RED Monstro 8K VV & ARRI Alexa LF 4.5K Camera Test 2019

[Phil Holland]

There is a Compact Drive Reader that is USB3 and about $600.

You're welcome..... :001_cool:

PS - Of course, the software is Mac only. So there is that.....

That's what I was alluding to. Especially with their zip, erm... HEDT thing.

 

[Hrvoje Simic]

That's what I was alluding to. Especially with their zip, erm... HEDT thing.

Did you do comparisons of reader transfer times, RT playback and renders ?

 

[Ido Karilla]

Hi,


1. Can someone explain why faces seem longer on the Alexa?

2. It seems to me gamma is more accurate with the Alexa, textures are more prominent on it. Is there an other explanation?


Thanks.

 

[Sareesh Sudhakaran]

Great test, Phil! Thanks for putting it together.

For what it's worth, I prefer how Red handles darker skin tones straight from the camera (IPP2) and how Arri handles lighter skin tones. Of course after grading it doesn't matter, but I find it interesting. Even in Phil's test the lighter tones tend to stand out with the Arri, but the darker tones look more natural to me on the Red (e.g. the shot with the four women).

Last year we shot this experimental piece, and the way the Dragon sensor handled skin tones (IPP2, Log to LUT in Resolve 14) was just amazing:

On the other hand, last month we shot a fashion film with the Alexa LF and here's a frame grab of my DP, Sivanesan, sitting in while roughing in the light:

His skin tone from the Alexa LF just doesn't look natural (I have other samples, but I can't post it here) straight off camera (LogC to 709, Resolve 15). But it looks amazing with fairer skin (our model was Caucasian). If somebody can share their experiences here that would be great.

And if anybody's interested here's my quick review of the Alexa LF (I also explain midway why I decided to use it instead of the Monstro for this project):

Ultimately, with my limited experience with both cameras the argument for skin tones can be made both ways! However, as Phil's test shows, the Monstro is 'more camera' than any other.

On a different note, I wonder how much of a difference it would make if Red offered an oversampled 8K in S35 mode in the Monstro, similar to how Arri did with the Alexa (3.2K)?

 

[Hrvoje Simic]

For what it's worth, I prefer how Red handles darker skin tones straight from the camera (IPP2) and how Arri handles lighter skin tones. Of course after grading it doesn't matter, but I find it interesting.

His skin tone from the Alexa LF just doesn't look natural (I have other samples, but I can't post it here) straight off camera (LogC to 709, Resolve 15). But it looks amazing with fairer skin (our model was Caucasian). If somebody can share their experiences here that would be great.

That depends on quite a few factors, white balance and light being some of them. Also, the example you posted has generic LUT applied. All these tests are not tests of base camera image but tests of camera + transformation.

Red has superior separation in shadows so that helps with darker skin palette but final colour results depend on transformation from log/ wide gamut.

 

[Jacobo Martinez]

Hi,


1. Can someone explain why faces seem longer on the Alexa?

2. It seems to me gamma is more accurate with the Alexa, textures are more prominent on it. Is there an other explanation?


Thanks.

This is a characteristic that I also noticed. Why is this ? I feel Alexa face comes toward you, people look thinner.

 

[Hrvoje Simic]

This is a characteristic that I also noticed. Why is this ? I feel Alexa face comes toward you

Tonal distribution affects perception of volume.

 

[Phil Holland]

Back from NAB, onto a weird shoot until the 12th, but in town.

Have a good holiday weekend.

 

[Tim Sessler]

Thanks for this test Phil!

Whats the deal with the grid like noise pattern in the Xyla chart? I'd recently posted about that, asking about it as I had noticed the same pattern in very high ISO footage. The LF on the other hand seems to have a much "cleaner" noise pattern. Curious to hear your thoughts about that!

Thanks,
Tim

 

[Jacobo Martinez]

Thanks for this test Phil!

Whats the deal with the grid like noise pattern in the Xyla chart? I'd recently posted about that, asking about it as I had noticed the same pattern in very high ISO footage. The LF on the other hand seems to have a much "cleaner" noise pattern. Curious to hear your thoughts about that!

Thanks,
Tim

I would also like to know this .

 

[Jake Wilganowski]

I think you are seeing one of the differences in the cameras

 

[Aharon Rothschild]

Pretty standard practice here on reduser to ask for a raw frame whenever someone posts about sensor issues like noise etc. I think we can clear up a lot with a xyla frame for Monstro and LF.

 

[Jacobo Martinez]

Thanks for this test Phil!

Whats the deal with the grid like noise pattern in the Xyla chart? I'd recently posted about that, asking about it as I had noticed the same pattern in very high ISO footage. The LF on the other hand seems to have a much "cleaner" noise pattern. Curious to hear your thoughts about that!

Thanks,
Tim

Does it have to do with compression , how R3D works?

 

[Phil Holland]

Pretty standard practice here on reduser to ask for a raw frame whenever someone posts about sensor issues like noise etc. I think we can clear up a lot with a xyla frame for Monstro and LF.

Not going to happen, also there is no "single xyla frame", that is not how you use this chart. What you are seeing is a composite of patches from each camera composited together. If you expose the chart all at once without utilizing the sliding blinders the light itself contaminate the results.

I'm on shoot until the 12th, but I will post some log3G10/LogC side by sides as requested when I have a moment.

Whats the deal with the grid like noise pattern in the Xyla chart? I'd recently posted about that, asking about it as I had noticed the same pattern in very high ISO footage. The LF on the other hand seems to have a much "cleaner" noise pattern. Curious to hear your thoughts about that!

For the warehouse scene, I wouldn't exactly "trust that" as it was a early demo Ranger and moderately all bets are a bit off. But the Xyla's and ISO Color Stabilty + Texture are regular DSMC2 Monstro, my cams. Kicking myself for not using my vanilla Monstros for a take of that shot as I suspect something is funky.

For those charts you're looking at a log3G10 image pushed to ISO 3200 and matched to the curve of LogC to have a 1:1 ratio on where those patches exist in both spaces, checkout my previous Monstro DR stuff for some expanded thoughts:
http://www.reduser.net/forum/showthread.php?165811-RED-Monstro-Dynamic-Range-Examined


Something really to note regarding this test is finishing for 4K and usable ISO Range without noise reduction. Everybody will decide where they land on this, but I have now participated in 3 studio tests that more or less tell the tale on what is preferred among current cameras. Getting the shot is always the most important thing, but visual quality targets are a very real thing for certain entities and that is often overlooked.

I'll just say what I know on Monstro to get that ball rolling. Even at 5K ISO 1600-3200 is rather okay for a 4K finish in SDR, with the potential to stretch to ISO 5000-ish for the Low Light Optimized and Standard OLPF. Interestingly the native color of Monstro's sensor plays better with the Skin Tone - Highlight OLPF. General logic for me is working typically at somewhere between ISO 400-1600 and finding that sweet spot that works for what you are doing. Things are even better at higher resolutions like 6K through 8K. As for HDR, ideally you'd tend to prefer a lower ISO rating as there typically is more visible shadow detail. Much of the HDR finishes I've done in the last year shot on Monstro have been in the ISO 400-1600 range however. Which sort of mimics some recent studio test results. Also not shocking, the general usable range be default on the cameras is ISO 250-3200. Which make sense. I wouldn't mind 200 being in there for sanity's sake, but that's nitpicking.

I really would try hard to not push the LF beyond 800 if you can avoid it if you intend on finishing 4K. Maybe stretching to 1600 when needed, I would avoid ISO 3200 no matter what. It's not really the amount of noise, but rather the size of it relative to a 4K finish from max resolution which makes NR sort of funky and you will 100% be adding grain if you are doing that to make it look decent. Generally speaking and "the way" some narrative LF shoots have been going thus far are mainly ISO 400 and 800.

That all said, if you've got an image not terminating into the noise floor on the exposure, you have more flexibility. Which always gets us back to best lighting and exposing practices really. Oddly somebody asked why I didn't add fill light for the comparison and I had to explain that this was a side by side and not about doing my best to make each camera sing. That's not really what a camera test should be about. We know the cameras make pretty images, what is needed knowledge is when those images fall apart or where certain things become a concern. And everybody will have a different tolerance on a few of those topics.

 

[Michael Wahba]

Great test Phil. Thanks for sharing. I wonder how the Gemini fits in the mix and whats upcoming for that sensor's future.

 

[Hrvoje Simic]

I really would try hard to not push the LF beyond 800 if you can avoid it if you intend on finishing 4K. Maybe stretching to 1600 when needed, I would avoid ISO 3200 no matter what.

I would not go beyond 1280 on that sensor.

 

[luigivaltulini]

Not going to happen, also there is no "single xyla frame", that is not how you use this chart. What you are seeing is a composite of patches from each camera composited together. If you expose the chart all at once without utilizing the sliding blinders the light itself contaminate the results.
.

Yep .. although I noticed that with the sigma cine I had no problem with flare or anything ...
It would however be better to shoot more patches.

[/url]IMG_0945 by luigi valtulini, su Flickr[/IMG]



The problem Phil is that people want to see Raw files, so they can judge if you're saying the right thing.
You're touching a very delicate subject, Red vs Arri and many users
they do not like RED and it is difficult for them to accept what you are showing ... And even in front of the evidence in some fields they will always say that Alexa is better than in Monstro ..
I know you probably won't, you will release the usual prores ..
But maybe one day I'll post them

anyway, good work, I wait for my ALexa mini Lf and then I'll do some tests too ..
Although I always prefer Monstro as a camera in all situations.

 

[Aharon Rothschild]

From the Weapon Monstro VV operations Guide p.78

"Range is ISO 250 to 12,800. Default is ISO 800"

Phil what's the dynamic range at 800 iso only?

What's being used to debayer for Arri LF and Monstro for the side by side xyla demo?

 

[Phil Holland]

From the Weapon Monstro VV operations Guide p.78

"Range is ISO 250 to 12,800. Default is ISO 800"

Phil what's the dynamic range at 800 iso only?

What's being used to debayer for Arri LF and Monstro for the side by side xyla demo?

The Default ISO Range within the menu when you purchase the camera or update your firmware is ISO 250-3200, that is the Recommended Usable ISO Range. The cameras can go to ISO 12,800 just fine, but nobody here or in the world would recommend filming at ISO 12,800.

I'm not here to enforce anybody to shoot at a particular ISO, but I am extremely happy to help recommend best practices as what has worked for me on various productions as well as other's productions. At the moment I'm working with the Skin Tone Highlight OLPF for instance at 800 for my current shoot. My last shoot was all ISO 400 on the Standard OLPF. I select and rate, light, and expose to what I'm trying to achieve visually; both in look and in technical execution.

The Dynamic Range remains the same at all ISO Ratings. It's more of the displacement of tones. My older Dragon Xyla should help visualize this concept, note that this is prior to IPP2 and log3G10 and is REDLOGFILM:

Debayer was performed via RCX and ARC then checked against Resolve and Premiere for thorough testing. Oddly I do that about 4 times a year to ensure the SDKs are rendering things properly. There have been times in the past where you catch something fuzzy, usually the 3rd party's fault.

 

[Aharon Rothschild]

From the RED VV manual p.78
"ISO NOTE: To see the full range of ISO ratings, select ISO in the Upper Status Row, select Edit List, and select Custom. Select the camera ISO rating. The sensitivity value increments in 1/3 stops. When the ISO rating is adjusted, the camera logs the change as metadata and the monitor path reacts accordingly. Higher ISO values lead to brighter images in the monitor path, and vice versa. RED recommends setting the ISO to the default, then adjusting the aperture, lighting, and ND filters to match. The ISO can later be adjusted around one (1) stop for fine-tuning. Range is ISO 250 to 12,800. Default is ISO 800."

If the dynamic range is the same and the default is 800 why are we debayering at iso 3200?

Also how is there any way to compare the two camera sensor sensitivities to a xyla chart when Monstro has a custom LUT on top of Log3G10?