crf 18?

[Fishman0919]

If crf 18 at default setting is considered transparent, would --preset slower --tune film --crf 18 --level 4 --merange 24 be over kill?

[Asmodian]

No, "transparent" is not well defined and crf 18 default is often not transparent in my experience.

[Neroldy]

In my opinion, transparent depends on the source, your device and your requirement. crf 18 is just a bitrate factor. And --preset slower --tune film --crf 18 --level 4 --merange 24 is just make the encode better...

[EncodedMango]

As mentioned by others, 'transparent' really depends on the quality of the source video. That said, I personally find CRF18 to be transparent for lossless self recorded sources.

[movmasty]

If crf 18 at default setting is considered transparent, would --preset slower --tune film

--crf 18 --level 4 --merange 24 be over kill?No, "transparent" is not well defined

and crf 18 default is often not transparent in my experience.
I would say that overkills, but Doom's users really dont like to compress movies :p
i never go below 23 ;)

CRF is not the same as quantizer, and then if Q 18 is equal to old Q 2, and Q 2 is good, and then CRF should be 18.
A CRF of 18 should give about Q 12 for I and 19 for B frames,
Maybe lower than the original BR.

CRF is almost an average for all I-P-B frames,
to have an average of 18 for I frames(that give the "tone" to all the movie if there is enough of them) you can use 23/23.7

Even if, i watch at a scale where 18 isnt =2 but 2.24
The usual conversion formula
12+6*log2
has some problem:
Why the default min Q is 10? What is the sense to compress at less than Q 1??
Then the max 51 would be around 90, too high,
and to store number 90 you need 7 bits, but 7 bits can go up to 127....

if we assume 10=1 and 51=63 (6 bits 0-63), we'll have
10+6,86*log2
here 16.86=2 - 18=2.24 - 23.72=4 - 30.6=8
The low quantizers are no so low, and the higher not so high.
These are not round numbers like the wrong conversion, but i think more precise.

[movmasty]

And...as an average CRF depends from IPB stucture

if you have B=nonB, and nonB Q=18 B=24, then CRF ~21
but if B=2nonB, at nonB Q=18 B=24 CRF will be ~22

[kuchikirukia]

Preset placebo is overkill, that's why it's called placebo. Preset slower doesn't even have 5 consecutive b-frames, and most sources will regularly go out to 5.


consecutive B-frames: 4.0% 4.1% 12.7% 20.7% 12.6% 32.8% 4.4% 4.4% 2.4% 0.7% 1.2%


CRF doesn't necessarily result in the same quality for all settings, either. Ultrafast is not going to look like Very Slow at the same CRF.
Resolution also plays a part. You need closer to CRF 13 for DVD. 720p is around 16/17.

[benwaggoner]

As mentioned by others, 'transparent' really depends on the quality of the source video. That said, I personally find CRF18 to be transparent for lossless self recorded sources.
Frame size is also a significant factor for required CRF. 2160p can look great with a higher CRF than 240p, since any artifacts would be so much smaller.

[benwaggoner]

You need closer to CRF 13 for DVD. 720p is around 16/17.
For perceptually lossless? For identical in an A/B comparison of paused frames?

"Quality" itself is ill defined. The QA manager at banking software company I worked at 20+ years ago said the only broadly applicable definition was "quality is fitness for use."

So we really need to define the use case to know what adequate quality is. The difference between "when playing at full speed" and "when paused on a frame" alone has a huge impact on needed CRF.