magic lantern

i’ve been a fan of shooting on canon cameras since i was just starting out in the film program at byu. digital slr (or dslr) cameras offered the option to shoot on aps-c sized sensors (or larger) with the ability to utilize quality lenses at a really affordable price. however, dslr video image quality stagnated for years with no significant improvements being offered on new cameras, and even when the 5d mark iii was released it offered only incremental gains, and yet carried a hefty price-tag ($3500 when the mark ii could be found for around $2000). when rumors surfaced that the camera had been hacked by the folks at magic lantern, i had to give it a shot.

the install took a couple of hours. there were too many tutorials that were just slightly different that it wasn’t as straight forward as it could have been. with raw being in “alpha” it’s not officially supported and therefore the raw module is not in the official magic lantern build. however, one of the best things about the mark iii hack is that you can keep magic lantern on an SD card in the camera, and then record to a 1000x cf card (previous cameras required the magic lantern firmware to be present on the cf card you were shooting on).

when it came time to shoot, following the tutorial helped us get up and ready to shoot in no time. we were limited to an 800x card which meant we could only shoot 1920×818, but that is still a pleasant 2.35:1 ratio, and that was a compromise the director was willing to make.
we shot for over 12 hours on the camera, with only one card, and had no magic kantern problems the entire day. the only issue we found was that whenever we ejected the cf card to dump, the camera switched automatically to the sd card for recording, even after we re-inserted the cf card. so it quickly became a habit of checking that setting after each mag change. It also became apparent that because we were formatting the cards in camera (as fat-32) we were limited on file sizes. the camera can continuously record 1920×818 on our speed of card, but it would break up longer takes into two separate files. those files were not easily workable in post, and it turns out you need to merge them with additional software (or with Windows command line prowess) for them to be usable.
the shoot consisted of 2 locations: one was a dark club scene, and the other was outside on a sunny day, which allowed a great variety of shooting circumstances to test the sensor. the raw capabilities of this camera are amazing. while the image that is seen is fairly crushed, when pulled into post (adobe camera raw) the latitude might well match the alexa’s amazing 14 stops.

the post work-flow was a little cumbersome as there were some additional steps needed at the time we used the hack. We first had to convert the recorded file into a series of dng images (with an included magic lantern executable file) and then color the first image of each sequence in adobe camera raw, synchronize the remaining images, and export a tiff sequence to be pulled into adobe premiere (though many people are using after effects).
at the time of our use, camera picture profiles and color temperature settings were disregarded by the camera, and the image comes out looking like it’s in the “standard” picture profile, and with some arbitrary color temperature applied to the metadata of the raw image. however because it was raw, all those settings were easily adjusted to be more desirable.
i noticed the image was quite a bit noisier than the h264 equivalent, but adobes built-in denoiser quickly got rid of undesirable noise, and when applied aggressively actually produced a very pleasant skin-smoothing effect.

i don’t think i’d ever want to shoot h264 on a canon camera again, even with the much better all-i codec now available on the mark 3. seeing the wonderful image the raw hack produces, the extra effort and expense for 1000x cf cards makes it well worth it.
personally, pixel-to-pixel i like the image far more than what the Epic produces, and maybe even more than the alexa.
The hacked 5d mark iii is perhaps best compared to the 2.5K black magic cinema camera (bmcc) in both price and espoused image quality. When used in crop mode (sampling a 1:1 area on the sensor), the canon should be able to match the resolution of the bmcc, with a similar crop to the sensor.
having used both I’d say the bmcc produces a more film-like image right out of the box, and they both achieve a similar amount of latitude when working in post. a colleague of mine would say the canon image is less sharp, but i’m not entirely sure that’s accurate.
soon we will be conducting a test comparing the bmcc 2.5k, 5d mark iii, and r3ds mx sensor, the results of which i’ll post.

i’ve included a link to a google drive folder that contains images from the shoot we did. the settings i ended up using for the club scene were:
2800K color temp, 0 tint, +.5 exposure, +10 contrast, -50 highlights, +25 shadows, 0 whites, -10 blacks, +15 clarity, -5 vibrance, -5 saturation. then switch over to the detail tab and have 50 value for luminance under noise reduction (this is beyond what is needed to get rid of the noise, but I loved the effect it had on people’s skin).

sample dng frames

i’d love to hear from others what settings they like, quality issues they see, or anything regarding the ml hack.

update: i’ve written an article about a shoot i did in hawaii with the camera and shooting at full 1920×1080 with magic lantern, you can read it here.