Click To Enlarge

Last year, shortly after the Sony HVR-Z1 camera shipped, I borrowed one from Sony to test every low-budget filmmaker's dream: a blowup to 35mm. I used the Z1 to shoot a short film, Windsor Knot, and two different labs (DV Film and Heavy Light Digital) blew up HDV test footage to 35mm using the same negative stock and the same print stock. A panel of experienced film DPs then screened the results in the Main Theatre at the North Carolina School of the Arts School of Filmmaking. The results were quite revealing. The article appeared in the January 2006 issue of DV, and you can read it online at DV.com.

JVC HD100: indie film contender

Almost exactly a year later, I was ready to try the same type of test with HDV footage from a JVC HD100 camcorder. Two quick notes here. First, while JVC has now introduced the HD110, the HD110 and HD100 are essentially identical in terms of filmout. Second, while JVC has branded their solution as ProHD, the recording format is still HDV. If you faithful DV readers have been following the saga, you know the HD100 has produced intriguing test results in shoot-outs with other HD and HDV camcorders (for details, see Adam Wilt's tests in the May and September 2006 issues of DV, and online at DV.com). While all the 1/3-inch chip cameras are more similar than they are different, the HD100 brought in surprisingly good resolution numbers in these evaluations, especially considering that it is a 720p camcorder. The Sony and Canon units both record in 1080i, technically a higher-resolution format-at least on paper. But "on paper" and "real world" are often different things, and so it has proved with the HD100. The camera has also produced a lot of buzz and many fans in the filmmaking world, because it is 24p native (Sony and Canon both function in interlaced mode and use different technical fudges to create an artificial 24p output) and because it is the only camera of the three to offer a manual lens, which is generally preferred by DPs.

So, armed with an HD100 from JVC, I set out to create another short film. Titled Passing Moment, it is a narrative drama that explores the ways in which a patient's death affects a young doctor. It was my first-ever experiment with a truly no-budget film. All the actors were volunteers, the locations were free, and the costume items were borrowed. During my almost-30 years of creating video, I'd never actually done this before. It was an interesting experience, and truly amazing to find really good actors who would go far out of their way at their own expense to be part of a script they believed in.

Click To Enlarge Still life shot used to evaluate detail and saturated colors. I adjusted the Sharpness setting down and boosted the red channel to get the look I wanted out of the JVC HD100.

Camera tweaks: setup for blowup

Of course, optimizing the camera's menu settings prior to shooting required several important decisions. For this test, I'd decided to take my selected scenes to DuArt Film and Video, a venerable film lab that has a long track record and lots of experience with feature films. I'd already had some discussions with DuArt and various other labs about adjusting the HD100's settings, so first I turned Sharpness down to Minimum. This setting, technically known as aperture correction, artificially boosts the high frequencies in the picture-sort of like a stereo tone or treble control turned up too high. If not turned down, the resulting edginess increases artifacting and makes the picture look "video-ey," if that's a word.

Second, I boosted the red channel in the custom color matrix. For some reason, the factory settings have a very weak red component, which often leaves skin tones looking pasty and sickly looking. I increased the red to +3 and the overall color intensity to +5.

Third, I ran some advance tests with and without the stock Cinegamma setting, then tested the latitude (contrast handling) in different situations. The internal Cinegamma is a nice quick solution for output to video, but it is overkill for filmout and seems to decrease the camera's latitude. Additionally, a rule of thumb for both "film-like" and film output shooting is to be terrifically careful to avoid crushed blacks and overexposed whites. So after some experimentation, I settled on the stock master black with a black stretch of 1, and I set the knee to manual with a setting of 85 percent. For those unfamiliar with these functions, the black stretch lifts the lower end of exposure to preserve details in the very darkest areas of the picture. This avoids having dark areas "crushed" or solid black without detail. The knee compresses the upper end of exposure, preventing early overexposure and white clipping. These settings combine to create an artificial but very practical expansion of latitude.

Shot selection

Upon completion of Passing Moment, we selected a few scenes for the filmout test, including a couple of perfectly exposed shots and a couple of shots that had been underexposed and had to be stretched in post. We then shot a highly saturated still test with some fine detail (similar to a shot tested in the Z1 blowup last year) and an exterior panoramic shot that included a pond, grass, and lots of trees-in short, a lot of detail. I added a couple of greenscreen shots that were just a shade off from perfect to see how they would look on the big screen. My concept was to include a range of scenes: some that would show off HDV footage at its best, some that were a notch or two off from perfect, and some that would stress the codec with lots of detail. For comparison purposes, I included a couple of short clips from last year's Z1 footage.

During this project, I was able to take advantage of DuArt's direct experience with the HD100. The company has worked closely with JVC to optimize a workflow for performing filmouts from HD100 footage. Andy Young, DuArt's vice president of special projects, has been through the process himself-shooting a number of his own projects with the HD100 and blowing them up to 35mm. Andy was very helpful, and far more concerned about some of the technical details of the footage than even I had been. DuArt prefers to work either from camera-original M2T files or from uncompressed files, so that they have the maximum latitude for color correction and optimization of the files. The edit of Passing Moment had already been completed in the CineForm codec, so I exported the scenes as an uncompressed file (the 2-minute clip took up a full DVD) and rendered the clip out as an M2T. The results were truly fascinating.

Click To Enlarge A scene from Passing Moment. Skin colors are natural, detail in hair is good.

The results

Once again, the Main Theatre at the North Carolina School of the Arts School of Filmmaking provided the big-screen venue for the test. In addition to the audience of staff and students, several local filmmakers--including Arledge Armenaki (The Howling V, Blackout, Club Fed) and Richard Clabaugh (Children of the Corn IV, Phantoms)--offered their expert eyeballs and comment. The overall reaction was quite positive. Carefully exposed medium and close-up shots looked absolutely beautiful, easily as good as those commonly seen in the average 35mm theatrical distribution print, and better than some released films shot on 16mm. Color was natural and matched the original files very accurately. One complaint I've had about film prints from video sources (done by several different labs) is poor black density and detail, but this print had both good density and good detail in the shadowed areas. Skin and hair detail were well preserved also. In short, DuArt did a splendid job on the blowup, giving the submitted footage its best chance to look good on the big screen.

Generally speaking, the DPs (and the many students who attended the screening) were quite impressed with the blowup. Last year, Arledge Armenaki compared the HVR-Z1 blowup performed by Heavy Light Digital to a good 16mm blowup, but felt it fell short of Super 16. For this experiment, he compared the HD100 footage to a good Super 16 blowup, but noted that it visibly fell short of 35mm quality in the wide shots.

Surprisingly, the underexposed scenes that had been stretched in post (and over which DuArt's Andy Young had expressed concern) did not look bad at all. A definite increase of noise occurred in the darker areas, which was then exaggerated by the film grain, but not to a level that it compromised the scenes. Although most shots held detail and color well, the panoramic shot of a pond and trees, which contained lots of detail, looked soft-as if the focus was slightly off. While not bad enough to attract the untrained eye, it was different enough from other shots to receive comment and complaint from most of the camera operators and DPs present.

When last year's Z1 footage received similar comments, the assumption was that the camera operator had just not quite nailed the focus. Without a full-resolution HD field monitor, getting the focus right in HD can be a difficult feat. This was the immediate reaction from several viewers at this screening as well, and seems to be the prevailing discussion in Internet forums. You'll see lots of comments online about "soft-focus" issues with HDV footage. However, my subsequent tests suggest that this effect is caused by a combination of factors beyond the operator's focusing skills, including the camera's lens, the CCD function, and the HDV codec itself.

Devil in the detail

HDV, like DV, is a format with a constant bitrate that must be rigidly maintained. The format works differently from DV (which uses only intraframe compression similar to JPEG), because it also introduces interframe compression over several frames. The base HDV format calls for a 15-frame Group of Pictures (GOP) with only the first being a fully digitized actual frame. Subsequent frames are described by changes from the initial, or I-frame. This is a stupendously efficient compression method, allowing a 2-hour movie to fit comfortably on a DVD that can accommodate less than 2 minutes of uncompressed HD-but it comes at a price.

Click To Enlarge This green screen shot has problems when composited with a single-pass keyer. The sizzle on edges and schmutz over the background are subtle enough to be acceptable in SD but when blown up on the big screen become quite noticeable.

When presented with a scene of great complexity (lots of sharp detail), especially one that contains motion, the codec must find a way to compress that GOP into the fixed bitrate. While DV codecs will produce color blocking (posterization) and the familiar mosquito netting artifact around overly sharp edges, the MPEG-2 codec will tend to soften detail so that the GOP will not exceed the fixed bitrate. In many cases, this softening is barely noticeable; in others it is marked and obvious, especially when blown up to the big screen.

Such was the case with the pond and trees shot from the HD100. Exactly the same issue occurred in the opening wide shot of Windsor Knot, shot with the Sony HVR-Z1, which was one of the Z1 test shots that I had DuArt include in this year's reel. Both shots looked like the focus was soft, but the camera operator carefully focused the pond and trees shot during shooting, using both lens marking and focus assist (peaking display) to check. And in fact, at f/8 on a 1/3-inch chip camera, it is pretty hard to get a wide shot out of focus.

Click To Enlarge This green screen shot was composed without up-res'ng the undersampled 4:2:0 chroma. While most of the picture works fine, certain transition angles (notably the young hunter's shoulder) show visible jaggies on the big screen. Compositing destined for the big screen must use some method of smoothing jaggies and must be done with great care for detail.

Most of the test footage looked good enough on the big screen to show anywhere to anyone, but the panoramic shot was a visual problem-and an issue that HDV filmmakers need to take into account. Too much detail, especially detail with motion (waving leaves and branches, rippling water), will introduce some softening to HDV and the skill of the camera operator cannot always compensate for it.

Detective work: codec vs lens

This raised the question: how much of the softening in this test shot with the HD100 was truly the codec and how much was the stock Fuji Th16x5.5BRMU lens? This lens, while nice for the price, isn't a high-end HD lens. It performs satisfactorily in the middle range, but the extremes of the zoom range exhibit pretty marked chromatic aberration (CA). CA is clearly identifiable in macro close-up shots. However, in wide shots, CA may appear to simply soften focus, especially in a scene like the pond-field-trees shot, which had few hard-edged white details. Plus, it seems that the lens simply has a lower resolving power at its full wide position.

Click To Enlarge Comparison of the pond and trees shot from JVC HD100 (left) and Canon XL H1 (right) captured to Wafian disk recorder. The Canon was captured via HD-SDI at full 1920 x 1080 and scaled down to 1280 x 720 to match the JVC footage.

I did some side-by-side tests with the Canon XL H1 (which clearly has a better lens, even if it is servo-controlled), including capturing footage from both cameras into a Wafian HR-1 disk recorder. The comparison was instructive, particularly when capturing from the Canon at full 1920 x 1080 via HD-SDI.

The difference between the full 1080i capture from the Canon and the Y/Pb/Pr component 720p capture from the JVC was astonishing. Even when scaled down to 720, the Canon footage held significantly higher detail. The Canon footage exhibited clear CA at the edges, but not toward the center. Unfortunately, 60i footage isn't ideal for film transfer. The difference narrowed substantially in similar tests in Canon's 24F mode. In tests of the actual HDV capture of the camera, the difference also narrowed to the point where I was convinced the difference was simply a better lens.

It would be really interesting to test a Canon with the factory alteration allowing 50i operation, captured at full 1920 x 1080 on the Wafian, then deinterlaced properly, converted to 24 fps, and output to film. The results would be stunning, but the process would be quite inconvenient. There are rumors of a true 24p camcorder from Canon, but at the moment, this information isn't confirmed or official.

So I was still left with the question of how much of the softening was contributed by the HDV codec rather than the lens. I took the pristine 1920 x 1080 HD-SDI capture from the Canon, scaled it to 720 and recompressed it to HDV, and then examined frames from the resulting M2T file. Only minor degradation occurred: a slight color shift, increased blocking in dark shadowed areas, and a very tiny amount of softening. However, the differences were minimal and significantly less than I had expected. So once more, HDV has made me eat crow: my theory that the softening of the wide detailed shot was primarily attributable to the HDV codec was wrong. Most of the softening appears to come from the stock lens, and the same would apply for the similar problems with the wide footage from the Sony camera last year.

Unfortunately, I could not get one of the upgraded Fujinon Th13x3.5BRMU 13:1 wide angle lenses for this test. I hope to get one soon, so I can do side-by-side comparisons and report the results. Those who have done tests of the 13 x lens report that it is substantially better than the stock lens, particularly at the extremes.

Compositing concerns

The other area of concern for film blowup was the greenscreen footage. I specifically selected the test shots because they looked nearly perfect in SD but exhibited visible problems when viewed at full resolution on a good HD monitor. When blown up to film and then projected on a theater-sized screen, these problems become quite eye-catching.

One of the scenes was difficult to do in a single pass (the way most folks will create a color-based matte composite) because it included a shiny sword that reflected the green backing color at a high level. I composited this scene using a single pass of Primatte, but the compromise settings that allowed the sword to remain visible also left what I refer to as schmutz (small but noticeable amounts of noise in areas that should be clean background) and sizzling edges in other areas of the picture. Compositing this scene cleanly for the big screen would typically require a multilayer precomped matte; at a bare minimum, separate core and edge mattes would need to be pulled, probably using different keyers for each.

In a second instance, I used Keylight, one of my favorite keyers, but it doesn't have a facility to up-res undersampled chroma to see how the 4:2:0 HDV color component would work without help. While most of the picture actually looked quite good (the actress's hair is excellent), there were certain areas where the jaggies showed up very visibly. A keyer such as Primatte, which has an excellent chroma up-res function, will produce better edge results using 4:2:0 footage.

This hammers home the fact that producing for film blowup requires substantially better craft and more attention to details than production for the small screen, especially the standard-def screen. Minor pixel-level mistakes that disappear into invisibility in NTSC SD are blown up to the size of a small third-world nation on the big screen and become glaringly obvious.

However, don't get the impression that you can't do good compositing in HDV. You can, but the big screen requires some effort. Passing Moment used one color composite in the final edit (the main character driving in her car at night), and it turned out quite well. However, it was done with precomped multilayer core and edge mattes, and the edge matte used a chroma up-res to smooth the 4:2:0 undersampling.

The bottom line

This test showed that the JVC HD100 has substantial chops for low-budget productions that hope to eventually achieve distribution and a 35mm blowup. When the camera is properly tweaked and operated, most scenes will look quite good on the big screen. This would be particularly true for an actor's film, where the vast majority of the story consists of medium and close-up shots of the cast. Without a better lens, the HD100 is clearly not the first choice for a film that is full of sweeping panoramic shots with lots of detail. However, there are both better lenses available and adapters for those better lenses (though most of these adapters work better for telephoto than wide angle). Some ground glass adapters, such as those available from Redrock Microsystems or P+S Technik, allow the use of 35mm lenses on pro video cameras.

The test also confirmed that close communication and advance tests (using the intended camera settings and lens) with the chosen lab are essential-repeat, essential-to get a finished product that stands up to big-screen projection. And finally, the old GIGO computer axiom holds true: Garbage In, Garbage Out. Your production craft must be excellent and your attention to detail high to achieve the best results.

The Rev. John Jackman has been involved in video production since the mid-1970s. You can read more about edge and core mattes in his new book on digital compositing, and you can reach him online in the DV.com forums, where he moderates the Camera Forum and The Craft of Lighting Forum.