What codec are you editing?

[Evan Donn]

Originally Posted by Alexander Ibrahim

Not strictly true. In fact its only true if you have a 4:4:4 codec for the timeline.

Any effect you add to a time line is calculated based on uncompressed video... but at the bit depth and chroma sampling of the source sequence.

Motion, Shake, After effects and other graphics/compositing software behave differently. They do everything at 4:4:4 uncompressed, then throw away data right before output, exactly as you suggest.

Do you have a reference for this? I could swear that I've read in FCPs docs that all internal processing is done at 4:4:4 (at whatever bit depth/precision you have selected in preferences) and then downsampled to the timeline's format.

[David Schmerin]

First please let me say that compared to guys like Steve Mullen and Alexander Ibrahim, I consider myself to be a relative thickie. 90% passes so far over my head that I look at you guys in wonder...

When I edit EX1 HQ footage with FCP, I myself stay with the native codec. I ran a small test whereby I brought a 5 second video of a flag into FCP. Then I rendered the file out as H.264, ProRes, and Photo JPEG.

All render settings were set to best or as unrestricted as possible to allow the codec to produce the highest quality files. In every case I found the rendered files quality to be pale in comparison to the original.

If any FCP users would like to take a look at the results of my flag test, the raw, H.264, ProRes, and Photo-JPEG are posted on my FTP site at

Host: download.gotfootage.com
User: vegas
Pass: demo
Port: 21

Or you can paste the following link into Safari

ftp://vegas:demo@download.gotfootage.com

Test File Names:
flag.raw.mov
flag.H.264.mov
flag.ProRes.mov
flag.pjpg.mov

David Schmerin
www.GotFootageHD.com

[David Schmerin]

First please let me say that compared to guys like Steve Mullen and Alexander Ibrahim, I consider myself to be a relative thickie. 90% passes so far over my head that I look at you guys in wonder...

When I edit EX1 HQ footage with FCP, I myself stay with the native codec. I ran a small test whereby I brought a 5 second video of a flag into FCP. Then I rendered the file out as H.264, ProRes, and Photo JPEG.

All render settings were set to best or as unrestricted as possible to allow the codec to produce the highest quality files. In every case I found the rendered files quality to be pale in comparison to the original.

If any FCP users would like to take a look at the results of my flag test, the raw, H.264, ProRes, and Photo-JPEG are posted on my FTP site at

Host: download.gotfootage.com
User: vegas
Pass: demo
Port: 21

Or you can paste the following link into Safari

ftp://vegas:demo@download.gotfootage.com

I would very much to know if anyone else feels the same as I do about the rendered video files.

Test File Names:
flag.raw.mov
flag.H.264.mov
flag.ProRes.mov
flag.pjpg.mov

David Schmerin
www.GotFootageHD.com

[Alexander Ibrahim]

Originally Posted by Evan Donn

Do you have a reference for this? I could swear that I've read in FCPs docs that all internal processing is done at 4:4:4 (at whatever bit depth/precision you have selected in preferences) and then downsampled to the timeline's format.

The sequence settings panel, which I've a attached a screen grab of for you to peruse, supports part of what I am blabbering on about.

You can set rendering precision. 10 bit materials are rendered in 32 bit floating point (FP). You can force 8 bit materials to render in 32 bit FP

Now, I think some confusion sets in in that FCP creates all GENERATED imagery in RGB at 10 bits per channel. (32 bit RGB)

Another point that might confuse a reader is that in Volume 3 page 650 the manual says, "By default, render files are created at full quality, but you can speed up rendering by choosing lower-quality options in the Render Control tab and the Video Processing tab of the Sequence Settings window."

This has nothing to do with bit depth or chroma subsampling.

Finally if you are stupid just like me, as opposed to in your own peculiar way, then you may have confused material you read in the Shake manual with stuff you read in the FCP manuals. Man that still drives me crazy.

If you do a crossfade, that's done at the color space of the sequence. If you want that to happen in 4:4:4, then you have to have a 4:4:4 sequence.

This is standard amongst NLE's Quantel, AVID, FCP you name it.

This is a smart optimization by the software engineers. The majority of "rounding errors" will result in color levels that are unsupported, not colors out of gamut- even in 4:1:1 or 4:2:0 color spaces. So, you waste CPU cycles on extra bits and that makes a difference. Wasting them on increased colorspace is actually a waste. You are limited by the output codec- so never bother exceeding its capabilities.

The relevant section is in the Final Cut Pro Manual Volume 3 page 659.

If you are going to 32 bit processing though you may want to read Volume 3 pages 241-252 which contains notes on which effects are disabled when you are working in high precision YUV modes. HINT: Most of them.

Compositing software, like Shake, NUKE or Flame works in 32 bit FP all the time. Of course they work in RGB space, causing a whole different set of headaches.. but hey that's the game we are in right?

If you need an NLE that works this way then you really need a "Finishing system." I think AVID DS, Autodesk Smoke, Autodesk Fire, and anything from Quantel are the only NLEs right now that can work in 10 bit with all features all the time.

Avid Symphony Nitris is mostly 8 bit, but it offers 10 bit titling- sorta like FCP just wicked fast. Of course 8 Core Mac is also wicked fast- but Nitris can be wicked faster.

Enjoy

[Alexander Ibrahim]

Originally Posted by David Schmerin

First please let me say that compared to guys like Steve Mullen and Alexander Ibrahim, I consider myself to be a relative thickie. 90% passes so far over my head that I look at you guys in wonder...

Awww... shucks.

Well just keep on working- this stuff just seeps into your head as you slog through project after project.

In every case I found the rendered files quality to be pale in comparison to the original.

Well... H.264 looks like sh*%e of course. The ProRes and photo jpeg versions look fine on my system when I played them through Quicktime. They are a bit more deeply saturated than the XDCAM file.

Check your monitor calibration. Also make sure that high quality is set in the Quicktime movie properties panel. (I think its that way by default on QT7, but check it.)

[Sebastien Thomas]

Hi,

I would like to say, then, what is the right setup ?

Stay in XDCAM-EX (4:1:1) or go to prores (4:2:2) ?
or maybe some other format ?

Then how ? Is it just a matter of changing the sequence setting ?

Is this workflow good so :

import XDCAM-EX into FCP
use the media manager to re-compress as prores (4:2:2 1920x1080 HQ)
create a new sequence, set it as prores.

Will this improve the plugins/color correction you may do ?
What will go to Color then ? Will you get the same quality when coming back to FCP ?

I asked this kind of questions, to myself first, with the manual, then on some forums. Nobody have been able to answer yet. I may not habe the time nor the experience to try all this myself.

Thanks.

[Alexander Ibrahim]

Originally Posted by Sebastien Thomas

I would like to (ask) what is the right setup ?

For 90% (WAG) of the work people will do:
Import your footage as XDCAM.
Drop it in a sequence that is Set to ProRes, DNxHD, Cineform or some other high quality codec.

That's it. You will get VERY good results that way. Chances are you can stop reading now.

IF You plan on extensive adjustment or alteration in post production then you should get your material into a high quality codec as early in the production chain as possible.

Before I go into that- I need you to realize that I am talking about an edge case. I do that a LOT on this forum. 90% of videographers should never use any of the information that follows.

Every time you do any effect, from a crossfade, to adding titles, to heavy compisting and DI you reduce the quality of your image. Especially if you are working across different facilities with lots of generations of processing. These things are bad for the technical quality of the image, but are essential to achieving our practical results. Balance is the key.

The very best thing you can do is skip XDCAM altogether, and just use a superior codec from acquisition. If you can hook the camera up to a higher end capture solution, like the Convergent Designs Flash XDR, or AJA ioHD, or the proposed Cineform solid state recorder, those are all excellent options. I guess (wildly) that about 10% of the EX1's users will need to do this.

If you know your footage is going to be "abused" then uncompressed capture over SDI is the right thing. I estimate that 99% of users of the EX1 will never EVER need uncompressed. Of the remaining 1%, less than 2% of our footage will need to be captured uncompressed. The rest falls into the "capture with a better codec" category above.

Failing that you will get MARGINALLY better results by transcoding to a high quality codec. Encoding in XDCAM already threw away a LOT of sensor data- and there is nothing you can do to get it back. You are trying to stop the bleeding.

Often we end up back at square one- the best way to stop the bleeding is to leave the footage alone.

If you follow this path you have to consider what is going to happen to every bit of your footage. The goal being to transcode it as few times as practical- and then to the codecs most suitable for continued work on that bit of footage.

It is hard to draw this line- and only lots of experience will help you draw it intelligently. In the meantime, try what you need to do with the footage in its source codec, when that fails try improving the codec. Notice when this helps and try to figure out why.

HINT: It doesn't help that often.

If your footage will go through ten or more generations of processing then you need to consider converting to uncompressed.