500 clip. These links are valid as of 2018.01.09 but are not guaranteed to work forever.)

I was able to see 14+ stops in V-Log, HLG, and the two “EVA1 Look” gamma curves, and darned near 14 stops in the two default “video look” curves, albeit with those final two or three stops compressed within an inch (2.54 cm) of their lives. You wouldn’t want to shoot any of the WYSIWYG gammas other than HLG and attempt to pull a lot of detail out of the extreme highlights or deepest shadows, but it’s good to know that Panasonic seems to use as much of  the camera’s native dynamic range as possible to create those gamma curves.

Resolution

With a native resolution of 5720 × 3016 photosites, the EVA1 should have plenty of pixels to work with. The general rule of thumb is that you’ll resolve about 80% of the raw pixel count on a Bayer-pattern sensor, so the EVA1 should be able to resolve around 4.58K, comfortably above its 4096 maximum resolution. EIS crops the active area 1.15x and uses roughly 4974 x 2622 photosites, still enough in theory to yield a true 4K image.

The 4/3 crop mode uses 4520 x 2384 samples, mixed down (presumably) to 2260 x 1192, so we should still get something like 1000 TVl/ph in crop-mode FHD.

Mind you, all of these “raw” frame sizes need to be scaled down to 2160p and 1080p for recording. That pesky non-integer resampling completely messes up the simple math, adding all sorts of opportunities for resolution loss and/or added aliasing.

Here are 1:1 details from UHD clips, using the “S35 5.7K” sensor mode. 4K clips yielded essentially identical results.

4K / UHD recording, 1:1, internal recording.
4K / UHD with EIS on, 1:1, internal recording.

When I wobbled the camera slightly to see aliasing, it looked like the camera does indeed see a full 2000 TVl/ph vertically and diagonally. Horizontally it seems that the limits are closer to 1600 TVl/ph, though it’s hard to tell. But, seriously, this is angels-on-the-head-of-a-pin nit-picking. When I shot real-world images with the EVA1 and looked at them from about six inches away on a 49″ UHD screen (well inside the Lechner Distance), I couldn’t see any resolution defects or artifacts to save my life.

And yes, aliasing artifacts in the EIS case set in about 15% sooner, as you might expect from the 1.15x magnification in EIS mode.

S35 5.7K mode may also be used to shoot 2K/HD. The resulting frames looked pretty much like the 2:1 downscaling you’d do in post to make HD from UHD: they looked just fine. However I didn’t grab any samples.

The “S35 MIX 2.8K” mode is used for creating 2K/HD images with high frame rate capabilities:

2.8K HD, 1:1 sample, internal recording.
2.8K HD with EIS, 1:1 sample, internal recording.

It’s slightly more aliasy as you might expect (and EIS aliases 15% sooner still), but it’s not horrid. 750 TVl/ph or thereabouts.

4/3 CROP&MIX 2.2K is used for even higher frame rates:

2.2K HD, 1:1 sample, internal recording.

Yes, it’s more aliasy, but considering you’re shooting this mode at up to 240fps, it’s a fair tradeoff for the price. 600 TVl/ph or so?

And remember, almost nothing in the real world looks as crappy as test charts; the slo-mo clips in Matt Allard’s review or the Z Systems review give you a good idea of how both 2.8K and 2.2K sensor modes look on practical subjects.

Rolling Shutter

In Sensor Mode S35 5.7K, rolling shutter is less than 1/60 sec. For S35 MIX 2.8K, you’ll get half that: slightly less than 1/120 sec. For 4/3 CROP&MIX 2.2K, the sensor reads out in around 1/250 second.

VFR

The camera shoots 4K / UHS a 1–30fps using the 10-bit, 4:2:2 codec. If you switch to 4:2:0 8-bit recording, you can capture 1–60fps. It’s the same sensor readout; the data rate for the codec is the limiting factor forcing the switch.

FHD recording in the S35 MIX 2.8K allows 1–120fps using a 4:2:2 10-bit codec. If you switch to 4/3 CROP&MIX 2.2K, you’ll get 1–240fps, but it’s always recorded with a 4:2:0 8-bit codec.

All those values assume a system frequency of 23.98, 24, 29.97, or 59.94 Hz. If you’re shooting at 25 or 50 Hz, the rates top out at 25fps, 50 fps, 100fps, and 200fps instead.

V-Log and HLG

The EVA offers log recording with the same 14-stop V-Log curve used in the VariCams, DVX200, GH4, and GH5 (technically the last three use V-Log L, a 12-stop version). You can choose to use a V-709 LUT on the LCD, the SDI output, and/or the HDMI output for more eyeball-friendly monitoring.

HLG is also available: the BBC/NHK Hybrid Log Gamma curve. It’s an HDR (High Dynamic Range) encoding designed for live broadcast, and reasonable backwards-compatibility with existing SDR displays. Unlike Dolby Vision or HDR10, there’s no need for scene- or show-specific metadata, only an HLG flag in the datastream: it’s HDR For The Rest Of Us.

HLG records a normal, gamma-encoded signal for the “normal” bits of the image: anything up to about 100% in SDR shooting. Highlights above that are recorded in log. The switchover point is at about 50% on the waveform monitor, which means that on the camera’s own monitor the image has normal contrast and color but looks about two stops underexposed, albeit with an unusual amount of highlight headroom compressed into the upper half of the tonal range. Here’s a demo (using a GH5, because I didn’t think to stick the bright LED in front of the charts in my EVA1 tests):

(These are frame grabs from FCPX, interpreting the clip as “raw” Rec.709 material, reflecting the levels in the signal itself and corresponding roughly to HLG appearance on both the EVA1’s and GH5’s LCD monitors. Proper HLG signal scaling for SDR display—whether done in an NLE or in an SDI/HDMI output—is a somewhat fraught topic for a different day.)

Squirt that same signal out to an HLG-compatible monitor and you’ll see a properly-exposed image with those same highlights expanded back out to the limits of what the monitor allows, with unclipped colors and brighter highlight hits. Drag HLG clips into an HDR-aware NLE like FCPX 10.4 and they’ll similarly expand to their full range.

I hooked the EVA1’s HDMI output up to a Sony X900E TV (the only HDR display I have handy), and the Sony didn’t automatically sense an HLG signal and flip into HDR mode. However when I manually set it to HLG mode the signal displayed in all its glory. I don’t know if the EVA1 doesn’t output the HLG flag or if it’s just the Sony refusing to see it; a GH5 outputting HLG did trigger the Sony’s HLG mode automatically.

The EVA1 even offers a separate knee control for HLG, to allow customized control of highlight rendering. Both knee point and slope are tweakable.

Recording

The EVA1 uses dual SDXC cards (speed class 10 for 50 Mbit/sec, V30 or UHS 3 for 100 or 150 Mbit/sec; V60 for 400 Mbit/sec) and supports pre-record (5 seconds at 4K/UHD, 10 seconds for 2K/FHD/HD), relay recording, and simultaneous recording. There isn’t a separate proxy-recording mode, as there is with VariCams, probably since the 150Mbit/sec data rate isn’t so stressful that you’d really need a lower-bitrate proxy.

Normal recording uses 4:2:2, 10-bit long-GOP H.264 MOV files at 150Mbit/sec, though high frame rates will push you into 4:2:0 8-bit modes at 100 Mbit/sec for 4K/UHD or 50Mbit/sec for FHD. There are also AVCHD flavors for 1080-line FHD and 720-line HD; these are 4:2:0 8-bit codecs at 28 Mbit/sec or less.

Future firmware updates will allow a 400 Mbit/sec intraframe mode, and some form of raw output to Atomos and Convergent Design monitor/recorders (but not internal recording).

I spent a fair amount of time pixel-peeping internally-recorded files and comparing them to ProRes files recorded externally using a Video Devices PIX-E5.

With one exception, I didn’t see any appreciable difference between them: the EVA1’s recordings are pretty much visually lossless, with no obvious macroblocking, mosquito noise, loss of shadow detail, or other codec-related defects.

The one exception is that internally recorded clips are slightly softer than externally captured clips. The resolution appears to be the same, but there’s a minor loss of crispness; it seems very high-frequency data is muted, probably to make the codec’s job easier. It’s the sort of difference you’ll see in a back-to-back A/B comparison on test chart images, but it’s not something that calls attention to itself otherwise.

4K / UHD recording, 1:1, external recording (I grabbed this with markers enabled on HDMI, hence the center cross).

I saw a similar softening on the VariCam LT, too.

Somewhat confusingly—and confoundingly, for my make-the-codec’s-job-easier hypothesis—the lowly GH5 doesn’t soften its images noticeably when recording internally, yet its files are as clean and artifact-free as the EVA1’s. I set both cameras up to look as alike as possible (shooting the GH5 at ISO 400 to better match the noise of the EVA1 at 800), and shot the same exterior shots of foliage with erratic camera motion (the GH5 was bolted atop the EVA1 so the motion would be the same). I also shot the same greenscreen test and shadow-detail test with both cameras, then compared the clips frame-by-frame. There was no noticeable difference I could attribute to the codecs; camera noise and minor black-level differences swamped everything else.

EVA ROP

If you install the $160 AJ-WM50 Wi-Fi dongle in the USB port, you can use the EVA ROP  (Remote Operations Panel) app on your Apple or Android tablet to configure the camera and trigger recording:

EVA ROP displayed on an iPad.

(The Apple version also supports iPhone, but it’s essentially a scaled-down, pillarboxed version of the same 4×3 screen, and unless you have tiny, tiny fingers it’s really too small and cramped to be usable.)

It uses a totally different layout than the HOME button’s SBSP; it’s optimized for a single-page, tablet-sized touchscreen. The larger real estate of the tablet screen means you have more status readouts and controls available at a glance—and at a tap—than the camera’s side panel and monitor allow.

Along with up/down buttons, frame rate and iris have vertical sliders. The iris slider lets you drag the iris open and closed in manual mode, or set an exposure offset in auto-iris mode.

Tapping the MENU button brings up a menu controller with up, down, set, and exit buttons, but you must look at the menu displays on the camera itself.

EVA ROP in menu-twiddling mode.

When EVA ROP is running a few of the camera’s own controls are disabled, though their functionality is replicated in the ROP. The camera’s MENU button can’t be used to summon menus, although once they’re displayed the camera’s other controls can be used to navigate them. IRIS, AWB, ND buttons, and one-push auto-iris controls are also only available in the ROP.

The THUMBNAIL button is disabled when the ROP is running and no playback is possible other than a quick REC CHECK function. More critically, FOCUS EXPAND isn’t possible at all while ROP is running.

Those hiccups aside, it all works brilliantly, and I found the ROP useful enough that I often had it running on my iPad even when I was working beside the camera. It was simply faster to use the ROP than to use the camera’s own controls.

I should note that the ROP is a control interface only. The camera does not stream video over Wi-Fi.

Zacuto’s EVA1 Z-Finder

Panasonic sent me a trio of Zacuto products pre-assembed on the camera: a $595 EVA1 Z-Finder, a $195 EVA1 top plate, and a $130 Z-Rail Axis Mount to attach the finder to the top plate. The Z-Finder is (at this time) the only loupe sized for the EVA1’s 3.5” monitor, and the flip-up loupe and the solid metal frame convert the delicate LCD into a rugged EVF.

The assemblage adds 1 lb 10 oz / 746 grams to the camera, and it creates a very rigid structure: I had no problem jamming the loupe up against my eye and using that pressure to stabilize the camera. You can adjust the EVF side-to-side by a couple of inches, or you can replace the supplied rod with any 15mm rod to go as far out as you’d like. The Z-finder has two clamping positions, fore and aft, and you can rotate the finder anywhere you’d like, but there’s no provision for rotating the monitor sideways once it’s clamped in the Z-Finder’s grasp.

The loupe flips open to allow monitor-style viewing and touchscreen operation, though finger access is a bit cramped due to the depth of the frame.

Conclusions

There are plenty of perfectly good cine cameras in the $5K–$10K price range these days; it’s not like Panasonic is filling a gaping hole in the market. Indeed, we’re at the point where it’s arguably impossible to make a bad choice in this market segment, as all the cameras in this price range offer superb resolution, wide dynamic range, and some form of log and/or raw recording. Instead, the choice comes down to feature-set tradeoffs and to matters of taste, like color rendering and operational design.

Why, then, might you be interested in the EVA1? I see three main points in favor of the beast.

Practical 10-bit recording: In normal shooting, and even at moderate VFR rates, the EVA1 offers internal 4:2:2 10-bit recording at 150 Mbit/sec or less on affordable SDXC cards, with no obvious codec-related artifacts. The abstemious bitrates pay increasing dividends as your shooting ratios climb: you need less disk space and shorter data-transfer times for the level of picture quality delivered. The EVA1 offers both V-Log and HLG modes with 14 stops of dynamic range to fully utilize all 10 bits, too.

Panasonic color: Panasonic’s color palette is neutral and naturalistic, and it’s often compared to ARRI’s color. It’s an unfussy yet robust rendition that holds up well under variable lighting conditions and survives being pushed around in post. It intercuts well with VariCam; it intercuts well with GH5. Art Adams raves about VariCam’s handling of subtle blues and cyans and I saw nothing in the EVA1’s pictures to indicate it’s inferior in this regard.

VariCam-style operations: With a VariCam-inspired SBSP display on its HOME button and its use of “preset lists” for shutter angles, frame rates, and white balances, the EVA1 can be driven like a VariCam. The “Vari” in VariCam refers to variable frame rates, and the EVA1 delivers those too. Focus squares, FOCUS EXPAND, open iris focus assist, and the uncluttered monitor display only increase the family resemblance.

It’s a mental shift from using other cameras, but once you understand the control flow and take a few moments in prep to define your presets, you can fly though on-set camera adjustments with very little recourse to menus. Operating a VariCam is fast, and the EVA1 operates much the same way: it’s designed as an operator-focused camera first and foremost.

On the flip side, the EVA1 doesn’t offer continuous autofocus, and the one-push AF it does have isn’t really usable while shooting. Raw isn’t available yet, and when it does arrive via firmware upgrade it’ll need an external recorder. A PL mount costs extra, and many people will need to lay out considerable coin for viewfinding and support kit.

For some, those limitations will be deal-breakers; for others, not so much or not at all: de gustibus non est disputandum. If you’re in the latter camp (and if you’re read this far, you probably are), it’s worth getting some hands-on time with the camera and seeing for yourself what all the fuss is about.

The good news is that, if you don’t have local dealer or rental house with a demo camera, EVA1s are available from LensRentals at non-frightening rates.

Pros

Cons

Cautions

More Info

DP Matt Allard has an excellent hands-on review at newsshooter.com with many sensible operational observations (and good slo-mo and low-light demo clips, too).

AU-EVA1 webpage — links to operations manual, brochure, firmware, etc.

AU-EVA1 mini-site — many useful details, Panasonic demo films, list of tested NLEs, and more.

AU-EVA1 Brochure (PDF) — interesting specs that don’t appear anywhere else.

Video: EVA1 vs FS7 vs FS5 — comparisons of color, noise, and HFR rendering.

Article and Video: AG DOK Kameratest 2017.2: Sub 10K Kameras (German; open in Chrome for automatic translation) — comparison of Panasonic’s VariCam LT, GH5, and EVA1; Canon C200 recording raw; Sony FS7, Kinefinity Terra 6K, Blackmagic Ursa Mini Pro 4.6K; and RED Epic Helium 8K.


Disclosure: Panasonic supplied me with an AU-EVA1, Sigma 18-35mm lens, and Zacuto top plate / axis mount / EVA1 Z-Finder for review; all that gear has been returned to Panasonic. DSC Labs provided test charts. The PIX-E recorder was supplied as a development device when I worked for Video Devices. The other Canon and Sigma lenses, GH5, Zactuo EVF, and various bits of support kit were all purchased at market rates at various times over the past several years.

Exit mobile version