Magic Lantern Raw Video

[SlanderShot]

7 minutes ago, kidzrevil said:

@SlanderShot ah man Im looking for something that works with windows ?

No problem i use Windows too and MLVFS is available on W7 to W10. https://www.magiclantern.fm/forum/index.php?topic=13152.0

[kaylee]

a year ago "4k" was announced for the 5d3

whats the situation? anyone have a progress update?

is it usable? 

[mercer]

On 4/2/2018 at 7:59 AM, kaylee said:

a year ago "4k" was announced for the 5d3

whats the situation? anyone have a progress update?

is it usable? 

I’m interested to know this too. Have you found any info on the internets?

I’m also interested to find out if the crop modes or experimental builds allow for any continuous slow motion. If I could get a constant 60p... or really just 15 seconds... at close to 1080p, I’d be more than happy. But I’d love 72fps if possible.

[Deadcode]

Storm is coming... 

SD Card writing speed hack

If you are familiar with the ML possibilities, you already know the best RAW capable cameras are those with CF card. The CF Card interface is capable of 70MB/s writing speed with 5D2, 75MB/s with 7D, and 95MB/s with 5D3.

6D / 700D which has SD interface is only capable of 40MB/s, limited by the camera and not the SD card.

Now the ML team are pumping up the SD card interface writing speed. It seems like the 40MB/s can be raised up to 70MB/s or even further.

That means the 700D will be able to record in 2520x1072 continuously

I hope they can make it stable

 

[kye]

5 hours ago, Deadcode said:

Storm is coming... 

SD Card writing speed hack

If you are familiar with the ML possibilities, you already know the best RAW capable cameras are those with CF card. The CF Card interface is capable of 70MB/s writing speed with 5D2, 75MB/s with 7D, and 95MB/s with 5D3.

6D / 700D which has SD interface is only capable of 40MB/s, limited by the camera and not the SD card.

Now the ML team are pumping up the SD card interface writing speed. It seems like the 40MB/s can be raised up to 70MB/s or even further.

That means the 700D will be able to record in 2520x1072 continuously

I hope they can make it stable

 

Holy wow.. that is excellent news!  I bought my 700D for stills and to replace my Panasonic GF3 as a travel camera more years ago than I can remember...

I think I read something about RAW compression (which I interpreted as lossless compression) which was also being worked on?

[Hans Punk]

[Stathman]

You forgot something...

"Foolishly I should have quit earlier to recover the logs from the SD card, but alas that Sandisk appears to be toast, hence the lack of precise data - just my anecdotal report."

"Do not try on expensive cards until we figure out what happened!"

[Hans Punk]

24 minutes ago, Stathman said:

You forgot something...

"Foolishly I should have quit earlier to recover the logs from the SD card, but alas that Sandisk appears to be toast, hence the lack of precise data - just my anecdotal report."

"Do not try on expensive cards until we figure out what happened!"

LOL

[squig]

3 hours ago, Hans Punk said:

Congrats, you made a cyberpunk earring.

Now if you'll excuse me, I've gotta go sacrifice my first born to the film gods for sparing my 128Gb extreme pro.

[kye]

This card-reader hack is at the bleeding edge of the bleeding edge.  When something hasn't been included in the experimental build yet.....  Still, lots of people getting >40MB/s readings sure is promising.  

When it's stable I'll be using it for (1728 / 16:9 / 10bit / 48MB/s) and (1792 / 16:9 / 10bit / 51.6MB/s)... but I'd be interested to see if they manage to figure out modes above 1728 without a sensor crop.

[kye]

First publishable ML RAW test.  700D, ML RAW, Sigma 18-35.  
Not a great example of any aspect of film-making, but might still be of interest as an example of lesser film-makers with lesser equipment

One thing I found surprising was that the ISO noise from the 700D (not the quietest camera of all time, especially at 800-3200 in this video) was quite organic and natural looking.  As the noise changed a lot between shots I had a go at the technique that @kidzrevil has mastered of NR+Grain to make the video more consistent.

Enjoy

[kidzrevil]

@kye looks really good !! What kind of grain are you using ? Filmconvert or an overlay ?

My latest test of the 5d raw w/ my DIGILOG LUT & workflow. I got better with my handheld technique  im no longer losing sleep over the lack of IBIS. Magic Lantern Raw has really matured. I use the regular nightly build since it is the most stable and use my post processing techniques to get faux 4k for streaming. 

 

[kye]

11 minutes ago, kidzrevil said:

@kye looks really good !! What kind of grain are you using ? Filmconvert or an overlay ?

Thanks!  That comment means a lot coming from you

At the risk of providing too much information, the total workflow was:

• ML RAW 1728 10-bit
• MLV App (Mac) ---> Cinema DNG Lossless
• Resolve Clip Node 1
  • WB
  • Basic levels
• Resolve Timeline Node 1
  • OpenFX Colour Space Transform: Input Gamma Canon Log ---> Output Gamma Rec.709
• Resolve Timeline Node 2
  • Noise reduction (Chroma only - the Luma noise in RAW is quite pleasant)
• Resolve Timeline Node 3
  • Desaturate yellows (Hue vs Sat curve)
  • Desaturate shadows + highlights (Lum vs Sat curve)
• Resolve Timeline Node 4
  • Slightly desaturate the higher saturated areas (descending Sat vs Sat curve)
• Resolve Timeline Node 5
  • OpenFX Colour Space Transform: Input Gamma Rec.709 ---> Output Gamma Arri LogC
  • OpenFX Colour Space Transform: Input Colour Space Canon Cinema Gamut ---> Output Colour Space Rec.709
• Resolve Timeline Node 6
  • 3D LUT - Rec709 Kodak 2383 D65
• Resolve Timeline Node 7
  • Sharpen OFX plugin
• Resolve Timeline Node 8
  • Film Grain OFX plugin: Custom settings, but similar to 35mm grain with saturation at 0
• Resolve Timeline Node 9
  • Glow OFX plugin
• Resolve Fairlight settings
  • Master channel has compressor applied to even out whole mix
• Render settings:
  • 3840 x 1632
  • H.264 restricted to 40000Kb/s

I credit the overall architecture of the grade to Juan Melara - I cannot recommend his YouTube channel enough.

To those starting out, in case that looks like a stupid amount of work, it's fast as hell once you save the structure in a Powergrade.  Once I'd converted to CinemaDNGs the whole edit process only took a couple of hours including music selection.

[kidzrevil]

@kye sheesh thats a lot of transforms for 10 bit low res footage. Your end result looks great but I challenge you to get the look you want in 3 nodes ! You can do it !!! Your 10bit footage will look a lot better trust me. You should only transform your color space once in your grading chain

[kye]

54 minutes ago, kidzrevil said:

@kye sheesh thats a lot of transforms for 10 bit low res footage. Your end result looks great but I challenge you to get the look you want in 3 nodes ! You can do it !!! Your 10bit footage will look a lot better trust me. You should only transform your color space once in your grading chain

Challenge accepted!

The workflow comes from here.

I'm currently trying to get a workflow that allows camera matching between all my setups and either this setup with the Colour Space Transform plugin or potentially ACES transforms seem like the best candidates.  Unfortunately I haven't found profiles for ML RAW, iPhone, or Protune, so I'm still left to rely on my (modest) grading abilities.

In terms of colour space transforms, don't get them confused with LUTs as they have advantages, not clipping the data is one of them.  The above suggests that they are lossless transformations so don't degrade signal quality (and Resolve has a very high bit depth internally).

[Deadcode]

38 minutes ago, kye said:

I'm currently trying to get a workflow that allows camera matching between all my setups and either this setup with the Colour Space Transform plugin or potentially ACES transforms seem like the best candidates.  Unfortunately I haven't found profiles for ML RAW, iPhone, or Protune, so I'm still left to rely on my (modest) grading abilities.

Filmconvert has very nice profiles for GoPro (just make sure you download profiles for Hero 3/4/5, the color is the same, but the profiles are for different settings, for example Protune Flat, Native WB is only in the Hero5 Profile) , and you can use the 5D Mark III BMD RAW profile for your 700D

[kidzrevil]

@kye the workflow before compression may be non destructive but you are still shifting a lot of data around. These transforms exaggerate noise and other artifacts in the image. Its better to use floating point transforms than the hardcoded transformations in LUT’s for color space conversions. Davinci Resolve Color Management & ACES are floating pointing operations that are far superior to LUT’s especially when matching cameras. Filmconvert does a good job too but I’ve found variations in their profiles.

[webrunner5]

Speaking of editing, just found this. Nice.

https://***URL removed***/news/4236484813/wd-s-new-blazing-fast-gaming-ssd-is-perfect-for-4k-video-editing

[kye]

8 hours ago, kidzrevil said:

@kye the workflow before compression may be non destructive but you are still shifting a lot of data around. These transforms exaggerate noise and other artifacts in the image. Its better to use floating point transforms than the hardcoded transformations in LUT’s for color space conversions. Davinci Resolve Color Management & ACES are floating pointing operations that are far superior to LUT’s especially when matching cameras. Filmconvert does a good job too but I’ve found variations in their profiles.

I did a bit of reading before my previous reply and I didn't get a straight answer on LUTs vs floating point algorithms.  The logic on LUTs was that they have two main problems: a lack of data points, and potential issues in-between the data points.  
The thread that I read included people saying that even a LUTs with thousands of data points are still the equivalent of a curve with only ten control points on it, so in terms of matching to the foibles of a sensor put through a cameras internal profile transforms, it is potentially going to have a bit of error.  
The second issue about what happens in-between the points isn't that interpolation isn't possible (with all the variations of linear / quadratic / polynomial / etc functions available) it's that there's no consistency between standards and so although your program might do a good job of interpolation, who knows what the software that the LUT was designed on was doing or if they match.  This would become much more important the lower the number of data points in the LUT becomes.

What I took from that was that RCM and ACES are ok and LUTs are a question-mark.  In the end the proof is in the pudding (as the common saying goes) so it's just if you prefer the final result or not.  When I think about this stuff I get enthusiastic about taking some time to try and reverse-engineer what a particular LUT or transform is doing so I can learn from it, but then life happens and my energy fades..  one day I might get around to it.

I haven't forgotten my homework.

[cantsin]

The problem is that LUTs can map multiple color value to the same color values, thus degrade footage and be irreversible, while a mathematical function (applied to floating point values) is always reversible.