Re: Terminator 1 & 2
Okay, to answer Dave H's question with what I promised:
Let's start with some history...
Note that T2 was shot in Super-35, which means it used a full-aperture 35mm film frame (with image exposed full-width, sprockets-to-sprockets, rather than losing over 10% of the image width to a soundtrack area as in the Academy aperture). However, because it's composed for a 2.35 aspect ratio, it's only using the center 57% of the available negative area for active, composed picture. The rest is available for recomposition in postproduction or when doing 4x3 video transfers (which are largely obsolete these days due to HD). This would lead to the obvious charge of a grainier picture than films not shot in this format, but while this charge is mathematically true, the quality of the image all depends on how good your filmmakers are --particularly the DP, which in this case was the excellent Adam Greenberg-- and T2 looked pretty damn good on film. So let's start there.
The T2 master was originally transferred in 2003 from a Super-35 IP. Prior to colorist Mark Nakamine performing the transfer at IVC, Lightstorm screened their preferred 35mm print of the film in the THX-certified screening room at the Lightstorm offices in Santa Monica. This completed transfer was then screened for and approved by Jim Cameron. This transfer master was comprised of the Theatrical Version (in seven 20-minute reels) plus a few more separate, compiled reels of Special Edition scenes IP.
In order to take advantage of the full vertical "negative area" of the 1.78 HD format, the film (which was composed for a 2.35 aspect ratio) was transferred FULL-HEIGHT to a D5 4:2:2 HD master; that is, the film image that would normally take up the center 1920x816 portion of the 1920x1080 frame was stretched vertically to fill the full 1920x1080 frame, thus maximizing the use of the HD storage format. In normal usage, the film would have been just transferred in its original aspect ratio that would fit horizontally into the HD frame, with black letterboxing above and below. Doing a full-height transfer means that there are 1080 lines of resolution used in storing the image, rather than just 816 lines with the rest being black letterboxing. So the good news is that there is more image resolution; the bad news is that it means that the master has to be reformatted to the normal letterboxed 2.35 aspect ratio whenever it needs to be used. You cannot use the original transfer master directly. Fortunately, because it is digital and theoretically has no generational loss (other than from the reformat scaling itself), the image quality should not deteriorate significantly when creating a compression master.
From this 24psF full-height transfer master, comprised of 4 tapes (3 for the original Theatrical Version, plus 1 for the SE/ESE scenes), I created an additional pair of full-height masters: a 3-tape edited master of the Special Edition, and an ESE/Omitted Scenes edited master which consisted of the "T-1000 Searching John's Room" and "Future Coda" ending. These edited masters were digitally edited from the original Theatrical and SE Scenes transfer masters, so there was no generational loss. So now we have a total of seven final D5 tapes: a 3-part Theatrical master, a 3-part SE master, and an ESE/Omitted Scenes master, which are the current archival video masters for the film. All home video releases from the past six years were created from reformatted dubs/clones of these approved masters.
Also note that these masters have what is called SEGMENTED timecode, which means that the timecode on the tape is only continuous on each individual tape. So the first tape has Hour 1 timecode (the feature starts at 1:00:00:00) and comprises film reels 1-3, the second tape starts at Hour 2, etc. The significance here is that the timecodes on the tapes are NOT continuous throughout the whole film, since the reels 1-3 combined never add up to exactly 1 hour... it will usually be shorter (which means a gap in timecode when you string the film together) but may also be slightly longer than one hour (meaning there will be DUPLICATE timecode in that area). This will get addressed when the compression masters are made.
It's at this point where we did the manual dirt cleanup to the feature masters at Fotokem in late 2008. This process entailed doing a detailed QC list of dirt/scratch/hairs/watermarks along with timecodes for exactly where they are. The film then gets copied from the already-digital D5 tapes into a server, and an artist manually goes to each timecode on the list and paints out the offending items. For T2, at my request, they focused specifically on obvious artifacts like hair and scratches, and did NOT do any kind of automated grain removal or dirt removal. These processes do exist and are often used --like the "Dust and Scratches" filter in Photoshop-- but they are somewhat indiscriminate, since fine detail or grain may be mistaken by the system for dirt or scratches... which is why I specifically asked them not to use such tools on the T2 master, what with all of the sparks and explosion debris, etc. So after this labor is done, the feature is laid back out to a new set of D5 tapes, and we did some spot-checking by going to various spots in the master that were listed on the timecode dirt list and toggling between the before-cleanup and after-cleanup versions to make sure the work was done. This also made it possible for us to insure that no automated grain remove or noticeable noise reduction was done to the image, since this would show up when toggling. It's an asymptotic process in that once you get rid of the large dirt, the medium dirt starts standing out, and then when you get rid of the medium dirt, the small dirt is more noticeable, etc. We even went back and did some more manual work to certain areas for that very reason. Obviously, this gets pretty subjective, but what you do is the most you can do with the budget and time allotted.
Once the manual dirt work was completed on these these full-height masters, a set of continuous-timecode, reformatted compression masters are made for DVD and Blu-ray use. This means that the full-height image is run through a high-quality scaler to squeeze it vertically to the correct letterboxed aspect ratio, and the laid back onto a new HD D5 tape with the letterboxing. At the same time, a new timecode track is added so that the film has continuous timecode from start to finish, which is very important if you're synching things like PiP and text data to the feature. In some cases, if a movie is not too long, it can be laid off onto a single longplay D5 master (not the case with all of T2, of course). In our case, we had a single longplay Theatrical master, a two-part longplay SE master, and a short ESE/Omitteds master. These are the D5s that went to the authoring facility (in the case of the Skynet Edition, it was Blink Digital Studios in Burbank) for encoding.
The original plan was to use the AVC codec, but a lot of factors weighed in on this... one of which was the fact that in Blu-ray, in order to do Picture-in-Picture, both the primary (feature) video stream and the secondary (PiP) video stream need to be encoded using the same codec. Also, the amount of features we were trying to put on this single disc was a LOT --seamless branching of three versions, all of the audio tracks, etc. -- and most of it had to be in sync with the film, which means we could not put it off on a separate disc, even if the studio had agreed to do a second disc. Here's what Drew Hunstman, Senior Director of Tech Ops at Blink Digital where we did the authoring and compression, had to say about the codec issue:
"Our workflow is a pretty typical one for BD: capture a source file in 4:2:2 or 4:4:4 in V210 codec via Kona3 cards, using a slight NR at this point, if necessary; dither the color space into the appropriate YUV format; make a few encode tests based on calculated bit rates, adjust filters, tweak specific scenes as needed and encode; lather, rinse, and repeat until the final encode looks as good as possible based on client recommendations and bit rate usage calculations.
"The VC-1 choice was based on how the PiP looked in AVC, which was poor and in VC-1 which look far better. If we had used the AVC, to get a passable look for the PiP, the feature bitrate would have been a few points lower. As you know it is a balancing act making all the parts and pieces look the best possible while optimizing bit rates and still fitting them all in the BD box."
One thing I noticed about the comments regarding the picture quality on the forum was that most of the negative comments came first from the UK, where the European disc was being reviewed. The later comments that were a lot more positive seem to come when the US Domestic version was reviewed. This makes sense in the light of the fact that the International versions had a lower feature compression rate than the domestic, based on the greater number of languages that had to be accommodated on the international versions. Here's Drew's take on the disparity between the domestic and international encodes:
"The source, while good, was a D5 <4:2:2> not HDSR <4:4:4>, which a great many BD disks have for source material and can be superior in quality. Some discs like Ratatouille were processed directly from DPX files and so the source quality is even better than an HDSR. The European encode and the Domestic encode used the same source files captured from tape, the same pre-filtering and a moderate NR applied, but the 5 Mbps is making the greatest difference. Keep in mind that the amount of content on either disk kept the bit rate lower than many other discs on the market and the Euro was lower still.
"There is some NR applied at the capture point and some applied at the encode point. This is to strike a balance between the amount of grain that the encoder has to deal while maintaining picture integrity. We have found, as many other have, if we slightly reduce grain at the capture, then we need FAR less at encode point. And we do use the Digital Vision NR for some of this.
"In short, the picture differential between the Euro and Domestic is only about 5Mbps, however with the low bit rates due to the amount of content this makes a greater difference.
If the overall bit rates were higher and we had the same 5Mb differential the picture would vary somewhat less. Figure that the lower the bit rate, the greater percentage of the picture quality 5Mbs represents. It is about a 6% drop…using present settings. If the overalls were higher, the 5Mbs would be a less percentage drop in quality between the two."
Okay, so as I understand it, this means that a certain amount of digital noise reduction was applied during the precompression process --not as a creative decision, but in order to hit the target bitrates, which were determined by how much material the studio wanted to include on the disc. We specifically had the feature encoded twice at different rates to best accommodate the differeing amounts of content for the Domestic and International versions. Given that, I think we did a pretty good job. ;-)
The Geneon disc, which was made from the same transfer master (but did not have the added manual dirt fixes as the Skynet Edition), was able to be encoded in AVC and at a higher bitrate because it did not have to include multiple versions of the film or as many languages on a single disc, and also did not have PiP or other such features. The inclusion of the standard-def featurettes in the Geneon versions still takes up a lot less room than including branched HD footage and PiP and BD-Live and all of the stuff that's on the Skynet Edition, so the Geneon version was able to do a higher bitrate. The rule of thumb for compression is pretty straightforward: the more detail you want to preserve, the higher the bitrate has to be. The challenge --and the art-- of compression is to find the best balance, where you get the maximum quality in the minimum amount of space. This all comes down to how much content is going to be included on the disc. And as I've noted before, the studios try to balance how much they put on the disc in terms of new features (and the costs for doing so) against the sales projections of how many more or less they will sell if those additional features are not included. And most indicators --like the sales of SuperBit DVDs-- show that folks will choose feature-laden discs over movie-only discs even if they are technically at a higher bitrate.
Can we do better in the future? Most likely. The digital technology gets better and more efficient as time goes on, and hopefully next time the film will be retransferred to a 4:4:4 HDcamSR, or better yet, scanned at 2K from the original neg and restored digitally. And if we're lucky, we'll be able to do more than one disc and keep the feature at a higher bitrate while not sacrificing other features. If we're REALLY lucky, we'll have those multilayer discs with hundreds of gigabytes of storage on a single discs and no one will have excuses not to have the best quality encodes possible...
Does this make the Skynet Edition worth your purchase? Only you can decide that for yourselves. All I can do is try my best (when given the opportunity) to create compelling content to complement the film and to try to make sure the film looks and sounds as best it can within the parameters of the space available.
So there you have it... hope this answers some questions!
V
