I recently upgraded my 5 year old Canon 7D and bought a 5D Mark III. I was interested in playing around with Magic Lantern’s RAW hack after seeing some examples of the improved video quality, but was a bit intimidated by stories I’d heard about it being complicated and time consuming.
I’ve read a lot of information on shooting with Magic Lantern, but I also found it hard to get detailed tips from in-the-field users, so hopefully this will shed more light on the process for you. Included at the bottom are more links that I found helpful when testing and researching Magic Lantern, including installation guides and samples.
The 5DMark3 normally records highly-compressed H264 .mov files, but Magic Lantern allows you to record the RAW 14-bit images from the sensor. These RAW files are called .MLV’s, and they contain both picture and sound.
Here are two screen grabs from my own testing of RAW vs H264.
H264 .mov (Canon Native Recording):
Magic Lantern RAW:
You can clearly see more highlight and shadow detail in the RAW image. Also, the RAW image is a bit sharper than the H264s. The RAW frame grab was made from a ProRes HQ Quicktime that I’d made via two pieces of software which I’ll talk about in a minute.
Of course, to record in RAW creates a LOT more data and requires expensive, high-speed CF Cards. I bought 5* 64GB Lexar 1066x CF Cards and one 128GB Komputerbay 1066x card for my kit. One 64GB Card holds about 12 minutes of RAW footage.
An opportunity to use the RAW feature on a real job presented itself in the form of a short film project titled ‘Sudden Reality’ directed by Matthan Harris. Matthan and I looked at the examples and thought, what do we have to lose? If it became onerous, we could switch back to shooting H264’s, but he was game to try it out and there weren’t any clients on set to embarrass ourselves in front of. We could take some risks on this project.
The first big hurdle we ran into was on-set playback. The camera can usually play back the last RAW file after recording, but if you want it in real-time, it’s in black and white and flickers terribly. The director was also acting in the short, so he’d need to see playback. We could have rented a KiPro mini or Atomos recorder to solve this by recording the HDSDI monitor (I have an on-board TV Logic monitor that converts to SDI), but we didn’t have the money for that, so we had a PA video the director’s 17” monitor with their cell phone for scenes where he needed to see playback because he was acting in the scene. It was cheesy but worked.
We had to constantly be transferring the CF cards on-set to Hard Drive because we shot a lot of footage and went through cards pretty quickly. With USB 3.0, we could transfer a 64GB card in 10 minutes, and since we backed up to two hard drives, needed roughly 20 minutes per card for data management. Also, we used ML RAW Viewer (a free program) to quickly check the .MLV RAW files after transfer.
One other bit of Data Management that really helped with the post-processing: I had the Data Manager format the cards on the computer when she was done transferring them to the two hard drives. She formatted them as ExFat instead of Fat32, and I enabled the Magic Lantern menu feature of allowing file sizes greater than 4GB. This kept the video clips from being broken up and DaVinci Resolve seems to really prefer this, especially when it comes to getting the picture and sound in sync. So instead of formatting the cards in the camera, we Erased them on the computer using Disk Utility and then just inserted them in the camera and pushed record.
Another thing that threw me off at first is the yellow and red and green ‘record’ icon. I thought it meant the camera was dropping frames until it turned green, but I later found out it indicates the status of the buffer. So the camera is at speed, no matter the color of the indicator, even though it was taking about 20-30 seconds for the icon to go from yellow to green. In the Magic Lantern menu, I made it so the camera would stop if it dropped frames so we would know about it.
One other ‘gotcha’ to watch out for is the 5D’s ‘Info’ button. If you switch the LCD view to Full Screen instead of the view that has all the Magic Lantern info (audio levels, frame rate, RAW, etc.), then the camera will record H264’s and not RAW. You may not realize this until later, so watch out for this. I found it after a few days shooting tests.
Also make sure to keep checking your Recording settings in the Magic Lantern menu. The Sound recording has to be enabled separately from the RAW\.MLV settings.
I found that if I ‘rolled out’ — the card filled up while recording — I didn’t lose the last take but I had to go back and put them through the post process manually (see below for my post processing). RAW Magic created the DNG’s, but DaVinci needed to add those clips separately because they were in folders. They didn’t make it through the post chain automatically, so this is something you want to be careful about. They’ll show up as Folders in DaVinci instead of Clips, and you need to ‘right click’ them and ‘Add Folder into Media Pool’. There may be some corrupted frames, and you may lose a large part of the last take, so be very cautious when you roll out. You might want to re-shoot the last take if that happens on set.
I was able to make the slow motion feature work and shoot 48 fps at 1600×560 resolution and then stretch it out in Resolve (1.61x Height adjustment). This is still sharper than the camera’s 1280×720 60 fps in H264, and it’s RAW. You have to change several settings to enable it, starting with the Camera’s H264 setting (change to 720×1280 60p), then make changes to the frame rate and resolution in the Magic Lantern RAW menu. Don’t forget to change everything back when you’re done, and re-check your RAW Sound Recording! Also, I suggest using separate CF Cards and create separate folders for the slow motion footage so you can deal with it separately in DaVinci Resolve and not mix it in with your real-time footage. You have to change its frame rate as well in Resolve.
Post Processing the camera media is a two-step process. First I used a program called RAW Magic ($30 in the App Store) that converts the camera’s .MLV files into stacks of .DNG files. The .DNG’s are individual frames. Raw Magic is a simple program that will automate the conversion and takes about 4 times real-time on my 15” USB 3.0 MacBook Pro, so 1 hour of footage takes about 4 hours to convert. The .DNG’s take up a little less disc space than the .MLV’s (maybe 80%?).
Here are my RAW Magic settings that we used:
The .DNG’s created by RAW Magic are then imported into DaVinci Resolve, and I tweak a few settings (using BlackMagic curves and highlight recovery) to get a very flat, high-dynamic-range look. Resolve then outputs ProRes HQ Quicktimes that can be sent to the editor. My laptop converts these files at a little faster than double real-time (1 hour of footage takes about 2 hours to process).
The ProRes files are able to retain the sound from the camera so syncing sound with the Recordist’s files is easier in post (via Pluralize or Premier).
Here’s a technical walk-through of Resolve with Magic Lantern:
In DaVinci Resolve, I added all the .DNG clips created by RAW Magic to the Media Pool, then selected all clips in the Pool and Created a New Timeline based on the clips.
I checked the timeline in the Edit window, then went to the Color Window. I selected the first clip, and changed the RAW settings to Full Resolution, Use Camera Settings for White Balance and ISO, and for Color Space and Gamma I selected ‘BMD Film’. I also raised the Saturation from 50 to 83 in the Color Wheel window. I saved this Grade in Memory bank A, then selected All the Clips in the sequence, and Loaded the Grade to all them. It changed all their settings in one go.
Then I went to the Deliver Window, selected the Advanced Option, and selected ProRes 422 HQ (could have also gone 444 if we really wanted), Render as Individual Source Clips, 1920×1080, 23.976fps, Enabled the Audio as Linear PCM (for camera audio reference). Then under File I selected Use Source Filename, Render as Unique Filenames (suffix), Force Sizing to highest Quality, Force Debayer Resolution to Highest Quality. Selected a Render To directory, Added the Job to Render Queue, and hit Start Render.
Ultimately the editor will work with these ProRes files and we can color them directly instead of going back to the .DNG files. I also want to try out Neat Video for noise reduction since I think the RAW files are pretty noisy, especially at 400 iso.
Overall I liked working with the RAW files and found them much less daunting after shooting a project with them. It’s a relatively cheap way to get much better images out of a Canon 5D Mark 3. I still think the Canon is a bit noisy and isn’t ideal for a lot of productions. Focus- pulling is tricky and the HDMI output sucks (but I’ve found a solution that helps from Small HD — an HDMI port port protector). Lack of decent playback is a deal-breaker for some clients unless you have a cheap outboard recorder and someone to operate it.
Here are a few more links I found useful when learning this process and prepping the project, as well as samples of H264 vs RAW comparisons.
https://www.youtube.com/watch?v=B5P13iyAJvg#t=10 (This one really sold me on it)
Graham Futerfas is a Los Angeles based Cinematographer. His work can be seen at http://www.GFuterfas.com