hyalinejim

23 minutes ago, Andrew Reid said:

That frame I could have sworn was from a Hollywood movie but it's actually just somewhere in England isn't it

Well, as a Cork man living in Dublin I'd say that pretty much sums it up

24 minutes ago, Andrew Reid said:

Is there any reason why you're shooting in 2x crop 3K and not the 1.5x crop UHD?

The high resolution crop modes in the crop_rec module of the "4K raw video recording" build have zero preview at the moment - just a bunch of garbled pixels.

But the regular 5x crop mode combined with the 14 bit lossless option now available in the mlv_lite module gives the slowed down grayscale preview, so you at least have some idea of what you're doing.

At the moment, these super-resolution modes are not usable professionally. But, if the LCD preview gets fixed so that it's in real time then absolutely yes.

@fuzzynormal log transcodes are the way to go. Ballpark WB on ingest, and do your grading later. Compare my log example vs original DNGs above. 

OK, I'm back at home. Haven't had a chance to have a look at these yet. These are the first DNG in each of the sequences I shot today. There may be motion blur in some, as it was handheld, and some were through a car window.

https://www.dropbox.com/sh/oqdeuipslwnvipq/AADPkIFTb-HCVPmdsZKz6AeEa?dl=0

All are 3168 x 1320. I used 5x crop with 14bit lossless. I see A1ex has posted a new build with all the new crop modes included. Haven't had a chance to play with those just yet. But these DNGs are continuous at 24p on a Komputerbay 256GB 1066x.

They're not particularly pretty. But if you don't have a cam and wanna have a play - jump right in!

EDIT: And if you're interested in checking out how Cinelog-C can contribute to a ML workflow, I've exported a 444 Cineform transcode of those very same DNGs in one file here:

https://www.dropbox.com/s/0o8ugt0s9yrxk45/ML%203K.avi?dl=0

PS: Lens was a Tamron 24-70 @ 4 to 5.6 mostly.

Extract files with new version of mlv_dump on the experimental builds page. Use this for batch processing:

http://www.magiclantern.fm/forum/index.php?topic=10526.0

Then work away as normal with the DNGs (Resolve / ACR / other software)

1 hour ago, jpfilmz said:

I'm in 1.13 and getting pink colored fames using MIRaw Viewer....but it's only with the 14bit lossless setting enabled.   

Try the mlv_dump from here:

https://builds.magiclantern.fm/experiments.html

Just on the way back from a Sunday afternoon walk. Brought the 5D. Will try to post some DNGs later.

36 minutes ago, Andrew Reid said:

What about the HDMI output? Does that offer a workable preview while the camera records 4K?

AFAIK it doesn't, but I don't have an external HDMI to check.

Paul Leeming posted this example of extreme VLog ghosting at high ISO on the GH5 facebook page:

Another user mentioned that Panasonic has said that if there is a demand to turn off NR completely, they will do it. Perhaps it's time to voice your concerns to Panasonic if ghosting like this is a problem for you.

Personally, I would prefer to control the NR myself, and I'm sure many others would too.

29 minutes ago, Chrad said:

While this is amazing, will no doubt be put to good use, and any democritization of prohibitively expensive technology should be welcomed with open arms...

 

...the elephant in the room is that if you can afford to do serious work with 4K raw, you can afford a more suitable camera that requires less hoop jumping than the 5Diii. The storage cost is astronomical. Even independent features with A-list talent choose to record in less space intensive formats for economic reasons.

Well, I've just finished a 17 minute documentary short. I shot 5 hours of ML 1080p for this. I had to buy an external hard drive to stick the original MLVs on as backup - around €80. And the log 4444 transcodes that I edited and graded from sit on my internal hard drive, taking up 400GB. These costs are not high. Obviously the costs will be higher for a longer project, but quite manageable. And the 3K sample I posted earlier is a roughly similar data rate, due to the 60% compression of the new lossless 14bit.

However, the "elephant in the room" for me is this:

Quote

* Canon preview is broken in most modes; you need to use the grayscale preview in the raw recording module.

ML users are familiar with this, but newcomers to ML may not realise that these advancements in resolution come at a cost: while you're shooting the LCD runs at about one frame every 2 seconds. This means that it's almost pointless to try to move the camera or to change focus while recording. The feedback is so slow that you're pretty much guessing and it's impossible to be precise. Will this ever change? Who knows?

Everyone should read a1ex's updated first page post:

https://www.magiclantern.fm/forum/index.php?topic=19300.0

The new modes are ideal if you're making work in the style of James Benning, though! (fantastic antidote to the hectic pace of contemporary media)

3K Cinemascope downsized to 2K DCP.

I missed the focus a little, but the colours are awesome. This is Cinelog-C to Impulz Tetrachrome Arri lut.

Still waiting on my pre-order

13 minutes ago, zmarty said:

I think there are motion problems, I don't know why.

Shutter speed is too fast, IBIS is not switched on, and camera movments are too fast.

How are you guys dealing with grayscale preview in high res crop modes? I regularly use the 3x full colour 1920 x 1080 crop_rec with a Tamron 24-70, which pretty much lives on my 5D, to get me that 70-200 reach without changing lenses. But I haven't found a use for the 5x zoom higher res modes too often, other than locked off landscape shots. I mean, you can't do reframing or focus pulls in grayscale.

Look, this development is potentially huge. It's generally agreed that 5D3 MLRAW gives a wonderful image, especially in terms of colour, and comparisons with the Alexa are often made. While it may not have incredible dynamic range, it's still very respectable at around 11.5 stops. To date, 5D3 ML footage has fallen down in terms of its resolution. Because of pixel binning, the 1080p image suffers from a general softness and slight but noticeable aliasing. Its deficiencies are clear when compared with the downsampled-from-4K 1080 of the C100, for example, or even the in-camera 1080 of the GH5.

Resolution for the sake of resolution is a fool's game. Recently, I was at a screening of the art collective AES+F. These guys' work is based on a slick aesthetic, in part commenting on the world of commercial advertising. The screen was huge. I was sitting in the middle of the theatre, and I was wearing glasses. There was a question and answer session at the end and somebody asked about 4K. AES+F replied that the work was in 4K, but had been presented in 2K because the projector was 2K, but it made zero difference to the appreciation of the piece. If the final image has enough resolution, it's incredibly difficult to separate 2K from 4K. The difference is there. But it doesn't matter.

Back to Magic Lantern and the Alexa. It has been noted that the Alexa's 2.8k is ideal for a 2K or 1080p delivery as the downsampling counteracts the softness that is inherent to the debayering process. And this is exactly what we're seeing today with Magic Lantern for the 5D3. As I type, my camera is recording a 3072 x 1286 losslessly compressed 14bit RAW file to the card. It's a 1.5x downsize from this resolution to the DCP spec of 2048 x 858, eliminating softness and giving a sharp, clear image. And the camera has been running for 30 minutes continuously.

Admittedly, I'm getting a greyscale preview with a low refresh rate, no sound, and a 1.8x crop of my full frame sensor. But if what A1ex is suggesting in his "April Fools" actually comes to fruition, these issues may no longer be issues. It's an exciting time to be an owner of a 5D3.

Respect to Magic Lantern

30 minutes ago, Stab said:

Does anyone know the native ISO of the GH5? I've seen different websites stating different results. I think it is ISO 400, not 200. Correct?

This should be easy enough to figure out. Stick your camera on a tripod in front of your monitor. Create a step wedge chart in photoshop or google one. Expose the camera at ISO400 so that some of the wedges are clipping. Drop ISO down to 200 and check results on a waveform. If 200 is a digital pull, then the clipping point will be the same as 400, just pulled down few IRE.

7 minutes ago, mercer said:

This is pretty amazing, am looking forward to future developments. After doing my research and watching a ton of videos... I really don't know why I would ever want to shoot on something else. The image is too undeniable. With Hunter's Alexa LUT, there is also some beautiful highlight and shadow rolloff. And with the different LOG conversions... you use CineLOG, correct? Ther are so many options for post workflows.

Do you know the max continuous resolution and frame rate when shooting in 2:35?

I'm getting continuous 24p at 3168 x 1320, which is 2.4:1. At the moment, vertical resolution doesn't go any higher than 1320. It's far from being usable at the moment, so hold your horses.

Agreed that the image continues to amaze and yes, I'm still doing Cinelog. What's the deal with Hunter's lut for Alexa - any links, particularly on how to develop the DNGs pre-lut?

@Filipe Samora I thought your name might be an anagram:

APRIL FO... (er)...A..I...M..S.... so not quite an April Fool's joke

I stopped recording after 11 minutes. It's safe to say it's continuous. So yes, there is a Santa Claus if Santa Claus = continuous 24p compressed RAW at 3168 x 1320 (max vertical resolution without going wider than 2.4:1). Here's a DNG:

https://www.dropbox.com/s/8trg0exly76nx9a/m01-1531_000000.dng?dl=0

Santa Claus is yet to bring correct preview framing while recording or sound. Let's hope it doesn't take until April 1st 2018.

4 hours ago, fuzzynormal said:

1st rule of fight club...don't talk about fight club. 

Perhaps! OR perhaps it's an ultra-sophisticated April Fools where the people calling April Fools are being fooled. As in, "Hey dude, I heard you like April Fools, so I put an April Fools in your April Fools so you can be fooled while you're not being fooled"... an April Fools Inception, if you will pardon the mixing of pop culture references.

Alex did post this in response:

Quote

Nope, I would rather say it's the whole truth as long as no one talks about continuous recording:

http://www.magiclantern.fm/forum/index.php?topic=18300.0

So perhaps there really is some consolidation of the incremental but significant developments that have been achieved over the last few months, like 10 bit, centered 5x zoom, etc, plus some new development (fixing the static preview problem in 5x zoom 10 bit, maybe?) that is genuinely exciting and going to make these higher resolutions with lower bit depths actually usable, albeit with time limit on recording. And Alex is riffing on ML's history of April Fools pranks by making a genuine announcement on April 1st.

Just some idle musings...

EDIT: In fact, this is looking quite likely. Alex appears to be in the early stages of delivering lossless DNG compression

https://www.magiclantern.fm/forum/index.php?topic=18443.msg182074#msg182074

EDIT 2: I downloaded the test build and am 5 minutes into a 3168 x 1320 24p crop mode recording at 14 bit lossless as I type. Preview framing is not correct so it's only good for locked off shots in its current incarnation. Also it's MLV lite, so I think there's no sound.

31 minutes ago, Hanriverprod said:

Yeah, at first, I was lukewarm, then hyped with some of the footage, and now it's a no go with that ghosting.

Yes, it's there. But have you ever spotted ghosting in any of the real world low light tests completed so far?

I've seen ghosting when freeze framing some footage (the low light V-Log stuff from Emmanuel Pampuri) - but it's not noticeable to me during playback from what I've seen so far. And ghosting is something that really bugs me, having wrestled with it extensively on the XC10. Even in this video, which is the classic wave-the-camera-around-to-show-ghosting, you have to kind of concentrate to spot it.

Check the date!

Yeah, I spotted that earlier today, but he left the knee on auto.

24 minutes ago, hyalinejim said:

 If I had my GH5 I'd be balls deep in the menu, massaging Like709's beautiful knee.

Maybe I spoke too soon. Paul Leeming finds no difference in DR between CineD and Like709:

Quote

OK, so I've done a test of Cine-D, Like709 with knee adjusted to give maximum dynamic range, and V-Log. All done ETTR to normalise the highlights and let the shadows fall where they may. Thanks for the tip Mike

Dynamic range results, best to worst:

1. V-Log by about two stops.
2. Cine-D (just a bit better than Like709, but not by much)
3. Like709

Basically, with the knee control set to Point: 80 and Slope: 99 (showed maximal gains), it looks very similar in curve to Cine-D, only with a touch less dynamic range in the shadows. I mean, a really little bit of difference. I even wonder if Cine-D is just Like709 with those knee settings built in, but it seems a touch more refined. The whole Cine-D curve is slightly lifted compared to Like709, which is where I think the slight gain is coming from. I think in the real world you won't see much difference between the two when set up for maximum dynamic range.

V-Log is really in another class though, showing more highlight retention and even cleaner shadows!!

Based on the above, I will stick with Cine-D as the linear profile going forward.

https://www.facebook.com/groups/pansonicgh5/permalink/1688892324743802/?comment_id=1689255548040813&notif_t=group_comment_follow&notif_id=1490971906901837

How is the ghosting in Like709? This could be a great alternative to V-Log at high ISO. See page 175 of the manual to dial down the contrast of this profile using the knee settings.

I still can't believe that people aren't all over this profile. If I had my GH5 I'd be balls deep in the menu, massaging Like709's beautiful knee.

OK, but at what ISO? I would expect to see very little to no ghosting at base ISO, and for it to increase as ISO increases.

If it's any consolation, having been through this issue with the XC10, ghosting is a hell of a lot more noticeable in a still image than a moving image. But once you see it, it's hard to unsee it. I don't have a GH5 yet so I can't corroborate for you. But if you'd like others to test if their units are affected (and my guess is that it's universal) then you need to tell us what ISO you're seeing ghosting at.

To minimise it:

1. Keep ISO low where possible
2. Turn NR to minimum (theoretically this should decrease it)
3. Find the profile that has the least ghosting
4. Keep movement to a minimum
5. Try not to worry too much about it. 99% of the population won't be able to see it.