Handbrake encoder settings

[umbala]

I've spent a good amount of time trying to figure out the best settings for encoding my video collection. I mostly have movies, TV shows and animations. I want the results to look the same as the source without any noticeable difference in quality while still reducing size as much as possible. Encoding speed is not important to me, but would like it to be reasonable. I use Android based TV boxes to play back content and backward compatibility with older or low spec devices is not important. My settings are already very aggressive and I'm looking for some advice on whether I went too far in some cases or perhaps not far enough.





Originally I was going to go with a Constant Quality Rate Factor of 16, but after doing some tests I decided to settle on 18 because I couldn't see a difference in quality. Any reason I should be going back to 16? Any settings I could change to further increase quality or reduce file size without encoding time getting really out of hand?

My biggest concern is some of the settings that were added manually such as mixed-refs, vbv-bufsize, vbv-maxrate, and rc-lookahead. Are they too aggressive or would they even have a noticeable impact on the final quality or file size? Should I use separate settings for film and animation or would these work equally well for both?

Due to these settings the encoding speed is fairly slow, but I suppose that's the price for quality. Any advice on how to increase encoding speed without any impact to quality or increase quality further while keeping or reducing file size would be greatly appreciated.

[KarMa]

Just use the x264 presets, using the slowest setting you can stand the better. CRF 18 is generally considered transparent for HD, with SD requiring a slight lower CRF than HD for similar quality.

[umbala]

I'm not trying to get "good enough" results, I'm trying to get the best possible quality and the smallest size. Using presets won't give me that.

[sneaker]

Your settings are almost identical to preset veryslow. Actually, preset veryslow uses higher rc-lookahead and merange. no-dct-decimate is part of tune grain. Hard to tell from your screenshots but you may also be missing out on high profile and "overshooting" level ref limit.

Don't shit on the presets if you don't understand them. There's a reason we recommend using presets + tunings.

[KarMa]

I'm not trying to get "good enough" results, I'm trying to get the best possible quality and the smallest size. Using presets won't give me that.

Use placebo then, it's pointless but it's there.

http://dev.beandog.org/x264_preset_reference.html

[derSimon]

I'm not trying to get "good enough" results, I'm trying to get the best possible quality and the smallest size. Using presets won't give me that.

Use placebo then, it's pointless but it's there.

http://dev.beandog.org/x264_preset_reference.html

AND one more encode with RF quality set to 51. And try to resize the resolution as much as possible, let's say 16*16 Pixels? That should take care of your request for "and the smallest size". There really aren't any presets for that.

[Bernix]

Hi KarMa,
i think he thinks Handbrake presets....
Best quality for H264 is somewhere around CRF 12, if you want lossless, you can achieve minimal difference at cost of very much big file. Even 12 seems to me be too low, CRF 14-16 should be good enough, especially using placebo, which is nonsense, slow should be enough.
HERE is time encoding example by presets. Video has 180 sec. On average computer, don't remember if H264 or H265, but relative times should be in same or in worst case similar ratio.

[Attachment 44437 - Click to enlarge]

Edit And compare statics images from video, with minimal difference is also IMHO nonsense.

Bernix

[Hoser Rob]

People new to video encoding often post here looking for that perfect set of xd264 presets. They don't exist. Each video has to be taken on its own terms.

And, since video encoding is actually very complex ... when those codecs were developed they were done so by pros FOR pros and the average user ripping their DVDS didn't enter into it at all ... the advice about using levels and tunes is good advice.

[sophisticles]

The absurdity of these types of questions, some guy has a "collection" of videos, which we all know are downloaded via torrents and ignoring the legality of doing so for a moment, they are always already encoded using lossy settings. The he wants to re-encode them but he wants the smallest possible size along with the high possible quality and when you give him advice he just dismisses it and best of all he's using Handbrake to so his encodes with.

Just LOL.

[_Al_]

Yes, if you want to push boundaries and pressing compression efficiency even further and not to see quality drop much, it has to be done in fazes. Unless you hack x264 yourself. You encode to just highest possible quantizer for most scenes. Then watch the video again, and make a timetable on a piece of paper what scenes need to be fixed and mark also degree of a problem. Then Encode video again using zones, where you increase bitrate (lower quantizer or multiply bitrate). Or encode it separatelly and use stichable. Or devide movie into clips/parts (this particular thing is where I do not understand why it did not pick up yet in general, technology is here)

When an engineers builds a complex machine, they do not design it from the scratch and let it "print" again. They find a problem and try to localize it and fix it. Different part of an object needs to be reinforced or changed. Unless the whole design is fundamentally wrong. But we know that x264 "machine" is one of the best things out there. So fix the broken parts , not designing it over and over again. Yes, it takes time. Doh. If you do not have that time settle with compromise and you will NOT find visually that "better" yet compression for a video.

[_Al_]

Classic example with visual feedback like that is in movie Ex Machina. That movie would return very nice result for certain CRF. Except those three or two scenes where there is no power. Those scenes would look like crap. So one can go and lower quantizer overall and look if that scene looks ok. But you also increase overall bitrate. Instead you just encode those scenes with different settings. That movie is an extreme and beautiful example for "lab experiments/ school experiments". Most movies are not like that but anyway.

We can spend a days to encode videos like that, but it is not worthy, especially encoding movies, movies that have milions of copies somewhere, not unique videos. And unique videos on the other hand, you do not care about bitrate. So how to encode with best settings is basically pointless. You just settle with a preset that overall returns ok quality and go on, use batch etc.