I have heard good things about Prelude CS6, another brand new addition to the suite, but I haven’t really wrapped my head around its capabilities.  It is a media ingest tool, but most of those features are already available in Premiere Pro, so I guess they are just trying to make them more accessible to non-editors.

Speaking of Premiere Pro, the CS6 version has a few new features, most of which are under the hood.  Adobe Mercury Transmit changes the way the application interacts with 3rd party I/O cards, which should improve both stability and performance on professional level workstations.  More streamlining of the underlying media engine software will offer smoother playback, especially at larger frame-sizes and higher frame-rates.  In my opinion, the biggest improvement in the CS5.5 release was undocumented: RT playback of multilayer 2K image sequences.  This was very helpful on the last steps of the Act of Valor online, since CS5 used to drop frames after a minute or two of 2K DPX playback.  I saw CS6 playing back 4K that way at NAB, so it appears that the software pipes continue to get bigger, to keep up with the new advances in hardware capability.

After Effects CS6 has a whole new ray-tracing 3D composite engine available, which I am sure is very helpful for advance visual effects work, but that is not something I usually delve into.  I am most excited about a relatively simple feature, variable mask feathering.  This single feature has been a major limitation compared to competing products for many years.  The new 3D camera tracker will allow simple compositing shots to be done faster, or at least give you a starting point to work from.  Better frame caching will also be helpful, to take advantage of all that RAM that is so cheap these days, as well as new card based SSDs.

Overall, it would appear that Adobe has not been sitting on their hands for the past year, but luckily their products have developed to a point where they no longer have gaping holes that absolutely require immediately upgrading to the newest version to alleviate basic workflow problems.  I am sure I will adjust to the newest version eventually, but I see no driving need to upgrade currently active projects until they are completed.  (Unless you are an After Effects artist, in which case, CS6 could change your life.)

Our solution was to do our entire post process on the full 16×9 image, with software letterboxing applied for monitoring.  This allowed us to maintain the full scope of our image throughout the post process, but required some creative project management.  All reframing had to be deferred until after we split out the versions for different aspect ratios.  That meant that all titles had to be applied after that point as well, so they wouldn’t get repositioned out of sight in 2.39.  And any changes we made to the movie after we split into different versions would need to be made to each version, and carefully tracked.

Using Premiere Pro CS5.5 to do the final online of the movie provided us with an unusual solution.  We kept the full aspect ratio footage in a source sequence, which is where we would usually add changes and VFX shots.  As we further processed the footage, new versions would stack up in this sequence. (Original shot, Twixtor render, colored-corrected export, etc.)  We would export from this sequence to send footage to color correction, or our texture and density passes.

Using a separate master sequence, with that source sequence as the main source clip on the timeline, I sliced up that instance based on timing from an imported offline EDL. (PageDown, CRTL+K, repeat)  Next I duplicated that sequence of clips across three layers, one for each aspect ratio.  I then went through one layer at a time, and used the motion effect to zoom in or shift any shot that needed adjustment for that aspect ratio. This was usually vertical shift for 2.39, zoom for 1.78, and horizontal shift for 1.33, but varied greatly for different shots.  For the 239 version, I brought in an Avid export as a guide-track, with a garbage matte to split the screen, so I could easily match the reframing Scott had done in the Avid.  For the other two aspect ratios, the focus was on cropping out anything that didn’t belong in the shot, and I reframed based on what I knew Scott’s intent was for the shots.  We also added the subtitles and other overlays on higher layers on this master sequence.  All preview outputs for review, and the final DPXs, were exported from this sequence.

When it came time to prepare slightly altered cuts for certain markets, we were able to do that by further using the existing “master” sequences as source in new timelines.  It was easy to trim out sections, and usually the new clips were added to an “additional source” sequence, which was then cut into the new timeline with a separate layer for each aspect ratio.  So as long as the correct aspect ratio layers were selected in BOTH the “master” sequence and the new altered timeline, it all synced up perfectly.

This allowed every change in the movie to be easily propagated across all versions automatically, which both saved time and prevented mistakes.  I was able to do simple dust busting fixes in Photoshop, even at the last minute, and as long as my changes were made in the source sequence, it would be corrected in every aspect ratio, of every version of the cut.

With this framework in place, we exported the source sequences to be processed in Cinnafilm’s Dark Energy software.  We used their tools to de-noise the H.264 based footage from the 5D, and add back in a level of synthetic film grain, to better match with the footage that had been shot on film.  Shane and Scott worked with our texture artist Monte Contractor to dial in the look they wanted, especially for the darker scenes, (where you find the most noise) which was then applied in varying degrees to the rest of the movie.  Once that process was completed, the exported DPXs were just added as a new layer to the source sequence, and were ready to use in any version we needed.  The same was true for when we had a separate film density pass done near the end of our finishing process.

Once all of our preparations were done, I just selected all the correct layers for a given aspect ratio across all the sequences, and queued up all of the DPX exports for that version of the movie in Adobe Media Encoder.  Once they were going, I would select the layers for a different aspect ratio, and get that queued.  We exported DPXs directly to external USB3 drives, to save time and space, with so many versions to deliver, at 1.5TB each.  I then used those DPXs as source for Cineform compressed AVI files, created by batching Cineform’s DPX2CF.exe tool.  This allowed us to keep a compressed version of our exports on site, and gave us the ability to verify that there were no render errors, since the outgoing DPXs were the source for our compressed AVIs.  I was able to do all of the final exports in a single weekend to three 3TB external drives, and schedule the read-back conversions to happen when other drives were being written to by Media Encoder, since the individual USB3 drives were technically the bottleneck.  Once Laser Pacific had all of our master DPXs, they created the DCP, did the HDCam-SR laybacks, and printed it out to film.

Twixtor is an amazing tool, but it is not perfect.  Whenever there is too much motion in a scene for the software to accurately track what is going on, it gives some strange results.  Since Act of Valor is a fairly dynamic movie, to put it lightly, we had our share of issues.  When issues with Twixtor were encountered, we had a fairly methodical, but time consuming solution.  We would bring three versions of the shot into the Premiere sequence: the original 30p version, which would drop frames to play at 24p, another copy reinterpreted at 24p, (so it was longer, but every frame was accessible) and the Twixtor interpolated render.  We would start by examining the Twixtor version, and replacing the bad frames with ones from the original file.  If there was enough action to make Twixtor totally useless, then that chaos was usually enough to hide the dropped frames anyway.  On shots that were smoother, where that would be too noticeable, we would sometimes have to use the reinterpreted version, and slow the shot down for a couple frames here and there to get through the rough spots.  This obviously took awhile, and I got really good at moving single frame clips around on the Premiere timeline.  Smooth playback with lots of layers and micro-cuts was critical for this type of work, so the fact that CS5 played back everything in real-time was critical, especially for rapidly comparing different variations.

For complex visual effects shots, our artists had to do this cleanup process on nearly every 5D source layer, before they could even start doing any regular VFX work.  Brett Novak did a lot of the early experimentation with compositing 30p Canon files with 24p film footage.  Andrew Furlong spent the better part of two years working on shots to keep pace with Scott’s editorial process.  Besides the 30p conversion issue, the visual effects were processed in a fairly traditional way, and exported to Cineform when complete.

We also dealt with a number of different types of rolling shutter artifacts.  My favorite ones were the muzzle flashes at night.  Because of the way CMOS sensors record light data sequential from top to bottom, a bright flash results in horizontal lines of varying brightness in the recorded image.  There is a lot of gunfire in Act of Valor, and most of it is in the dark, which exaggerated the issue.  This is one case where shooting 30p actually worked in our favor.  Since we had extra frames available, we would just reinterpret the footage, and manually cut out the bad frames we wanted to remove, and enough other ones to get it down to 24p.  If this didn’t result in smooth playback, someone would have to manually stitch together the good parts of consecutive frames, where once again, it was helpful to have 30 fps to start with.

We also had over a hundred subtitles, due to the Spanish and Russian dialog of the bad guys. (Some might argue that we should have subtitled the SEAL’s jargon)  These weren’t necessarily complicated, but had to be inserted into the workflow AFTER the color and grain passes, and the reframing for different aspect ratios.

We didn’t have much in the way of motion graphics, but treating them the same as visual effects was probably a mistake.  They were giving us issues right to the end, primarily because of all the revisions that graphics usually get, and because the original templates weren’t designed with our final 2K online process in mind.  Most of them were rendered as overlay layers with alpha channels, and composited over the backgrounds directly in the Premiere Pro online sequence, which gave us more flexibility to adjust them than rendering the backgrounds into the shots would have.

It was important to fix all of the visual effect, frame rate conversion artifacts, and rolling shutter issues before we further processed the image with color correction and our texture pass.  Once those modifications were rendered into the footage, pulling up source frames to cover up problems would no longer match.  Simple dust busting fixes on the other hand could easily be done on the corrected image.  So that was one of the last steps I did, pulling up individual DPX files in Photoshop for repair with the Clone-Stamp tool.

If your output deliverable is 1080p, there is very little practical benefit to monitoring 4K in post, although being able to shoot and process at 4K gives you more options, and possibly better quality.  The increase in quality from processing only takes effect if you are zooming in on the image, or doing VFX.  In almost every other case, it makes no difference, so there seems to be no advantage to processing the full 4K instead of a half resolution decode.  Back when all of this was being “burned-in” to the file during a transcode prior to editing, this approach would have been throwing away all of the flexibility offered by shooting 4K.  Now that editing systems can decode 4K RAW on the fly (especially at 1/2Res), you still maintain that flexibility of having the 4K data available if needed.

So what does all of this mean from a practical perspective?  If you are delivering 1080p for the foreseeable future, but want to take advantage of the new 4K shooting options, here are some ideas.  If you are able to record to 4K RAW, shoot with a wider lens than necessary, framing for a QuadHD center-cut if you can.  Import your footage and edit in a 1080p timeline, but do not scale to frame size, instead scale to exactly 50%.  If possible, set your decode resolution to 1/2.  Take advantage of any extra image outside the 1080p frame to tweak the exact framing of your image.  One issue with this workflow, at least in the current implementation of Premiere, is that setting your 4K footage to 1/2 resolution decode also effects any native 1080p footage, significantly lowering quality.

The same principles hold true, if not even more so, when shooting over 4K with the new Red cameras.  There is zero image quality increase of oversampling beyond twice your target resolution, so shoot with the intent of center cutting to QuadHD, by framing much wider.  The only exception would be shots that you want the option of being able to zoom into in post, but that is really the same approach, from a certain perspective. (Treat your target zoom level as the QuadHD framing point, if possible.

Now this only applies to shooting 4K RAW, but that includes most 4K cameras that are currently, and will likely include the new Canon C500 and the Sony NEX-FS700, once true RAW recording options are made available.  I predict that most deliverables will remain 1080p or 2K for many years to come, even though many 4K acquisition tools are hitting the market, so this approach will probably remain relevant for quite a while.

HP announced the long-awaited next generation to their line of professional workstations last month, and those systems are showing up in a variety of booths, now that they have started shipping.  With Intel’s new 8-Core Xeons, combined with Hyperthreading, you can process 32 threads on a total of 16 cores on the highest end Z820 systems.  There are four memory channels per socket, for a total of eight, meaning we will return to systems having 16/32/64GB of RAM instead of 12/24/48GBs.  The included support for PCIe 3.0, USB3, and SATA/600 will be very helpful, and free up slots for other uses.  The big thing missing, that I was expecting to see, is the Thunderbolt interface, which is absent.  The number of new products being released that utilize that technology has me convinced that it must be coming to PC soon, but I haven’t heard much about it.

I am looking forward to the next generation of NVidia Quadro cards to go with these workstations, but they are probably a ways off, since the GeForce 680s are barely available.  Support for the new GPU-Direct technology is mentioned in a few booths, but no one really seems to understand what it is going to do for the end user.

Red is showing a prototype of their 4K laser projector.  While impressive on paper, at 120FPS 4K on a 15′ screen for $10K, the version they are showing was only 2K, and didn’t really stand out or differentiate itself from others.  The final 4K version could be one of the most affordable full resolution 4K viewing solutions available, but it’s not there yet.  Full resolution monitoring is currently the missing link in the 4K workflow, and I anticipate there will be a variety of more affordable options available by next year.

Canon announced two new 4K cameras, the C500 which I referenced above, follows the recently released C300, and the new EOS 1D-C, which has a more traditional HDSLR form factor.  The 1D-C records 4K internally to 8bit Motion-JPEG files, while the C500 requires an external recorder, like the Ki Pro Quad, to capture the full 4K image from the sensor, but this approach allows access to the uncompressed RAW 4K data.  Both cameras have a variety of new HD 1080p recording options, that should improve the quality of the final picture, including options for image sensor windowing or scaling, and frame rates up to 60p.

Blackmagic Design’s big new announcement is that they have developed their own digital cinema camera.  It has a 2.5K sensor, and records to CinemaDNG, ProRes, or DNxHD to removable SSDs.  It kind of looks like a toy, but it packs a lot of existing functionality into a small box, including an Ultrascope display for live monitoring, and similar recording options to their Hyperdeck Studio.

Adobe announced CS6 last week, so most vendors are demonstrating CS6 support for their products.  The inclusion of Speedgrade CS6 adds a new option to the end of their workflow pipeline, and the new Prelude CS6 metadata application adds a new option for the acquisition stage.  I am looking forward to the improved 3^rd party integration and performance through the new Adobe Mercury Transit architecture, which will do away with conflicting sequence edit modes, and improve access the Mercury Playback Engine functionality with third party hardware.

That is all I have had time for so far, but more news coming soon, as well as some deeper analysis of how these new workflow options are going to fit together.

As scenes were finished, the post team would get lists of the Canon source files, and start queuing up the necessary Twixtor conversions in After Effects CS4.  Travis Schoen and Lance Holte helped me setup hundreds of these AE renders, and run them every night, on each workstation once our operators had gone home for the evening.  These conversions were rendered to 24p Cineform HD .AVI files.  We were originally required to convert entire source files in order for the online process to linkup to the new files automatically.  But once a file was converted, it would automatically work in any revised edit we got from the Avid.

When CS5 was released, it handled native 5D .MOV files much better, and more accurately, than going through Quicktime, the way CS4 did.  This resulted in files being decoded into the “correct” color space, which looked different than it had in CS4.  We had already “burned in” the old CS4 color space into all of the Twixtor conversions that we had made up to that point, so we decided to systematically replace all of those encodes when we had time.  The good new was that CS5 and the new version of Twixtor processed twice as fast.  The bad news was that there was no Cineform export plug-in for After Effects CS5 at that point, so all of the conversion compositions had to be imported into Adobe Media Encoder in order to render them to that codec properly.

The edit was reasonably “locked” at this point, so we developed a faster and more efficient way to process the clips that needed to be Twixtored.  Linking directly to the 30p Canon .MOVs in Premiere Pro gave us reasonable playback, now that we had CS5 with the CUDA accelerated Mercury engine.  Once we had a segment exactly the way we wanted it, we would copy the clips on the timeline, and paste them into After Effects.  This would create a comp with all of the clips we needed, and only the segments we were using, which was much faster to render, and saved space as well.  The disadvantage was it flattened that part of the edit into a single file with no handles.  This would have been a major problem at the earlier stages of the prrocess, but by the time we were doing the CS4 to CS5 color space replacement, the individual scenes were pretty solid.

At that point it was decided that we wanted to rescan the segments of film for the online at 2K.  We were able to use Avid’s film toolset to determine what to scan, but the resulting DPXs had to be cut into the online by hand, because there was no way to automate that in Premiere.  Miles Michaud spent a couple of long days manually building those sequences in Premiere, and that is one workflow issue that even since then, I still have not been able to develop an efficient solution to.

We would then scale the Canon footage to 2K to match the film scans.  Only having a 40Mb compressed HD image to start with meant that we had to be very careful how we processed it.  After doing many tests, we settled on using Cinnafilm’s Dark Energy to do the upres-scaling, as the directors liked the resulting look the best.  Travis Schoen managed to find a way to script this aspect of the process, since every time we made a change we had to rescale the modified clips.  This would eventually lead to us using that application (Cinnafilm Dark Energy) to de-noise and re-grain the non-film sections of the movie as well.

This process resulted in a full pass of the entire movie in 2K DPX files.  The work was all done with a software 235 preview matte, but we processed the entire 16×9 image for future flexibility.  We delivered the whole movie to our colorist as 2048×1152 DPX files.

On the boat, which was a $20 million yacht, I setup it the dining room, which was a circular room right at the front, with a 270 degree view of the Keys, and models sitting in the windows on the front deck.  And I had an array of laptops and drives stacked on the table, trying to get every card backed up to two locations before formatting it.  We would be recording constantly for 30-45 minutes at a time when the SEALs would run their operations, so that was a lot to keep up with.  Shane Hurlbut and his Elite Team operated in all sorts of challenging environments, and their methods evolved as they got more experience working with the 5D.  They have a pretty finely tuned system going now, but we were still in the developmental stages of DSLR filmmaking back then, so there were major changes in how we operated between every shoot, as we learned from our experiences.

That first shoot was also back when we were under the illusion that we were going to keep everything perfectly organized in real-time.  I was listening to a wireless feed from our sound recorder Gene Martin, who was on the other end of the boat.  While they were slating each shot, he would call out the name of the sound file, and I would write all of that down on my log sheet in Excel.  Then when I got the Canon files, I would add the appropriate file name (MVI_0392.MOV) for every camera that was shooting that take, based on the slate info and the onboard audio call outs.  I would then assign a unique identifier to each file based on an 8 character convention that we came up with.  The end result was a log that had the take info, sound file name, and all .MOV names.  This log could be converted into a .BAT file for automatically renaming all of the MOV files with the unique identifiers.  This is all while I was backing up more data from five cameras across two systems.

Between then and our next major shoot, we sorted through all of our data, and determined which steps were unnecessary, or could be deferred.  This is good because there is no way that method would have held up through our next shoot, with two weeks in the swamps of Mississippi.  By this point we had at least ten cameras, but we had 24 cards, at 16GB each.  I pretty much kept busy backing up and logging the footage in an RV a few hundred yards away and no longer tried to track the shooting process in real time.  We got better at labeling the cards since there were more steps between them leaving the camera and getting to me.  This was the first time I got to use my ExpressCard based card reader, which dumped the data to my laptop at over 50MB/s.  So now the USB2.0 drives were the chokepoint in the flow of data, which continued to be the case until USB3.0 drives became available a year later.

Our next shoot was on an island in the Pacific, mostly at night.  I worked out of one of the huts on the set, borrowing power from the lighting department when I could.  The challenge here was turn-around time.  Shooting in the dark, they wanted to review their footage as quickly as possible to make sure the 5D was picking up enough image detail.  This was the first time I brought the HP Dreamcolor display onto the set, and it was commandeered by the camera department, for real time monitoring, before the end of the shoot.  Shane has since become a big proponent of the Dreamcolor displays, as his eye into what the camera is seeing.

The last major shoot was two weeks out in the desert.  By then I had assembled what became known as the monitor cart, which was a Dreamcolor mounted to a hand truck, with a UPS battery at the bottom, both for stability and an hour of unplugged use between charges.  I had another Dreamcolor, connected to a dual processor editing station in my trailer.  I still used notebooks for the initial data dump from the cards, because the ExpressCard reader had no desktop equivalent.  By that point we had almost twenty cameras, and numerous cards floating around, making things all the harder to organize.  But with a reliable power source, I could use faster 3.5inch external drives, and a gigabit network to copy the files around.  And with a full editing workstation and Dreamcolor attached, the directors would come in during breaks and at night to watch my sequences of selects.  That was the closest we came to “editing onset.”

Throughout all of these shoots, I was constantly shipping drives of media back to Lance Holte at our office in LA.  He would ingest it all, finish any logging, and make sure it was securely backed up.  Siobhan would then import the footage into Avid as DNxHD files, and start syncing the different angles to the external audio, so that Scott would have media to work with by the time he returned from the shoot.

Mouse McCoy had long before envisioned someday being able to shoot professional video on a camera with a DSLR form factor.  The Nikon D90 was released in April ’08 and we had purchased one immediately to try it out.  We liked the form factor, and 24p was great, even if it was limited to 720p.  But high motion shots clearly didn’t hold up at all, with the level of compression it used to record to Motion JPEG.  Since everything we shoot is action-packed high-adrenaline footage, this limitation was a deal-breaker.  The D90 was clearly not the fulfillment of McCoy’s vision, but the technology was getting closer.

The Canon 5D MkII was released in Sept ’08, and it took us a little longer to look into that after our first disappointment.  Kevin Ward, who had been the DP on many of our previous projects, brought one by the office to play with one afternoon.  Clearly this had more potential, but there had been a lot of positive feedback and buzz after the initial release, so camera bodies were hard to find.  We ordered our own as soon as I found a decent deal that was actually available.

Our first major dedicated shooting test was an all night endeavor north of LA, with night-scopes and a variety of lenses and lights.  The image quality of the videos from the 5D was amazing, (once you converted it to something that could actually playback on an edit system) and it really performed well in dark scenes.  Plus it could use many existing lenses, but there were some downsides.  The 30p limitation was the biggest one by far, but there was also the lack of control over the exposure, without manual control of the shutter speed or the aperture.  By adapting it to use manual Nikon lenses, we were able to over-ride the aperture limitation, and then the shutter speed could be temporarily “locked” while aiming closer or farther from the lights, but this had to be done again for every take.

Our first major project that was shot primarily on the 5D was done with Shane Hurlbut.  The results were beautiful, but we were constantly tricking the camera into doing what we wanted.  And reinterpreting every shot to 29.97 and re-syncing the audio was extra work for post, but the native H264 files didn’t work well in any NLE software, so we were already transcoding everything anyway.  All of Bandito’s previous commercial work had been natively 24p, but for a few months, all our cameras were set to 29.97 to be compatible with the 5D footage.

Once the 1.1 firmware update was released for the 5D, that gave us manual control over the exposure.  The process on set became easier, but the post issues from shooting 30p remained.  We started looking into motion compensated frame blending software, so we could return to shooting and editing at 24p, and still use content from the 5D.  I asked around at NAB09 and found ReVisionFX and their plug-in Twixtor, which I may have used more than any other person on the planet by now.  Back in early 2009, it took an hour to convert one minute of 5D source footage to 24p, on an 8-processor Core2 based Xeon workstation.  Both the hardware and software have improved since then, so it is about 10 times faster than that now.

The other thing we had to look at was our creative editorial solution.  All of our commercials up to that point had been cut in Premiere Pro 2.0 and CS3, on Matrox AxioLE and Cineform ProspectHD based systems.  Still running 4GB RAM on XP32, those systems were maxed out on our commercials, with only 10 hours of source material.  We knew Avid was the only reliable option at the time, for a project with hundreds of hours of source footage, but nearly all of our existing tools and experience were with Premiere Pro.  It was at this point that we were introduced to Siobhan Prior, who became an integral part of the post team for the rest of the project.  Her experience using Avid on tradition feature films was the exact opposite of mine with using Premiere Pro to integrate all sorts of unusual formats into short commercials.  Our long debates over the “best” way to do things from our contrasting perspectives, eventually led us to a pretty solid workflow solution after a few months of discussion and experimentation.

I setup our first Media Composer system in the spring of ‘09, and started learning how to interface it with the Adobe Creative Suite.  The biggest issue in that regard was that we were changing the time-base of our source assets from 30p to 24p during the edit, so it was a bit tricky to figure out how to keep things straight for the online process.  Importing 30p material into a 24p Avid project causes it to drop frames as it converts to DNxHD36 MXF files.  The resulting 24p EDL from Avid doesn’t match the time-code of the original 30p Canon .MOV files, but it DOES match the frame-count of the Twixtor processed 24p conversions in Premiere.  This is because Premiere can ignore the “real” time-code of a file, and just convert the desired in-point to a frame count for the beginning of the file.  That allowed us to move sequences from the Avid to Premiere with relative ease as long as all of our source assets were pre-converted to 24p, and they all had unique filenames.

The development of our file naming convention was probably the most widely debated workflow decision, with input from the camera department, editorial, VFX, post finishing, and even our executive producers.  The end result accomplished 95% of what we set out to do, since you can’t account for every possibility from the beginning.  I limited it to 8 characters, so that it would work with standard EDLs, but we packed as much information as possible into those 8 values.  We also had Ben Gullotti develop a web database to log and track all of the footage, the design of which I based on our file naming convention.

Once we had manual exposure control, and a post solution that would allow us to intercut the footage with film, there was a major conversation that took place, where we discussed whether or not it would be feasible to shoot our upcoming feature film on the 5D.  It is funny to look back now, and realize how little experience I had, to be qualified to answer that question, but we eventually agreed that is was the best option.  Jacob Rosenberg, as the head of our post department, was confident that we would be able to overcome whatever issues would result from our decision to shoot with the 5D.  There was some concern about the lack of real time-code, and the issues of CompactFlash failures, but I was most concerned about the 30p issue, and processing times involved for 24p conversion.  I was assured that we would get a render-farm of powerful systems to do that when the time came. (In reality we ended up commandeering every edit station, every night, for six months, but it got done.)

Certain parts of the movie were still shot on film, and that footage was transferred to HD-Cam SR.  These tapes were ingested with our Mojo DX directly into the Avid as DNxHD files, to intercut with the dropped frame footage from the 5D.  The segments from the tapes were eventually recaptured with AJA Kona cards, into Premiere as Cineform444 files, as needed based on the Avid EDLs.  So we had a workflow ready to go that allowed us to edit in Avid and finish in Premiere, and could handle both Canon 5D source files and SR based film transfers.  Once things fit together successfully after our first shoot, we were pretty confident that we had it all figured out.

In the meantime, were are a number of other sources of information on the project and how we did it:

We were the cover story for POST Magazine and there was a piece on Creative Cow a while back, and there are a few others in sites that I am less familiar with.

Both HP, NVIDIA, and AJA were technology partners in the post process, and AdobeTV has a bunch of stuff about what we did, and how we did it.

This last one is not about the technical aspect of the project, but this movie review really hits the heart of what we were doing, and how it is different.