Blackmagic Forum

Hi! As the subject says im trying to figure out what codec is the best choice for exporting my timeline.
My files are 4k mp4 100mbit h264/65
Exporting to h264/65 seems not to be the right one as im loosing a lot of Detail :/.
My final Videos will be uploaded to vimeo, YouTube, Facebook and some other files for customers.

Export ProResHQ or DNxHR HQX and then ffmpeg, Vdub2, Handbrake etc to h264 with eg CRF=15 setting. It will be relatively small but good quality master for upload to youtube. With good bandwidth you can push ProRes itself.

Okay i see.
ProRes and DNxHR should be used when the material has to be edited again in a later process?

Try giving the ProRes master to YT as an upload... Depending on the lengths of your clips (thus on the size), it can be accepted.
I always made fine h264 especially for YT/Vimeo, but finally found out that the full ProRes "mother" export was ok too, despite its xxx Go. I have 400 MBs upload/10 GBs DL bandwidth and don't
Just try...

Well I'm using H265 (the NVIDIA option on the studio version) at defaults and it's looking perfectly adequate for YouTube uploads -- because YouTube is going to do much worse things to your footage than the NVIDIA H265 encoder does.

H265 works better (for me) than H264 because, for the same size of rendered file, the quality is much higher. Also, the NVIDIA H265 encoder runs like a scalded cat at 100-200FPS and I hate sitting around waiting for a video to be rendered

As others have said, you can upload in ProRes or other less-compressed formats but I doubt you'll see any difference in the footage that YouTube eventually spits out.

Well, I think you may overlook the reality of the whole process, the problem is indeed that YT won't take your footage for granted...
YT will surely rotten your clip ? Yeah, that's right, for sure; but even if you send it something meeting its basic/medium/high requirements aiming at something close to "WYSIWYG @ home", what's going to happen more specifically ?
YT will re-encode your upload anyway, even if it was super-ok in the first place. And this is going to be a second encoding actually, see ? Consequently, well, maybe you fortunately don't see any difference, but what the nvidia encoder produces is altered anyway (yeah, it's disappointing) , so what's the main point at the end of the day ?
All that regardless to the webm codec Google has been promoting for years, whispering on its own because of the MPEG-LA mastering the widely adopted h264/5 format, another - boring - topic

Imho, if your clip is not too long which would mean oversized too, also try your main uncompressed codec export once.

Webm is a container not a codec. The codec is VP9, but of course with youtube the codec actually used depends on the capabilities of the client that plays the video.

Thank you all!
Exporting with Dnxhr and using handbrake to encode with h264 works great .
However i dont understand why resolve has no options for h264 for better quality.

I cannot find prores. Do i have to by it, and it will then be avaiable in resolve?

If you are on WIndows then you will not see ProRes show up.
That option is available only under macOS.

Update your profile signature with the specifications of your system (like Processor, RAM, Graphics Card, OS etc..) and it will make life easier for other members trying to support you.

Take QuickTime and the MPG4 codec.

Resolve does not produce good h.264 videos. They have artifacts all the time until you set key frames to 1 (i.e. each frame). MPG4 size is smaller (than h.264 with keyframes at 1) and has better quality - even after uploading on YouTube.

h.264 really sucks. I do not use it anymore.

Try uploading a DNXHR export direct to youtube/vimeo. They definitely support Prores and DNXHD uploads because thats what I do. DNXHR might be too new, but ffmpeg supports DNXHR and theres always a chance youtube/vimeo are using some fork of ffmpeg.

I try to avoid h264 unless its an end usage (i.e. purely for viewing and not re-encoding/archiving)

For some reason, YouTube video quality appears to be a high topic of discussion of late.

RCModelReviews wrote:... YouTube is going to do much worse things to your footage than the NVIDIA H265 encoder does...

I wouldn't say YouTube's encoding (whether vp9 or AVC) is "bad" per se, it's simply that it is (necessarily) bitrate-constrained. In my estimation (based on quality metrics applied to downloaded YouTube encode files) the (maximum) VP9 encode quality for 4K is typically (depending on content complexity) around that of x264 encoded at CRF 27 - 28. Here's an example:

https://forum.blackmagicdesign.com/viewtopic.php?f=21&t=83493&p=464648#p464456

I often post footage to YouTube that involves a lot of small detail and rapid movement -- this inevitably stresses the codecs they're using to the point that massive compression artifacts are the issue.

Flying a small quadcopter at speed over autumn leaves through dappled light is a perfect example of this. Often, the footage I upload to YouTube looks pretty good after rendering (H265 at "best" quality) but inevitably the VP9 (or whatever) encoding that YT applies makes it look like a muddy mess once it's been uploaded and "mashed".

Hence, I don't see much advantage of using any format other than the NVIDIA H265 encoder, at least for most of my videos.

RCModelReviews wrote:I often post footage to YouTube that involves a lot of small detail and rapid movement -- this inevitably stresses the codecs they're using to the point that massive compression artifacts are the issue.

That's why we use MOV with MPEG4. No isses any more. h.264 is not good at all.

Bruce is referring to compression artifacts associated with the YT encoding.

JeffreyWalther wrote:That's why we use MOV with MPEG4. No isses any more

I don't see how exporting MPEG-4 (Simple@L1 profile only) for upload to YT would mitigate that.

Upload in ProRes/DNxHR, no point to give them H.264 as it's going to be transcoded into VP9 anyway. You get more bits per pixel if you output HD timelines at 4K, too.

Jack Fairley wrote:Upload in ProRes/DNxHR, no point to give them H.264 as it's going to be transcoded into VP9 anyway. You get more bits per pixel if you output HD timelines at 4K, too.

That used to be the case -- not so much any more. I did some tests a while back and the bitrate for 1080p from YouTube was the same whether you uploaded in 1080p or 4K. If you want reasonable upload times then a high-quality H265 file at 1080p is going to give better results than an upscaled 4K file of the same size (at an obviously lower quality).

RCModelReviews wrote:That used to be the case -- not so much any more. I did some tests a while back and the bitrate for 1080p from YouTube was the same whether you uploaded in 1080p or 4K. If you want reasonable upload times then a high-quality H265 file at 1080p is going to give better results than an upscaled 4K file of the same size (at an obviously lower quality).

h.265 is the best option unless your clips are only a minute or two long, anyone who says otherwise doesn't really know what they are talking about imo.

Render out a ProRes or DNxHR master, and then transcode to h.265 in crf mode, with one of the many free tools available.

RCModelReviews wrote:I did some tests a while back and the bitrate for 1080p from YouTube was the same whether you uploaded in 1080p or 4K. If you want reasonable upload times then a high-quality H265 file at 1080p is going to give better results than an upscaled 4K file of the same size (at an obviously lower quality).

I don't find that to be the case with x264 at least, based on quality metrics.

Here I took a 2160/50p Cineform (Filmscan 2) transcode of the CrowdRun sequence used in the test series I referred to above:

https://forum.blackmagicdesign.com/viewtopic.php?f=21&t=83493&p=464648#p464456

I down-scaled/transcoded the Cineform file to 1080/50p x264 at CRF=14. The bitrate of the resulting encode was 90Mbps. So I transcoded the file to x264 again as 2-pass encode with a target bitrate of 90Mbps at both 1080/50p and the original 2160/50p resolution. The resulting bitrates were confirmed to be 90Mbps.

I then uploaded both (2-pass) encodes to YouTube and downloaded the YT 'converted' 1080/50p files revealed by 'YouTube Info'.

https://www.h3xed.com/blogmedia/youtube-info.php

In fact, with 4K uploads, YouTube encodes both vp9 (webm) and AVC (mp4) files at 1080p, but with 1080p uploads it only encodes AVC mp4.

I then ran SSIM quality metrics on the downloaded 1080/50p files (using the 90Mbps 1080/50p x264 upload file as reference).

With the 1080/50p x264 (90Mbps) upload the SSIM score of the 1080/50p AVC (mp4) YT download was 84.197. The file bitrate was 5899 Kbps.

With the 2160/50p upload, the results were:

1080/50p AVC (mp4) download, SSIM: 84.194, Bitrate: 5895 Kbps
1080/50p vp9 (webm) download, SSIM: 86.874, Bitrate: 6427 Kbps.

So there was no difference in the metric quality (and bitrates) of the YT converted 1080/50p AVC mp4 files, but the 1080/50p vp9 (webm) encode from the 2160/50p upload was measurably higher quality and a slightly higher bitrate.

Here are the respective YouTube videos:




Incidentally, the uploaded 1080/50p x264 CRF=14 transcode (90Mbps) produced similar results to the 90Mbps 2-pass x264 uploads:

1080/50p AVC (mp4) YT download, SSIM: 84.531, Bitrate: 5926 Kbps

Edit: I also uploaded a 2160/50p x264 CRF=14 transcode of the original Cineform source. Obviously the encode bitrate was much higher - 509 Mbps. The downloaded YT 1080/50p files gave the following results:

1080/50p AVC (mp4) download, SSIM: 84.635, Bitrate: 5955 Mbps.
1080/50p vp9 (webm) download, SSIM: 87.351, Bitrate: 6546 Mbps.

So uploading at the higher (5.7 fold) bitrate did bring about a small improvement in the (metric) quality of the YT 1080/50p encodes, the vp9 file more so than the mp4 file.

Again, here are the respective YouTube videos:

My preference is to export Cineform YUV files. The final YouTube results look great.

That's what that earlier series of tests looked at:

https://forum.blackmagicdesign.com/viewtopic.php?f=21&t=83493&p=464648#p464456

And opinions differed on whether a 2160/50p Cineform (10bit 422, Filmscan 2) upload produced a better looking 4K YouTube video than a transcoded x264 CRF14 (8bit 420) upload.

Note: the bitrate of the 2160/50p x264 CRF14 transcode there was a bit higher (526Mbps) than in this series (509Mbps). In those first tests I created the Cineform file from the original 10-bit RGB SGI image sequence. In these tests I used an 10-bit YUV422 MagicYUV archive of original sequence. For some reason it generated a Cineform file with a slightly lower bitrate (3124 Mbps).

There is simply no way you will see the difference in "home watching" as x264 at CRF=14 is going to be very good quality. Youtube will destroy it so much that any source difference will be "masked".

Andrew Kolakowski wrote:There is simply no way you will see the difference in "home watching" as x264 at CRF=14 is going to be very good quality. Youtube will destroy it so much that any source difference will be "masked".

Yep, and i posted research here a while back that showed it for h.265, an h.264 is no different.

I use H.264 and select Best Quality. I upload the videos to WeTransfer.com(2 GB upload are free). I email the videos to my clients.

Andrew Kolakowski wrote:There is simply no way you will see the difference in "home watching" as x264 at CRF=14 is going to be very good quality. Youtube will destroy it so much that any source difference will be "masked".

I agree. There was an opinion that the motion cadence of the YT video produced from the x264 CRF=14 upload was more blurry and 'jumpy' than the Cineform one.

https://forum.blackmagicdesign.com/viewtopic.php?f=21&t=83493&p=464648#p464574

Personally I couldn't see a difference viewing on YouTube or the downloaded YT (vp9.webm) files.

JeffreyWalther wrote:Take QuickTime and the MPG4 codec.

Resolve does not produce good h.264 videos. They have artifacts all the time until you set key frames to 1 (i.e. each frame). MPG4 size is smaller (than h.264 with keyframes at 1) and has better quality - even after uploading on YouTube.

h.264 really sucks. I do not use it anymore.

Mike1938 wrote:I use H.264 and select Best Quality. I upload the videos to WeTransfer.com(2 GB upload are free). I email the videos to my clients.

I've tested Resolve H264 and MPEG-4 also. Used the Media Mgt > Transcode function to transcode the 2160/50p Cineform FS2 'source' to H264 'Best' quality with key frame interval set at default (Auto) and 1 (All-Intra), and MPEG-4 'Best' quality, all in MOV container.

As before, uploaded the files to YouTube and downloaded the converted 2160/50p vp9 (webm) files via 'YouTube Info'. Ran SSIM tests using the Cineform FS2 files as reference. The results:



The SSIM score of the MPEG-4 render was a fair bit lower than the other upload files, but the resulting YT download scored only slightly lower than those of the H264 uploads. I've yet to visually examine the files close-up. Here are the respective YouTube videos:

Cineform (FS2)


x264 CRF14:


H264 'Best':


H264 'Best' All-Intra:


MPEG-4 'Best':


Feel free to scrutinize. Being a creature of habit, I'll stick with x264 CRF14.

Bryan Worsley wrote:

Andrew Kolakowski wrote:There is simply no way you will see the difference in "home watching" as x264 at CRF=14 is going to be very good quality. Youtube will destroy it so much that any source difference will be "masked".

I agree. There was an opinion that the motion cadence of the YT video produced from the x264 CRF=14 upload was more blurry and 'jumpy' than the Cineform one.

https://forum.blackmagicdesign.com/viewtopic.php?f=21&t=83493&p=464648#p464574

Personally I couldn't see a difference viewing on YouTube or the downloaded YT (vp9.webm) files.

Personal imagination

Andrew Kolakowski wrote:Personal imagination

Confirmation bias!

I thought the same, but how would he know which video was which...unless he went to the trouble of downloading the original SGI sequence and replicating the tests himself ?

Even got me wondering also it's possible for someone to find out the original name of an upload file. The received wisdom is that it is not, because a unique YT name is assigned to the uploaded file for processing. But if you upload the same raw file twice YT notifies you that "the video is a duplicate of a video you already uploaded" and the processing is aborted, so evidently the raw file name is retained and internally referenced.

Bryan Worsley wrote:

.... I've yet to visually examine the files close-up.

Couldn't pick out overt compression artifacts in the upload files; the Resolve MPEG-4 render was a just a teeny bit softer than the others. As for the downloaded YouTube vp9 files, well of course there was visible degradation - here including the original Cineform (FS2) source file for comparison:



After opening, click on (+) cursor to enlarge.

Bryan Worsley wrote:opinions differed on whether a 2160/50p Cineform (10bit 422, Filmscan 2) upload produced a better looking 4K YouTube video than a transcoded x264 CRF14 (8bit 420) upload.

I think that's the wrong comparison. Cineform can be had directly out of Resolve. You need to compare that to H.264 coming directly out of Resolve.

Jim Simon wrote:

I think that's the wrong comparison. Cineform can be had directly out of Resolve. You need to compare that to H.264 coming directly out of Resolve.

I wouldn't recommend anyone use H.264 directly out of Resolve. You are very limited with regards to options, and imo seems to do a not so great job.

Why would anyone use H264 when H265 is more efficient size/quality-wise?

Remember the title of this thread when you answer.

Jim Simon wrote:

Bryan Worsley wrote:opinions differed on whether a 2160/50p Cineform (10bit 422, Filmscan 2) upload produced a better looking 4K YouTube video than a transcoded x264 CRF14 (8bit 420) upload.

I think that's the wrong comparison. Cineform can be had directly out of Resolve. You need to compare that to H.264 coming directly out of Resolve.

+1
And ideally: take the same source.

Because there: no way to compare what could be comparable.
(Just my humble opinion )

Jean Claude wrote:

Jim Simon wrote:

Bryan Worsley wrote:opinions differed on whether a 2160/50p Cineform (10bit 422, Filmscan 2) upload produced a better looking 4K YouTube video than a transcoded x264 CRF14 (8bit 420) upload.

I think that's the wrong comparison. Cineform can be had directly out of Resolve. You need to compare that to H.264 coming directly out of Resolve.

+1
And ideally: take the same source.

Because there: no way to compare what could be comparable.
(Just my humble opinion )

Well, the test series in the other thread that quote was taken from was perfectly valid as it examined the relative intrinsic merits of uploading to YouTube a large file mezzanine 10bit 422 format (like Cineform) vs x264 encodes derived from it - something that could be applied in any context.

Fair points about comparing Resolve's H264 (and MPEG4) in that way though. It was really an after-thought add-on in the light of comments raised.

The CrowdRun 'master' I have archived is in MagicYUV 10bit YUV422 format which was convenient for converting to Cineform externally (with VDub2). To compare Cineform, H264 and MPEG-4 Resolve exports from the same source I'll have to look at converting the master to a lossless format that Resolve will import - Uncompressed 10bit 422 (v210), basically. I'll retest when I have more time.

RCModelReviews wrote:Why would anyone use H264 when H265 is more efficient size/quality-wise?

Remember the title of this thread when you answer.

Because h264/5 has only meaning if you have crazy slow upload speeds and need to shrink your master to very small size. Otherwise you can upload ProRes, Cineform, high bitrate h264/5 etc and you will get visually the same quality as youtube adds a lot of compression on top of every uploaded source.

It has been also discussed countless times, yet every few days there is the same question. It's getting boring.
Should be enough to use search option to get an answer, but people are to lazy to do even this.

Bryan Worsley wrote:The CrowdRun 'master' I have archived is in MagicYUV 10bit YUV422 format which was convenient for converting to Cineform externally (with VDub2). To compare Cineform, H264 and MPEG-4 Resolve exports from the same source I'll have to look at converting the master to a lossless format that Resolve will import - Uncompressed 10bit 422 (v210), basically. I'll retest when I have more time.

Don't waste your time- you will get the same results.

Andrew Kolakowski wrote:

RCModelReviews wrote:Why would anyone use H264 when H265 is more efficient size/quality-wise?

Remember the title of this thread when you answer.

Because h264/5 has only meaning if you have crazy slow upload speeds and need to shrink your master to very small size.

That's not the only reason. YT has an upload limit of 128 GB.

So, if you're working with UHD 30p footage and render out a master as DNxHR HQ you will hit the limit at ~21 minutes. Higher quality codecs or frame rates will cuts this down even more.

its a real issue if you are doing any kind of long form work.

In this case you don't upload DNxHR for sure, but high bitrate h264 or h265. For HD CRF=14 should give you good few hours before you hit 128GB, so not an issue. 128GB will hit you sooner with UHD.
No one stops you to upload h265 if you happy with export times.

Andrew Kolakowski wrote:

Bryan Worsley wrote:The CrowdRun 'master' I have archived is in MagicYUV 10bit YUV422 format which was convenient for converting to Cineform externally (with VDub2). To compare Cineform, H264 and MPEG-4 Resolve exports from the same source I'll have to look at converting the master to a lossless format that Resolve will import - Uncompressed 10bit 422 (v210), basically. I'll retest when I have more time.

Don't waste your time- you will get the same results.

As it serves to illustrate that point:



Note: Cineform FS2 (Filmscan 2) = 'Best' quality in Resolve.

Yes, I know, I probably could have gone on to run the x264 and x265 encodes in two-pass mode to match the bitrates, but I'd had enough.

Edit:
With x264 and x265 encoded from the Resolve Cineform FS2 export:



With x264 and x265 encoded from the v210 Master:

Bryan Worsley wrote:Don't waste your time- you will get the same results. As it serves to illustrate that point:

@Juan Salvo has said the same thing several times, that YouTube recompresses everything uploaded anyway, and your download numbers would appear to confirm this. Whether you upload 3292mbps or 359mbps, it still comes out as 31.5mbps. So YouTube stomps on it pretty hard.

My workaround is to avoid YouTube and to instead use Frame.io and Vimeo Pro, which give you more bandwidth and more control over what you upload. Frame.io has the added extra of allowing clients to comment while the video is playing, so you can download a list of their fixes and issues.

There is a theory that 4K uploads are not as compressed when you watch them in HD, but this is anecdotal so far. YouTube's own suggestions for recommended upload bitrates are here:

https://support.google.com/youtube/answer/1722171?hl=en

Marc Wielage wrote: YouTube's own suggestions for recommended upload bitrates are here:
https://support.google.com/youtube/answer/1722171?hl=en

If YouTube's bitrate recommendations are followed for this source the resulting YouTube video will be pretty degraded.

Here I exported the v210 Master to H264 with the bitrate limited to 68000 Kbps (i.e. Constrained Constant Quality). The resulting encode bitrate was 69.4 Mbps. I also exported to Cineform FS2 ('Best') and transcoded to x264 CRF 23 with FFMPEG. Resulting bitrate was 66 Mbps.

Upload: Resolve H264, Bitrate: 69.4 Mbps, SSIM: 89.29
YouTube VP9 Download, Bitrate: 30.7 Mbps, SSIM: 86.51

Upload: x264 CRF 23, Bitrate: 66Mbps, SSIM: 90.42
YouTube VP9 Download, Bitrate: 25.9 Mbps, SSIM:87.61

SSIM measured with the v210 Master clip as reference.

Here are the cropped frame grabs from the YouTube downloads.