After evaluating some aspects of the image quality of Sony's new NEX-7 it's now time for a closer look at the video quality by comparing it to the current flagship of the MicroFourThirds (mFT) system cameras, the Panasonic Lumix GH2.
In order to compare it to a GH2 with current V1.1 firmware as well as to a GH2 with a patched V1.0 firmware, three cameras were taken to a side-by-side video shootout. As a movie speaks louder than words, let's quickly come to the video result and take some time for further explanations and a conclusion afterwards.
First an important note:
The Vimeo video that you find embedded here is NOT applicable to judge the quality of the ouput of these cameras due to the fact that the original FullHD 1080p video was rendered with 28 Mbps to 25 frames per second (fps) and the video that Vimeo streams is re-rendered by Vimeo at about 4.5 Mbps. As a result, you will see a video with less details and a lot more artifacts. So, to get a realistic impression, you will have to download the master file that was originally rendered and uploaded to Vimeo. If you are a registered Vimeo user, you find a download link on the Vimeo video page (please right click that link and select "Save to..." in order to download that file). This resolution and this high frame rate requires very powerful ressources from your playback equipment. Only systems that support hardware H.264 decoding on their graphics card will be able to play it without judder. The MediaPlayer included in Windows 7 supports many hardware accelerations of today's graphic cards, so chances are good with that player. VLC's support for H.264 hardware decoding are still experimental and must be explicitely activated in the VLC player's settings. If you have very powerful ressources you may try the 1080p 50 fps @ 32 Mbps version as well, that you find here: http://vimeo.com/35823620 (but please also download the source file, as the Vimeo rendering to 30 fps is awfully choppy with a lot of judder). For the sequences that were recorded initially in 25fps, the 25 fps version is fairly enough, only the 50 fps sequences loose some information accordingly.
Setup and recorded content
There were three cameras used, one NEX-7 with Voigtlander Nokton ASPH 35mm/F1.2 (adapted with a Novoflex adapter), two Panasonic Lumix with Leica GH2 25mm/F1.4 DG ASPH. The focal lengths resulting from the sensor crop factors have a fullframe equivalent focal length of about 53mm compared to 50mm but due to the larger area used on the GH2 multiformat sensor for 16:9 videos, the standard cropfactors (2 vs. 1.5) can not be applied. You recognize this when you compare the size of the objects in the side-by-side videos. One GH2 contained the current firmware version 1.1 with the new "high bitrate" mode (1080p25 @ 24 Mbps), the other one contained a hacked V1.0 firmware with a patch for the 32 Mbps 1080p24 mode. There are three sequences in which each time the NEX-7 is compared with one of the GH2s mounted side-by-side on a tripod. The cameras were set to default contrast, sharpness and film mode at basi ISO (160 for the GH2, 100 for the NEX-7), white balance was set to AWB.
In the first sequence the 720p50 mode of the GH2 (V1.1 with 17 Mbps) is compared to the 1080p50 mode of the NEX-7. The section of the GH2 video was subsequently scaled up to 1080 lines. The shutter speed was set to 1/100s. The second half of the sequence contains a relatively swift swing as panning for the passing barge, which clearly benefited from the higher frame rate. The aperture was set at F5.0 on the GH2 and to about F4.5 at the NEX (due to its lower ISO sensitivity).
In the second sequence the 1080p25 modes (each camera with 24 Mbps) are compared. The shutter was set to 1/50s, aperture F7.1 at the GH2 and appxoximately F6.3 at the NEX-7.
In the third sequence the 24p cinema mode of the GH2 (V1.0 wit a "mild" 32 Mbps patch) is compared to the 1080p25 mode of the NEX-7 (24 Mbps). The 24p video is sped up by 4% in order to preserve a smooth rendering without the sporadic inserted frame repetitions when rendering at 25 fps or 50 fps. The two sequences were cut to one another such that, in the middle part they are in about the same time.
The GH2 provides a visibly clearer video at 1080p, the gap increases somewhat with the use of patches to higher bit rates. The NEX-7 also produces visible moiré (e.g. on the brick walls of the channel-celled opposite wall). The NEX-7 videos look a bit warmer whereas the GH2 have more contrast - however, these differences can be relatively easily compensated in the postprocessing. Already with the standard (unhacked) firmware 1.1, at 50fps, the GH2 delivers with 720 lines (subsequently scaled to 1080 lines) nearly the same image quality as the NEX-7 with 1080p50. The NEX-7 profits by it's larger dynamic range and preserves highlights and shadows better. In terms of detail resolution, further improvements can achieved on the GH2 with current firmware hacks. So the NEX-7 has quite a large backlog in the video section compared to the GH2.
One additional note regarding the display and the viewfinder: The live view of the NEX-7 with default settings is significantly darker than that of the GH2 as it can be compared at the end of the video. The preview image of GH2 corresponds more realistic to the final recording. This resulted in significant overexposures during the first experiments with manual settings on the NEX-7. Therefore I recommend to raise the brightness of the display of the NEX-7 in the menu in order to receive a more realistic live-view.
As some readers had doubts about correct focus selection or bad influence of a too shallow depth of field due to the slightly larger APS-C sensor and the need to select an about 1/2 step wider open aperture on the NEX-7 due to its lower ISO sensivity, I went back to that location to take a series of still photos with the same focus and aperture settings as used in the video (simple to reproduce because the focus of the manual lens was set to the end position for "infinite"). The images were bilineary scaled (without sharpness transformation) to 1920x1280 pixel afterwards and finally cropped to 1920x1080. Here you can see the result:
The image was taken with F4.5 which was the most open aperture used during the video test recordings. You can see the sharpness that potentially could have been achieved and you can see that unsharpness due to out-of-focus rendering were nothing to doubt about. Further images taken with other viewing directions can be seen here.