MP4 vs AVCHD

[Chris Borjis]

It’s usually best to acquire / record in the highest bit rate possible.

[Andy Lewis]

Doesn’t lumetri handle colour in 32-bit float? Surely this means that there is no colour space advantage in converting to Prores.

[Tero Ahlfors]

[andy lewis] “Doesn’t lumetri handle colour in 32-bit float?”

It’s using internal 32-bit floating point math when doing colour changes. It has nothing to do with the actual bit depth of the footage and it doesn’t make bad formats better.

[Andy Lewis]

[Tero Ahlfors] ”
It’s using internal 32-bit floating point math when doing colour changes. It has nothing to do with the actual bit depth of the footage and it doesn’t make bad formats better.”

I realise it doesn’t make bad formats better. I’m saying that conversion to Prores will make no difference to the eventual bit depth at which colour is processed.

[Jeff Pulera]

AVCHD, MP4 and in fact most common camera formats are heavily compressed using Long-GOP. Without going into too much detail (you can Google it), a GOP is a Group of Pictures.

Using the HDV format as an example, there is only one “full frame” of video every 15 frames. The frames between the full frames contain only partial data – think of them representing the difference in the image from that full reference frame – what is changing in the picture.

You have a camera locked on a tripod pointed at the street. Nothing in the frame is moving, so basically no difference frame to frame. Then a car drives by left to right. From one frame to another, the car is changing, but the rest of the image is still static. So only the data for the moving car needs to change from one frame to another – the background is the same and that data does not need to be encoded/saved for every frame, as it can be borrowed from neighboring frames. Encoding only the changes between frames saves space = compression.

So for playback, most frames need to reference other frames (preceeding and following) to put a full frame back together from bits and pieces. This is called an inter-frame format. The problem is that with high detail and fast motion, the codec can’t keep up. Think of shooting a water fountain or leafy trees blowing in the breeze. Basically every pixel is changing from one frame to another, and then the image starts to get full of artifacts/blocks because the partial-frame scheme is not able to fully encode all that ever-changing data and it just gets sloppy.

So as others have stated, if you have the chance to go from 25Mbps to 35Mbps, every bit helps!! Besides losing parts of each frame to compression, color data is also thrown away to save space. AVCHD is going to use 4:2:0 color. The human eye doesn’t notice the missing color fidelity (the “0”), however when you try to do chroma keying or color correction, it doesn’t work as well since a lot of color data is just not there for the NLE to work with!

Now let’s talk ProRes. It used Intra-Frame compression, meaning every single frame contains all the data to rebuild that frame – it doesn’t need to borrow info from any neighbors. And the data rate is much much higher, again increasing the fidelity of the image, being much truer to what the camera saw (when recording direct to Prores such as using an Atomos recorder). You are not “throwing away” tons of data and averaging things out and compressing into the “blocks of similar color” that you might view as compression artifacts with lesser codecs. Full HD ProRes might run roughly 100Mbps to 220Mbps depending on quality level chosen – quite an improvement from 25Mbps!

Another benefit of ProRes is 4:2:2 color, much truer to the original, providing more color info that really benefits keying and color correction. Can also be 10-bit, representing millions more colors than the 8-bit of AVCHD, to lessen color gradient banding and such.

So you asked if a 35Mbps codec stays 35Mbps when converted to ProRes and the answer would be no. You are taking those Long-GOP partial frames of AVCHD and converting each to its own unique full frame, fully encoded to stand on its own. Basically “uncompressing” each frame to a large degree, so obviously that will use more data. It’s not going to make any individual frame look any better than it started out, but it will play back easier since it doesn’t have to look back/look ahead at multiple surrounding frames before it can be played/viewed.

And you still only have the original 4:2:0 color data to work with. However, if adding color effects, titles, etc. to a ProRes clip then those will benefit from being done in the 4:2:2 color space with higher fidelity. And then if you do any compositing, such as exporting your edits to a new ProRes clip, and then doing further work in the new clip, you have the unique frames at a high data rate and with 4:2:2 color, so it all holds up so much better over multiple generations, unlike AVCHD that is so highly compressed that all the artifacts (quality issues) just pile up and get worse with each copy.

So, if you don’t plan on doing multi-generational compositing and such – you’re simply going to do some editing (cutting) of clips, add some transitions – and then just export to the delivery format, converting to ProRes probably isn’t going to make any difference, other than taking time and filling up your hard drive. If you have an older, slower computer that chokes on AVCHD, then ProRes should be easier to play back, but again as others mentioned ProRes needs faster hard drives to keep up, so there is a trade off.

Hope this kind of puts things in perspective. Do some Google searches for “4:2:2 vs 4:2:0” and “Long-GOP” and do some reading, it will really all make more sense then.

Thanks

Jeff Pulera
Safe Harbor Computers

[Andy Lewis]

[Jeff Pulera] “And you still only have the original 4:2:0 color data to work with. However, if adding color effects, titles, etc. to a ProRes clip then those will benefit from being done in the 4:2:2 color space with higher fidelity. And then if you do any compositing, such as exporting your edits to a new ProRes clip, and then doing further work in the new clip, you have the unique frames at a high data rate and with 4:2:2 color, so it all holds up so much better over multiple generations, unlike AVCHD that is so highly compressed that all the artifacts (quality issues) just pile up and get worse with each copy.”

If you are adding colour effects within Premiere / Speedgrade then everything is processed in 32-bit float so the conversion to Prores has given you nothing. Without conversion, the software takes that 4:2:0 long-GOP data and upsamples it to discrete frames at 4:4:4 on the fly. Note: I am well aware that no quality has been added – the software is simply reducing rounding errors.

If you have converted to Prores, Premiere will upsample from 4:2:2 to 4:4:4 on the fly. Same result but with the inherent quality loss of transcoding to Prores (another compressed format).

It’s absolutely true that 4:2:2 Prores will hold up better than AVCHD over multiple generations but what kind of workflow necessitates multiple generations? If you are working within Adobe apps that should never be an issue. If you are sending out to Resolve through XML then there will also be no generation loss as Resolve will process those AVCHD files in 32-bit float.

The only typical reason I can see to transcode to a mezzanine codec like Prores is if it plays back smoother.

[David Roth weiss]

[andy lewis] “It’s absolutely true that 4:2:2 Prores will hold up better than AVCHD over multiple generations but what kind of workflow necessitates multiple generations? If you are working within Adobe apps that should never be an issue.”

Andy, if you’re not thinking about generation loss you really should be, cuz any finished master you create will in all likelihood wind up being reedited or repurposed at some point later in history, and history lasts a long, long time, as in perpetuity. Of course, if you have archived versions of the entire finished project, and assuming the project file and all media are intact and readable at that later date, whenever it might be (which is ALWAYS questionable), you could POTENTIALLY avoid generation loss. However, I’d hate to think that you and other Cow members are ignoring the long-term future of your completed video assets, which are all hard-won and very labor-intensive efforts.

As an example of what I’m referring to, a Cow member once asked me why he should save his ProRes master rather than a much smaller h.264 copy of the same. I asked him if, twenty or thirty years from now when a retrospective of his work is being played at the Academy, in 20K perhaps, would he want it to look like complete dog doo? Would you or your clients?

David Roth Weiss
Director/Editor/Colorist & Workflow Consultant
David Weiss Productions
Los Angeles

David is a Creative COW contributing editor and a forum host of the Apple Final Cut Pro forum.

[Andy Lewis]

Thanks David. I’m talking specifically about the use of a Prores transcode as part of a grading or compositing workflow, not about mastering or archive.

[David Roth weiss]

[andy lewis] “Thanks David. I’m talking specifically about the use of a Prores transcode as part of a grading or compositing workflow, not about mastering or archive.
“

I understand Andy, but honestly, the transcoding decisions we make today as part of a grading or compositing workflow, and mastering/archiving are really ALL very much tied together.

Yes, you can discuss each operation on it’s won, but ultimately, every editor should be thinking about the “entire” media pipeline, which often (or usually) stretches well beyond the more immediate decisions most typically think of.

As one of the now “elder statesmen” here, I just want to make certain that ALL Creative Cow users think about the long-term realities of asset management that typically only those of us with 20, 30, or 40-years of experience have seen through our careers. Make sense?

David Roth Weiss
Director/Editor/Colorist & Workflow Consultant
David Weiss Productions
Los Angeles

David is a Creative COW contributing editor and a forum host of the Apple Final Cut Pro forum.