I want to export a video from “Default” export options. I want to set “Default” export options to the highest quality. How could I achieve it?
NOTE 1: With “to the highest quality” I mean, a quality near the original video which loses no/little quality from original. NOTE 2: These are the values by default to: Default export NOTE 3: My Video Mode is set to: Master Properties (from: Master -in the timeline- --> Properties.)
If that is your aim you’re going to want to change the preset to H.264 High Profile. After that set the crf to 17 or 18 (64% to 67%) which will get you the video quality that you are looking for (“near the original video”).
I don’t know what would be the equivalent for audio. @Austin, is there a bitrate that is the audio equivalent of “visually lossless”?
Actually Default is 55% (crf=23). You can raise it to what @DRM suggested.
The highest quality is one of the presets in the lossless category, but it will create very large files that many media players cannot handle or use hardware acceleration on playback. If you are looking for highest quality that is about the same bitrate as the original, then we do not provide that. You can look up your source’s bitrate and try to target that, but if the source bitrate is low it will look like crap. Each decompression and lossy compression cycle is a generation lost, and the size of that generation depends on the amount of compression.
I got the source’s bitrate from my O.S. (Windows 8.1), but it is an overall bitrate, so, it gives me only a general idea about the quality of that video. Steps: Right-clic over the file --> Properties --> “Details” tab --> “Total bitrate”.
Then, as you can see in the picture, I should raise the quality value more than 67% in “Default” export to approaching to Original video’s Total bitrate. My question is: Could I raise it more or I could get some weird behaviour?
So, I chose some export methods (which I do not remember) and I exported a video file of 60KB approximately from this project. Then those export methods had an huge amount of decompression and lossy compression. Did I understand it well? If so, how could I fix that problem from those export methods?
Well, I actually recommended that if you want to get higher quality then use the H.264 High Profile preset, not the Default. Still, I am surprised that the Default at 67% gave you a smaller file. I don’t remember that ever happening to me. 67% at H.264 should give a file that’s at least bigger than the original. Did you by any chance have the hardware encoding on? Or did you switch the codec in the Default preset to H.265?
AAC 384k or AC3 448k is generally considered lossless enough for a final file. But if the file is going to be transcoded again and needs to survive another generation or two, then may as well max it out to 640k. (This would include uploading to YouTube.) To the OP… obviously, for highest quality, a lossless format would be used instead of the lossy Default settings, but that would make a bigger file.
As for video settings, a decision needs to be made here. Is quality more important than bitrate, or is it the other way around? The settings are mutually exclusive (for best results).
If using a quality percentage setting, libx264 will target a quality standard and will use whatever bitrate necessary to achieve it. There is no looking up the original bitrate and trying to apply it when using CRF quality modes. It will be whatever it needs to be to maintain quality, and bitrate is out of your control. But, if you’re targeting a bitrate, then quality will become variable (and compromised) in order to meet the bitrate cap.
Since your goal started out as “highest quality”, then @DRM’s recommendation will do exactly what you want. If the video needs to survive a generation (including YouTube), there could be benefit of even going to CRF 16.
However, if you need “highest quality” in the broadcast studio post-production sense, then like Dan said, you’ll probably need an intermediate or lossless file that’s huge, like ProRes or DNxHR.
“Highest quality” is pretty subjective and needs context before we can make solid recommendations. After all, a highest-quality encoding only makes good use of its huge file size if the original video came from a highest-quality camera. A lesser camera would create storage overkill. That’s why there’s “high quality” that’s plenty good enough for home video… then there’s high quality that’s required for commercial post-production houses that are doing additional editing… and then there’s high quality for archivists that can’t compromise a single pixel (and they pay an extreme storage price for this pixel hoarding). We would need to know which category you fall into. @DRM’s recommendation is ideal for home video, YouTube, and any other file that is final (as in, not going to be processed with further editing).
I exported the video from “Default” preset and “H.264 High Profile” preset to: 67% (as the quality value) and I got the same file (in size and total bitrate). Both presets have really similar values by default, here you can see the differences:
H.264 High Profile was created before YouTube and Default preset and kept for searching purposes. All generate high profile, which is something specific in terms of H.264. YouTube was added to give something closest to YouTube’s guidelines since quite a few people ask for that. Then Default was added so when people read “default” in tips and guides they can find something with that name (do not always understand default is implied).
That video contains photos about groups of people in different places: beach, mountain, … + text which go as slides through the video and I would like to create a: 3D image effect too. IMPORTANT: I want to keep their faces right. So, taking that in mind, I do not know whether that video would fit in home video category.
If the video is exported in CRF/quality mode, then matching bitrate with another video is a very difficult comparison. The real question is whether the CRF 16 files had quality near the original video, visually speaking. If they are visually of equal quality, then the file size and bitrate are immaterial. It simply means that libx264 was that much more efficient than the source video’s encoder.
I exported the video with 69% (crf=16) quality value (from “Default” preset) and they are similar (visually speaking). What about if I raise quality value to 90% as I did, would it has more quality than 69% (even if I can not appreciate those details)?
Technically, yes. But long story short: libx264 at 90% makes no sense in most cases. It makes the file larger but doesn’t make the image look better. This larger file is not suitable for any significant additional editing either, so there is literally no benefit to the higher bitrate. That’s a lose-lose situation.
libx264 (in this mode) is still a lossy encoder by design. It is taking advantage of weaknesses in human vision in order to save bitrate where the eye will not notice the difference. So this is where the choice of encoder depends on the situation.
If this video is the final video (with no further editing, and maybe one transcode generation for uploading to YouTube), then a lossy format like libx264 is acceptable and whatever looks good is good enough. Raising the quality to 90% will retain more detail, but your eyes would not notice or appreciate it.
However, if there is going to be additional editing on this footage (like it’s an intermediate or a segment for a larger project), then you need a lot more details than a libx264 encoding will get you. This is what the intermediate and lossless formats are all about. ProRes and DNxHR are examples of these formats. They produce much larger files, but also preserve sufficient detail that you can stretch colors in post-production without getting banding artefacts and other such problems.
libx264 by design throws away as much detail as it can in order to keep the file size down, even at high quality levels. This means any modifications made to colors by further editing (color grading, contrast, hue shift, shadow recovery, etc) will cause blockiness, banding, and many other unsightly problems. This is why libx264 is for final delivery only. ProRes and DNxHR focus on quality first and file size second in order to make further editing possible by retaining lots of detail. ProRes and DNxHR will also be smoother and easier to process during editing due to differences in their compression scheme compared to libx264.
So it all depends on your use case. There is a specialized tool for every situation.
Could a quality value between 80% and 90% be practical on large screens like: Smart TV, projector, … which have between 50’’ and 100’’ compared to 69%? I mean, I do not know whether the video file could lose quality (visually speaking) at 69% with a projector at 90’’.
NOTE: 50’’ = 50 inches.
So, if I uploaded that video file to Youtube, it would switch from libx264 codec to Youtube’s codec which would be one transcode generation, did I understand it right?
IMPORTANT: that’s the final video, so it will not be edited.
For anything less than a cinema screen, it is unlikely to make a difference. I would encourage you to test it on your videos to verify, though. There is a possibility that videos with lots of continuous movement or jitter could appear less “smeary” if the quality was slightly higher. libx264 knows that the eye can’t track detail in fast-moving objects, so it cheats on details there to reduce bitrate. These cheats could be more evident on larger screens, but again, that’s theoretical and not likely at proper viewing distances.
The general reason more quality isn’t always better is that the quality setting mainly determines color and brightness accuracy. The higher the quality setting, then the more data that’s available to pinpoint the color and brightness that should be shown for each pixel. Once the 68% quality threshold is passed, the human eye itself is not accurate enough to discern color variations in the encoding. The codec has surpassed human ability to critique it. This is why more quality becomes overkill for viewing, but not overkill enough for further editing that manipulates color.
Someone could make the case that increased luma detail could be noticeable on a 100-inch screen. They might be right if standing one foot away from the screen (and it would depend on the type of content). But if somebody is viewing the screen from a comfortable distance (able to see the whole screen at once for normal movie viewing), then they are much too far away to nit-pick individual pixel accuracy at that point anyway. It goes back to a non-issue.
If exporting for a cinema or IMAX screen, and for some reason the export was libx264 instead of the usual intermediate or JPEG2000 sequence, then 80% quality would make sense because there is a color transform to CIE XYZ color space required in order to match cinema projection standards. This is where that little extra color detail beyond human detection would actually be useful, to better survive the color space conversion.
What you might be wondering instead is, How do we get sharper images on 100-inch TVs? We get that through higher resolution source videos, editing, and exports. Resolution is a bigger factor in sharpness for large screens (to a point… there are steeply diminishing returns after 4K if the screen is only 100 inches).
Basically, all the quality setting does is determine the color accuracy within the export resolution chosen. And once that quality amount surpasses human color detection ability, then resolution has to be increased to get the next level of sharpness detail. Once resolution surpasses human vision, the next level of realism is an increased color gamut like BT.2020 to reproduce a wider range of colors and more completely mimic the world around us. I guess we could throw higher frame rate in as a step towards realism, but after that… the next step is bypassing the screen altogether and plugging directly into The Matrix.
Correct. Generally speaking, CRF 18 is considered “visually lossless” although it isn’t quite there if you know what to look for. CRF 16 usually is “visually lossless” even to the nerd-herd and is well-equipped to gracefully survive the one generation of transcoding that YouTube will do to your video as part of its standardization process.
Context: All the references to CRF 16/18/etc so far have been based on the assumption that the export is 1920x1080 or higher. For lower resolutions, the CRF has to go 14/12/10 the lower the resolution gets. If you’re curious, I’ll explain why, but it may not be relevant for now.