/mpv/ - the /g/reatest media player

*mpv.net

>>62333224
Haasn is not a .net retard but an haskell one!

>>62333192
Hi, mpv people!

In windows there is a player called splash pro ex. It somehow "adds" more framerate artificially to the video, I don't know how

Does MPV have a similar feature to that?
https://www.youtube.com/watch?v=ur8D-kWvsss

>>62333246
Who? Does he work on mpv.net?

>>62333281
you mean interpolation, in which case, yes.

video-sync=display-resample
interpolation
tscale=oversample

>>62333298
Okay, I got this now

x11-bypass-compositor=no
profile=opengl-hq
scale=ewa_lanczossharp
cscale=ewa_lanczossoft
dscale=mitchell
scale-antiring=0.7
cscale-antiring=0.7
dither-depth=auto
correct-downscaling=yes
sigmoid-upscaling=yes
deband=no
volume-max=100
hwdec=auto

volume=35

video-sync=display-resample
interpolation
tscale=oversample

>>62333335

dscale=mitchell
scale-antiring=0.7
cscale-antiring=0.7
dither-depth=auto
correct-downscaling=yes
sigmoid-upscaling=yes

delete this

who the fuck cares about mpv.net? it has no features or improvements over the default cplayer and uses the shitty broken wid embed mechanism.

>>62333335
antiringing is currently disabled for ewa scalers

>>62333350
Done

>>62333385
Good. They're already on

profile=opengl-hq

and >>62333371

>>62333413
Is there any config files to have these maximum tweaks on so that I can download this into the scripts directory and forget about it?

>>62333462
shiggy diggy

>>62333281

vo=opengl-hq:scale=ginseng:cscale=ginseng:swapinterval=1:temporal-dither:dither-depth=auto:fbo-format=rgba16f:scaler-resizes-only:sigmoid-upscaling:pbo:interpolation

>>62333512
upgrade your mpv to something from this year fucking hell

>>62333512
Hello 2015.

>>62333281
mpv does not have a feature like that built in but mpv can be configured to use SVP as input. SVP implements that feature as well as black bar lighting.

Here's video.
https://www.youtube.com/watch?v=Wjb6CSe4708

You can read more about it here:
https://wiki.archlinux.org/index.php/Mpv#SVP_4_Linux_.28SmoothVideoProject.29

Note, creating a profile is not the best way to use it and the mpv arch wiki page currently contains a lot of errors (the profile section in particular is wrong).

There's a lot of confusion in terminology since a lot of it gets used for different things in different circles. The sort of interpolation that Splash Pro's Motion2 and SVP implement is called "motion interpolation". The techniques they use however tend to produce a lot of bizarre artifacts that look like garbled limbs wobbly movements. For instance, in the youtube video you posted pay ver close attention to the human during the 0:19-0:21 period. For SVP you can notice the artifacts in the video I linked by looking closely at the credits text as the video plays. That isn't to say that good motion interpolation is impossible, but it is currently probably a few years away from being practical. Here is an example of a cutting edge technique using deep neural nets on high end graphics hardware that runs very far from real time.
http://www.sniklaus.com/publications/sepconv

Whether or not SVP is a good idea to use may depend on the content you are viewing (eg. action vs drama).

The normal behavior of video players is called a 3:2 pulldown that uses frame repetition to go from a lower framerate to your display's refresh rate (eg. 60hz).

>>62333298

mpv's "smoothmotion" technique is different from motion interpolation. Instead it is an improvement over 3:2 pulldown that interpolates some frames (actual frame itself, not motion within the frame) in order for the repeated frames to be played a more even number of times (eg. 2.5:2.5 pulldown).

>>62333649
>SVP is now closed source
dropped

>>62333712
https://www.svp-team.com/wiki/Manual:SVPflow

>>62333712
SVP has always been closed source

>>62334519
It seems pretty easy to decompile but who can like that soap opera effect!

>mpdn was/is a true shit
>madvr destroyed mpdn
>mpv destroyed madvr

>>62333649
>The sort of interpolation that Splash Pro's Motion2 and SVP implement is called "motion interpolation".
How do you know? I checked their website and found no explanation of what they are actually doing.

Also keep in mind that SVP is a fork of MVTools and while it's slower, it's definitely usable.
Could also be they are using something like http://bluesky23.yukishigure.com/en/BlueskyFRC.html (that is, AMD's FluidMotion) because their website only lists very specific conversations (24,30 -> 60) which is also the case for this AMD shit while SVP/MVTools can do arbitrary conversions.

Damn why is nnedi3 still that slow with madvr?
And damn why has he still not optimized his ngu-aa for animes to compete with ravu?

>>62334669
Isn't madVR way older than MPDN??

>>62334669
>mpff will destroy mpv

>>62334737
Maybe but madvr totally destroyed mpdn, but mostly because his dev gave up. Now mpv beats everything, ewa scalers are 4x faster than ones from mpdn and that's truly awesome for an opengl based program!

>>62334716
There are 50 billion motion interpolation techniques and on average, not a single one of them is going to be identical - or perhaps even comparable - to any other one.

What MVTools and SVP are doing is based on block-by-block motion vector analysis. They use a block search algorithm (similar to x264 etc.) to find similar blocks in both frames in order to determine the local motion vector for that block. This motion vector is then interpolated per-pixel while also accounting for stuff like optical masking and stuff; and then the pixels are reflowed along the interpolated motion vectors to form the intermediate image.

SVP also implements some heuristics to turn off the pixel-based interpolation for blocks that are too dissimilar; which you can tune via the settings. (Relaxing this heuristic makes the interpolation more aggressive but results in way more artifacts). Not sure if SVP does this but it would also be possible in principle to aggregate the per-block motion into a global motion vector to detect pans and only smooth out those; without affecting “action” too much.

>>62334785
Wait and see...

File: theblocks.jpg (2MB, 2000x2000px) Image search: [Google]

2MB, 2000x2000px

>>62334801
Natively inside mpv when?

>>62334790
> mostly because his dev gave up
Sad to hear, their devs did more good than madvr ever did. Their shaders are open source, bjin and igv based their work on theirs and further optimized algorithms for much better performance

>>62334844
I think .net was a bad idea...

>>62334801
Did you reply to the wrong person?
Or maybe I worded that bad. I wasn't asking what SVP is doing, I know that already (well, not really but I do know MVTools and just assume they didn't change too much). I wondered how you can tell that this mentioned media player is doing what SVP is doing. The spout a few fancy sounding buzz words but never mention what they actually do.
For all we know, they could do some cheap stuff like mpv's interpolation and use incorrect terminology for marketing purposes.

>>62334840
Would be easier if there were a decent API for MVTools. Haasn declines to touch it without one. But he also doesn't want to make one himself so...

>>62334844
Shiandow should give up too or join the mpv commiters.

>>62334844
Good? The whole project was cancer.
MadVR at least had the decency to have it all proprietary. The MPDN guy tried to have a proprietary player and at the same time leech work from the open source community without giving anything back.
Yeah, no thanks.

>>62334883
>Decent C API.
Is it even exist?

>>62334840
Whenever somebody gets bored enough to implement it, I guess. Block motion search wouldn't actually be too difficult, the difficult part would probably be pixel re-flowing while accounting for occlusion.

There was also this recent discussion about a paper that implements motion interpolation by using a convolutional neural network to generate a convolution mask directly, rather than using it to try and detect optical flow / motion vectors. This allows it to innately learn how to deal with masking/occlusion, blurry regions, artifact avoidance, etc - while also being easier to train blindly on regular video files rather than specially labelled optical flow training sets. The downside of this technique is that it would be pretty slow to implement without some major breakthrough; although the separable convolution approach might have some merits here.

>>62334868
>Did you reply to the wrong person?
I was replying to the discussion as a whole. My point was that just because two things both advertise “motion interpolation” doesn't mean they're even remotely doing the same thing.

File: 40202657.png (6KB, 210x229px)

6KB, 210x229px

>>62334930
Doable in one week by motivated haasn! ;)

>>62334941
And I'm claiming they might not even do motion interpolation, so once again I ask you: how do you know?

>>62335034
Know what??

>>62334906
yes, much more good. mpdn ported nnedi3 from opencl to shaders, implemented ssim shaders and gave them back. madshi leeched much more foss and did bare minimum to keep uphold licenses.

>>62335124
Are you a retard?

>decide to look through the mpv source code
>look for main function to start reading the code
>see this:
https://pastebin.com/raw/bYCVbDFd

what the FUCK guys, why are there so many?

which one is the correct main function???

best shaders for chinese cartoons pls?
I like em sharp and edgy, like my personality

File: 1391230395037.jpg (524KB, 1920x1200px) Image search: [Google]

524KB, 1920x1200px

What's the decoder+vo framedrop mode for? It only says it's not recommended on the manual. Is it simply less efficient than decoder or vo mode alone?

>>62335584
osdep/main-fn-unix.c, osdep/main-fn-win.c and osdep/win32-console-wrapper.c for windows, and osdep/main-fn-cocoa.c and osdep/macosx_application.m for mac os
why different ones for windows and mac os? the source code explains better than i can here; windows does funny stuff with regards to the differences between console and gui applications, and mac os needs the NSApplication singleton to use cocoa features.
the real cross platform entry point is mpv_main() in player/main.c

>>62335589
FSRCNN.

>>62335773
further clarification: mpv uses cocoa to get access to mac os features, and cocoa requires the start of execution to be on the main thread, and on windows, main-fn-win.c fixes stdout, stdin and stderr, fixes some windows nonsense (loading DLLs from %PATH%, heap corruption checking, removing some error boxes) and converts windows' UTF-16 command line to UTF-8.

>>62335773
>>62335840
I see, thank you for the explanation.

>>62335824
>FSRCNN
>Prescaler based on layered convolutional networks. Pretty slow, so mostly suitable for still images rather than realtime video playback.
anon why

>>62336273
anime watchers in this thread have vouched for it in this thread for realtime playback before

I have 750 Ti, anything you suggest adding/changing or is this fine?

profile=opengl-hq
scale=ewa_lanczossharp
cscale-ewa_lanczossharp
video-sync=display-resample
interpolation
tscale=oversample

>>62336616
>

cscale-ewa_lanczossharp

profile=opengl-hq
opengl-backend=dxinterop
opengl-pbo=yes
opengl-early-flush=no
scale=haasnsoft
scale-clamp=0.5
cscale=ewa_lanczossharp
sigmoid-slope=10.0
vd-lavc-threads=16
vd-lavc-dr=yes
opengl-shader=~~/shaders/ravu-r4-smoothtest1.hook

r8

>>62336616

cscale=bilinear

Hey this is pretty autistic and retarded, but I messily articulated why trying to loop mpv by default was giving me grey hairs >>>/qa/1566059. A few threads ago someone helped me get looping by default, so when the topic came up >>>/qa/1564836 I shared my awkward experience. Not trying to personal army request or anything but I figured those more tech aware can cringe at how retarded I was I guess.

>>62336748
interpolation

video-sync=display-resample
tscale=robidouxsharp
interpolation=yes
deband-iterations=2
deband-range=12
temporal-dither=yes
blend-subtitles=no

>>62336616
>cscale-ewa_lanczossharp
typo'd, but don't use ewa_lanczossharp for cscale, just increases ringing and looks worse than spline36 or haasnsoft

>>62336748
>vd-lavc-threads
>opengl-early-flush
unnecessary
>scale-clamp
>sigmoid-slope
placebo
>cscale=ewa_lanczossharp
use bilinear
>scale=haasnsoft
what the fuck
>>62336773
>tscale=robidouxsharp
>deband-iterations
>deband-range
>blend-subtitles=no
why

opengl-backend=dxinterop
scale=ewa_hanning
scale-radius=3.2383154841662362
scale-blur=1.055
cscale=haasnsoft
dscale=mitchell
correct-downscaling=yes
sigmoid-upscaling=yes
sigmoid-slope=10
deband=yes
deband-grain=32
deband-iterations=2
deband-range=12
deband-threshold=48
dither-depth=auto
temporal-dither=yes
video-sync=display-resample
vd-lavc-dr=yes
opengl-shader="~~/shaders/ravu-r4-smoothtest1.hook"

r8

>>62336812
>vd-lavc-threads
actually helps
>placebo
that's the point
>use bilinear
only when the source video is 1080p
>what the fuck
looks better than lanczos
>tscale=robidouxsharp
oversample is too juddery, mitchell is too blurry
>blend-subtitles=no
because i don't want my subs being affected by interpolation or other filters there's literally no reason for it.

watch out nerds

profile=opengl-hq
opengl-backend=dxinterop
scale=ewa_lanczossharp
cscale=bilinear
video-sync=display-resample
interpolation
tscale=oversample
screenshot-directory='~/Desktop'
msg-color
msg-module
no-osc
no-taskbar-progress
quiet
hr-seek=yes

>>62336273
Thats old version.

Here you go anon.
https://github.com/igv/FSRCNN-TensorFlow/releases
They are very fast.

>>62336877
RAVU > FSRCNN

>>62336861
>oversample is too juddery
Video with oversample cant have any judder. Unless its in the source of course. Yours is just more blurry therefore looks smoother.

>>62336877
Is it faster than NGU Sharp (Low)?

>>62336921
>Video with oversample cant have any judder
oversample doesn't mask the imperfect interpolation so you can see it skip or 'judder'

>>62336927
Yes or on par.
>>62336906
Theyre very different. Both great at what they do.

>>62336955
>Both great at what they do.
can you elaborate on this? they are both attempting to do the same thing so comparing them should be valid

>>62336616
tscale=mitchell :)

Madshi seems pretty quiet lately.
Is he concocting something crazy for us?

>>62337076
fuck off kike

>>62337076
Does he usually interact with people?
I don't follow madvr development.

Why so many video player devs are from Germany?

>>62337115
autismus

>>62334716
>How do you know? I checked their website and found no explanation of what they are actually doing.

You are right to doubt since their website is vague as fuck. The main way I "knew" was by looking at forum posts comparing it with other approaches and watching video samples. Which is to say I don't know for sure. As another anon pointed out there are a lot of approaches to this (many newer TVs even have one built in that can be enabled in the settings).

Also, I'd been away from this thread all this time. There's no need to be rude to other people as in: >>62335520 >>62335124 >>62335034 >>62334868

>>62334930
>>62334801
My understanding is that SVP is based on mvtools2 (or at least it was in earlier versions) so I'm not really sure that there's much to be gained from native mpv support since it produces quite a lot of artifacts as is. I am looking forward to newer techniques but as you said it's probably not gonna happen any time soon.

Personally I like the black bar lighting that SVP does far more than the motion interpolation. Now that is something I would like to see getting native support in mpv.

>>62336921
>>62336861
>>62336945
>>62336980

There's some interesting discussion on tscale settings here:
https://github.com/mpv-player/mpv/issues/2685

ffmpeg's new motion interpolation filter "minterpolate" isn't half bad.
https://www.youtube.com/watch?v=aT-oXygGf24

>>62337885
eta till we have the hardware to do it in realtime

>>62337885
just tried it and it was running with like 0.5 fps

>>62337885
Add that to mpv pls!

>>62337995
How do i enable it?

>>62338028
>>62338046
mpv --vf=lavfi=[minterpolate=fps=60000/1001:mi_mode=mci]

>>62337885
Looks pretty good. Way better than mpv native.

>>62338104
Dumbass.

>>62337885
Looks like arse on anime.
>>62338104
mpv doesn't do motion interpolation, only frames blending.

>>62337995
Why is it so slow? I have very low frametimes and CPU load yet it stutters!

>>62338313
Because it's not made for realtime.

>>62338355
Can i convert my videos through mpv with it?

>>62337995
It has no SIMD; completely unoptimized

>>62338383
Probably, but it'd be easier to use ffmpeg directly. It's going to take forever though.

>>62338355
Can i force big cache, play video at 0.5 fps and then rewind back and replay at full speed since it got cached?

>>62338429
yes if you wait a few hours until the cache is filled...

>>62338429
Even if you could (which I'm not sure) it wouldn't be worth it, be reasonable here.

>>62338453
forget what i just said, it wont work at all

>>62337885
how new is it

>>62338514
Not very, some guy was mentioning it a year ago
https://github.com/mpv-player/mpv/issues/2685#issuecomment-250985264

The main thing is, that mpv should implement a motion interpolation algorithm

haasn pls notice

>>62336906
RAVU = ewa_lanczos, retard

>>62338429
mpv doesn't cache decoded frames

Weird question here.

I'm using xrandr --scale to set my resolution at approximately 1080p on a 720p display. This means that if I download 720p media and play it at full screen then it gets upscaled to approximately 1080p by mpv and then downscaled back to 720p by my display. This is a waste of resources and it sounds to me like the only way to avoid this is to manually set xrandr back to normal when playback starts then upscale again once playback ends.

Is there a more sane way to handle this? Are there other potential drawbacks I'm missing by doing this?

>>62336877
Is igv the current fork to keep an eye on? Is it a good idea to add this to the Arch wiki? What sort of information should I include? Does it only kick in when video is being upscaled or is it always running (I haven't played around with this because upscaling isn't really useful on my shitty display unless I'm watching 480p video or something)?

>>62338943
>Is igv the current fork to keep an eye on?
Yeah.
>Is it a good idea to add this to the Arch wiki?
No, i dont think so. It will only confuse people. Maybe you probably should only include basic shaders like SSIM and SSSR since they will work automatically when needed and provide very visible improvement. IMO you need just to mention the ability to use shaders and link user scripts page from the OP.

>>62339128
>IMO you need just to mention the ability to use shaders and link user scripts page from the OP.
That's more or less what I did, but I had some doubts about that.
https://wiki.archlinux.org/index.php/Mpv#Some_simple_high_performance_configurations

Disclaimer: There's still a lot of things that need to be fixed on that page and my explanation of profiles is actually wrong (I've just been trying to decide how to fix it without making profiles sound impractical).

Me again, >>62339175

I know it reads badly as it's currently written. I do plan on editing it/rewriting it. I just wanted to redo other portions of the page first.

>>62338943
>Does it only kick in when video is being upscaled
yes

>>62339175
>The first option makes it so that if audio and video go out of sync then instead of dropping video frames it will resample the audio (a slight change in audio pitch is often less noticeable than dropped frames).
I think this is wrong. Doesn't it speed up/slow down the video to make it a multiple of your monitors refresh rate? It still drops video frames if you drop frames.
>It may be useful to hide the GUI window for video files. This can be done by using --no-video option.
This has nothing to do with the gui and plays back the file without the video track.

Also I think it would be useful to mention auto profiles like:

[extension.webm]
loop-file=yes

anyone have an example where SSimSuperRes looks good? it makes everything look oil paintingish

File: facepalm-.jpg (11KB, 280x210px)

11KB, 280x210px

>>62339395

>>62339360
>Doesn't it speed up/slow down the video to make it a multiple of your monitors refresh rate?
It does that too, but that's just an added benefit. The description on the arch wiki page is still the primary effect.

>>62339395
Just dont use it on low res videos.

>>62339175
>is not likely to make a significant difference in performance.
This is not necessarily true. For H.264 and VP9 on a typical desktop CPU, sure, hwdec is unlikely to help. But hwdec can make a substantial difference for slow CPUs or hard-to-decode content (e.g. HEVC).

>>62339243
Thanks!

>>62339360
>I think this is wrong.
I'm just going by what it says here:
https://github.com/mpv-player/mpv/wiki/Display-synchronization#display-sync
and here:
https://github.com/mpv-player/mpv/wiki/Interpolation#smoothmotion

It is definitely possible that I've misunderstood things.

>Also I think it would be useful to mention auto profiles like:
Yeah! I hadn't realized that this was something one could do without auto-profiles.lua (haven't played with this yet) until yesterday and I planned search the archives for some good example applications. Thank you for that one!

I saw the [protocol.http] stuff was useful for youtube-dl too (I planned to consolidate/rewrite the current stuff into a new section as well once I got some time to play with it some more).

>>62339594
I know 10-bit h264 isn't supported by hwdec (and isn't likely to be supported) but is HEVC supported? I'm not trying to be a dick or anything, I honestly don't know. Any suggestions on what that section should say instead (the original version of the page recommended hardware decoding and it looks like previous editors thought no hwdec meant no hardware acceleration, the gentoo wiki mpv page also says something similar)?

By the way, I'm just some anon who really likes mpv and decided to update the Arch wiki because lots of people go there for information (it also gave me bad advice and it took /g/ to set me straight).

>>62339175
>Volume is too low
The quoted “reasonable amount” volume-max=600 is absolutely insane. That's basically a 50 dB amplification (linear gain 216x) and will clip absolutely every file in existence or your money back. It's like setting VLC to 22,000% volume.

A much more reasonable value would be volume-max=130. Also, volume-max is *never* the right way to fix “volume too low”. I would expand this section and also add af=loudnorm, af=drc, af=acompressor, af=dynaudnorm etc. examples as those are all much better ways to fix too-quiet audio.

>>62339175
Something that would actually be useful to mention is suggesting users try out vd-lavc-dr and opengl-pbo, both of which can significantly improve uploading performance.

>>62339661
>I know 10-bit h264 isn't supported by hwdec (and isn't likely to be supported) but is HEVC supported?
Depends on your GPU. Newer GPUs generally have HEVC support, yes.

File: mpv-shot0001.png (2MB, 1902x1080px) Image search: [Google]

2MB, 1902x1080px

>>62339395
This looks like oil painting? Are you retarded?

File: mpv-shot0002.png (2MB, 1902x1080px) Image search: [Google]

2MB, 1902x1080px

>>62339788
Compared to this (without)

>>62339788
>artifact enhancer
now try RAVU

>>62338813
>RAVU = ewa_lanczos
??

>>62339671
Yea, that looked totally retarded to me too. I didn't write this section but it is an example of the terribad advice I'm trying to remove.

Unfortunately I don't know enough about audio filters to expand on that myself right now (I'd have to spend some time reading/testing). One concern I have is that some may require a recompiled (or maybe some aur) version of ffmpeg (some like loudnorm don't appear in mpv -af=help with default ffmpeg on my setup)? I also remember mpv having some built-in options to change how it does things when downscaling audio (i.e. normalizing first to avoid clipping) but I don't know much.

>>62339693
Right! I actually posted asking about the current state of both of those a few days ago while trying to decide whether or not to add them (or if I do add them whether or not I should advice people to use the mpv-git aur package instead of the official release).

I currently use direct rendering (vd-lavc-dr) and it works great.

For the last few days I've been looking through git issue pages and stuff for information. Here's one on opengl-pbo asking for more testing (seems it's possible that in some cases it may reduce performance, maybe).
https://github.com/mpv-player/mpv/issues/3858

There's also this pull request relevant to all the Wayland stuff that never got merged for some reason.
https://github.com/mpv-player/mpv/pull/4384

File: mpv-shot0003.png (2MB, 1902x1080px) Image search: [Google]

2MB, 1902x1080px

>>62339854
RAVU

>>62339976
This looks like oil painting in some places + errors

>>62340028
Bullshit.

>>62339921
loudnorm requires ffmpeg built with libebur128

These are the options I use/recommend:

af-add=@audnorm:!loudnorm=I=-25:TP=-1.5:LRA=1
af-add=@dynnorm:!dynaudnorm=f=200:g=5:r=0.1

The syntax here adds them with a given label and makes them off by default. You can enable them with keys like this:

n af toggle @audnorm
N af toggle @dynnorm

(Not strictly part of what I was suggesting, but an interesting example of “advanced” mpv usage nonetheless)

File: thank you all for helping this retard.jpg (28KB, 640x480px) Image search: [Google]

28KB, 640x480px

>>62339726
>>62339594
So maybe I should add a disclaimer that says something like.
>On systems with a slow CPU but fast GPU hardware decoding may make a substantial performance difference depending on the content being decoded. In such cases it may be beneficial to enable it in general and disable it on a case by case basis for files where bugs are encountered.
>Note: Ensure you do not have vo=xv in your configuration since it is not compatible with hardware decoding.

I should also add a message to the vo=xv section (Fix jerky playback and tearing) pointing out that that VO isn't compatible with hardware acceleration (I have no idea how accurate the rest of the information in that section is).

Since I have so many people's attention. I'd like to ask if anyone uses mpv on a laptop with both an integrated and discrete video card. Should mpv be run in some special way on those (like Nvidia + bumblebee people would maybe use optirun or primusrun)?

>>62340091
Yeh, RAVU is bullshit, it's good at removing aliasing tho.

>>62339921
>For the last few days I've been looking through git issue pages and stuff for information. Here's one on opengl-pbo asking for more testing (seems it's possible that in some cases it may reduce performance, maybe).
The only issue was some OS X shit and given how fucking broken that platform is (and the EDIT: by the author) I don't think it's that relevant.

It would strongly surprise me if opengl-pbo ever reduced performance on any platform that doesn't pereptually have about 20 active bugs.

>>62340130
GPU performance doesn't make any difference for hwdec. They use dedicated silicon decoder chips. The GPU itself is not involved in the hardware decoding at all. So the “fast GPU” makes no sense. It only depends on whether or not your CPU is a potato.

>>62340178
Okay, I'll test it on my system and add it.

>>62340286
Of course, that makes sense.

File: Capture.png (2KB, 304x58px)

2KB, 304x58px

Do I need to change something else?

>>62340112
What does the @ syntax do? It looks like it's naming the setting. Is that just a filter thing?

Thank you for sharing your settings. I'll read more on this and add something (unless you want to do it, which you are more than welcome to!).

>>62340431
AFAIK Adaptive is good enough.

>>62340431
>not maximum performance

File: 012311.png (600KB, 800x533px) Image search: [Google]

600KB, 800x533px

>>62340587

>>62340689
I dont even go above 30% usage. I average around 20%.

>>62340462
>https://mpv.io/manual/master/#video-filters-vf
>Before the filter name, a label can be specified with @name:, where name is an arbitrary user-given name, which identifies the filter. This is only needed if you want to toggle the filter at runtime.

I want mpv to start playing at 0 zoom without scaling down to fit my screen, overscan and all.

How do?

>>62342272
--video-unscaled?

>>62342436
Thanks. I was skimming everywhere and CtrlFing "zoom" in the manual because "video-zoom" ("video-scale"?).

>>62342769
When you change the resolution of the video, you either upscale or downscale it anon. It's the most common term.

>>62342801
I know but my fail was a combination of brainfart and incohesive term usage on mpv IMO. Perhaps instead of video-zoom it should be video-scale (the manual says "Adjust the video display scale factor.") (there's osd-scale=<factor>).

Also the manual says "video-zoom=0 is unscaled". And that's assuming the video isn't downscaled to fit the window/screen which would be the case by default. Threw me off.

>>62343322
I think --video-zoom makes more sense than --video-scale when you consider that the zoom is relative to the unzoomed video size; not the unscaled video size. (Since video is normally stretched to the window size. --video-zoom just adds an additonal zoom on top of that)

Is there a way to record/capture certain parts of a video on mpv? Would like to make some streamables. I don't think you can do it on MPC either, only VLC has this feature as far as I know.

>>62343797
see: one of the 50 webm scripts?

>>62343797
There are scripts that allow you to transcode or losslessly remux part of the video and there's the built in

record-file

option.

>>62343532
I see. If video-zoom is always relative to the window size then maybe there should also be video-scale relative to the native video size.

Also... I think Alt1 Alt2 Alt3 Alt4 should be video-scale factors.

>>62343848
I read that on the manual it's very experimental and can fuck up. Do I just put record-file in mpv config with the name of the video? It doesn't specify how to record certain segments too e.g. 02:00 to 02:30. I wish there was just a keybinding for this where I could just simply losslessly record. Don't want to use VLC

>>62343904
Seems reasonable. I can imagine generalizing --video-unscaled to --video-scale and making it that --video-scale=0 behaves the same as --video-unscaled=no (i.e. “automatic” scale); and --video-scale=1 behaves the same way as --video-unscaled=yes (i.e. 1:1 scale)

How can I get my media keys to work? NEXT and PREV don't seem to do the trick.

Do subtitles generally work fine out of the box? Do you guys generally set any subtitle options in your config besides slang=?

Noticed some iOS related commits, is libmpv iOS port used in any app yet?

>>62345322
Just font and size

>Interpolation
>MVTools
>SVP

You guys need to realize that those are just dirty fixes and the only way to get a truly smooth experience is to get a display refresh rate that is divisible by the framerate of whatever you are watching.

144 / 24 = 6
Problem fixed forever.

>>62346380
>madvr custom resolutions
>72hz for oldish monitors
>144hz for newest ones
>perfect playback
>bitstreaming friendly
>no waste in smoothmotion/interpolation shits
>full power for neural based algorithms

>>62346380
But what do you do for videos that are 23.98?

>>62346666
Underclock your display to 143.88Hz

File: haasnismadshi.png (203KB, 460x460px) Image search: [Google]

203KB, 460x460px

>madshi is inactive
>haasn is inactive

>https://github.com/rossy/mpv/commits/ra-d3d
What is the rossy's motivation?
The vo_vulkan is mostly finished and will work with windows too. No?

>>62347272
Will it work with d3d9 only graphic cards?
If so, madshi is on suicide watch!

>>62347292
madshi can just copy mpv's code since it is open source

he's playing 4D chess and making other people his hounds

Which one for 720p animes? I have a shitty GPU.
- FSRCNN_4-0-4_x2.glsl
- ravu-lite-r4.hook

>>62347494
profile=opengl-hq

>>62347494
ravu-lite-r2.hook

>>62347126
>both from Germany
>both visit this thread

>>62347553
I thought haasn is from netherlands or something

>>62347494
What GPU and what your monitor resolution?

>>62347561
His github profile says Germany.

>>62347568
Shitty iGPU with 1080p monitor!

>>62347592
How old is your CPU? Try profile=opengl-hq first. Even it might be too much for iGPU...

>>62347592
>>62347616
Set deband=no as well

>>62347710
B-but I can use debanding + lanczos3ar upscaling + smoothmotion with madVR!

File: ra-d3d11.png (2MB, 2507x916px) Image search: [Google]

2MB, 2507x916px

>>62347272
>>62347292
Yeah, if vo_d3d11 and vo_vulkan are ever released, they should be mostly redundant with each other on modern hardware. vo_vulkan already seems to perform well on Windows. vo_d3d11 will have a few advantages though. It should work on older hardware. Dumb mode works on feature level 9_1 and --profile=opengl-hq appears to work on 10_0 (ANGLE requires 9_3 and 10_1 respectively.) Also, integrating 10-bit hardware decoding with D3D11VA will be trivial in a D3D11 renderer, and it might be easier to hack in modern DXGI features, like HDR output.

The obvious disadvantage of vo_d3d11 is that it requires translating SPIR-V to HLSL, and SPIRV-Cross' HLSL support is pretty immature at the moment. Basic shaders work and they appear to compile down to pretty efficient HLSL bytecode. The EWA scalers don't work at all because SPIRV-Cross hasn't implemented the image load/store or textureGather intrinsics.

File: 1495985476506.png (3MB, 1411x794px) Image search: [Google]

3MB, 1411x794px

>>62348524
How'd I get the graphs on Windows?

>>62348653
shift + i

>>62348638
Can't promise anything yet bby ;) Shader support is still a big blocker because glslang (and SPIRV-Cross) are difficult to integrate in a C program and not very mature yet.

>>62348653
Install this:
https://github.com/Argon-/mpv-stats

File: 216px-MPlayer.svg.png (13KB, 216x185px)

13KB, 216x185px

Why does mplayer offer better playback on older systems than mpv? Even with vd-lavc-skiploopfilter mpv still drops more frames. I'm using the xv vo on both.

File: 1489554487745.png (943KB, 975x594px) Image search: [Google]

943KB, 975x594px

>>62348678
>>62348707
Thanks; I'm already that far but there is no graphical information, like the one I replied to.

>>62348737

plot_perfdata = true,
    plot_vsync_ratio = true,
    plot_vsync_jitter = true,

In the script.

>>62348707
Will it make interpolation as efficient as smoothmotion on madVR?

File: 1502327128667.png (3KB, 223x72px)

3KB, 223x72px

>>62348813
Fug, I don't know what I'm doing wrong. Tried a different series even and it's not showing up either.

>>62348857
mpv version?

>>62348871
mpv-x86_64-20170718, directly from lachs0r.

>>62348895
Mouahahahaha!

>>62348925
;_;

>>62348895
Something on your end, they're working fine with this build here too.

>>62348822
I don't know, but it would be possible to make tscale=oversample more efficient with D3D11 or Vulkan, because DXGI and Vulkan swapchains make it pretty easy to wait for vblank without rendering an additional frame, so mpv wouldn't have to redraw the intermediate frames.

>>62348992
>it would be possible to make tscale=oversample more efficient
madVR tier efficient? Ill remind you that madVR's smoothmotion doesnt give a fuck about v-sync intervals.

>>62349110
What does it mean?

File: this - Copy.jpg (73KB, 257x546px) Image search: [Google]

73KB, 257x546px

>>62349131
>Finally, --video-sync=display-* currently comes with one important drawback: Due to OpenGL's rather severe limitations when it comes to timing, the only way to reliably figure out when vsyncs happen is to actually draw a frame on every vsync. The consequence of this is that, even for 24 Hz video, you need to draw frames at 60 Hz even if they are the same frame over and over again - thus increasing power usage by a factor of 2x-3x in such a case.

No shit like that. Rendering time can be higher than vsync. i.e above 16K for 60Hz display. The only limiting factor is video's frametime.

>>62349247
Yes, so it should be fixed with the d3d renderer.

>>62349281
>should be fixed with the d3d renderer
Same with Vulkan? Then i can finally use mpv again on Linux.

>http://bugs.madshi.net/view_all_bug_page.php
>all this red
madshi can't fix madvr bugs anymore

>>62347434
>madshi can just copy mpv's code since it is open source
That's not how it works

>>62349380
Not that anon but you cant even check his source code since its protected.

>>62349366
He's busy working on vo=vulkan for mpv.

>>62349380
He can copy parts of it and no one would notice, or look at it and find out how vulkan stuff is done instead of trying to work it out himself.

Why do you think his code is closed source? Makes it easier to steal code.

>>62349247
That quote is no longer accurate. mpv already caches redraws from FBO. But that mechanism could be made a tiny bit more efficient with vulkan, yes - you could directly re-queue the rendered image instead of using a cache FBO. (Although that wouldn't let you update the OSD in-place)

More importantly, with vulkan you get to control the swapchain depth (e.g. --vulkan-swapchain-depth=2). In theory, with a swapchain depth of 2, you can render each frame for 30ms and it will just deplete the swapchain buffer to compensate for the increased duration, without resulting in a dropped frame. (However, it's possible that mpv will still measure a “mistimed” or “delayed” frame. But that stat would be a bogus measurement. Ideally the way mpv measures mistimed/delayed frames needs to be reworked for the considerations or dynamic swapchains)

Why doesn't mpv support animated WebP?

Works: https://www.gstatic.com/webp/gallery3/3_webp_ll.webp
Doesn't work: https://storage.googleapis.com/downloads.webmproject.org/webp/images/dancing_banana2.lossless.webp

Both work in google chrome.

>>62349593
>[ffmpeg/video] webp: skipping unsupported chunk: ANIM
gee what the fuck do you think this means

(also that second one doesn't work in my browser either)

>>62349646
I know it's not supported I'm asking why.

As for why it doesn't play in your browser:
https://developers.google.com/speed/webp/faq#why_should_i_use_animated_webp
>Animated WebP rendering requires Google Chrome version 32+

File: 1495616364428.gif (413KB, 640x633px) Image search: [Google]

413KB, 640x633px

how well would a gtx 460 and i5 2500k hold up with the good scalers and configs with mpv????

>>62349507
>He can copy parts of it and no one would notice
What parts would he copy and how would he integrate it?

- madVR is written in delphi and C++, mpv is written in C.
- madVR shaders are written in HLSL and OpenCL, mpv's are written in GLSL
- madVR uses static pre-compiled shaders, mpv uses dynamically generated shaders
- madVR uses fixed rendering paths, mpv builds the pipeline incrementally
- all mpv vulkan/opengl code sits behind the `RA` abstraction, which is specifically designed to meet the needs of mpv's dynamic rendering system
- lots of mpv's vulkan code uses fixed assumptions based on mpv's needs/requirements
- mpv's vulkan code is (currently) X11/Nvidia-only

Dissecting, understanding and reusing portions of mpv's code is likely going to be more difficult than implementing it from scratch. The hardest part about vulkan is understanding resource protection, cache invalidation and memory coherency - pipeline barriers, semaphores, fences, queue family transfer operations, etc. The second hardest part is memory/resource management and pooling. Without a solid understanding of that the mpv code is useless because adapting it to madVR would require understanding how it fits together. There's no “free lunch” here.

(cont)

>>62349774
>Why do you think his code is closed source?
Because it's shitty delphi? Why do you think most code is closed source? So the authors don't have to care about the quality of their code, since it won't be publicly displayed anyway. There's probably as much closed-source freeware as there is actual paid commercial software. Closed source code is a coping mechanism for people who can't handle criticism but don't want to feel inadequate when releasing their code.

Also, if he copies anything major enough to be obvious (e.g. the entire shader_cache or filter_kernels mechanism), he would have to go to at least as much effort to hide the “signatures” from it as just writing it from scratch. Plus, he doesn't seem to be interested in doing that sort of thing anyway. The code has been out there for years, yet he still doesn't see an interest in copying mpv's shaders or rendering mechanisms, even when those are either faster or more flexible than his. (Hell, he doesn't even seem to have an interest in optimizing his own kernels based on mpv's findings)

>>62349443
And? Code leaves traces. Wouldn't it be weird for a new madVR version to suddenly have support for customizable filter kernels, and then for that to mysteriously have the same tunable parameters as mpv's filter kernel generation? Wouldn't it be weird for a new madVR version to suddenly start supporting fixing prescaling shift, but then to have the same bugs as mpv's gl_transform mechanism? Wouldn't it be weird for madVR to suddenly start supporting vulkan, but the call order and usage patterns being exactly the same as mpv's (including bugs and exact parameters)?

>>62349774
>>62349819
Only madshi himself would know this much about both mpv and madVR.

If anything I suspect you even more now.

>>62349855
This. Very suspicious...

how can mpv even compete
http://screenshotcomparison.com/comparison/117039

>>62350389
looks like scale=sinc ngl

>>62350389
1. try using an mpv version from this decade
2. try adding RAVU or FSRCNN

>>62348735
Where does `perf top` say the time is being spent? Is it obviously different for mpv and MPlayer? (Are you sure mpv is actually using the `xv` VO, and not falling back to vo=opengl?)

File: 55062635.jpg (31KB, 499x451px) Image search: [Google]

31KB, 499x451px

>use madVR
>turn smoothmotion ON (only when motion judder without it)
>make sure that rendering time is lower than frame time
>done

>use mpv
>turn on video-sync=display resample, interpolation, itscale=oversample (always ON, even for videos with FPS even multiple of your display Hz for some reason. Off only for videos whose FPS and display Hz matches.)
>have to calculate your v-sync interval (1000/ display Hz) and have your rendering time to be below it. Have to decrease your settings accordingly because v-sync interval is always much lower than frame interval (i.e 40ms for 24FPS video in madVR and 16ms for 60Hz in mpv).


Something needs to be done...

>>62350784
Then do something about it? The code is open source, you're welcome to improve it

File: i.png (2MB, 1920x1080px) Image search: [Google]

2MB, 1920x1080px

Cool.

File: 2017-09-10_19-16-40.png (1MB, 1600x900px) Image search: [Google]

1MB, 1600x900px

>>62351072
My mpv looks worse.

But maybe your dragonball ep is higher quality.

>>62351072
What's with those color differences? Different source or?

>>62351072
> he fell for ewa_lanczossharp meme

File: mplayer-mpv.png (34KB, 884x533px) Image search: [Google]

34KB, 884x533px

>>62350466

It says [xv] in square brackets like that, that means it's using xv right? I'm not sure how to read the perf top output, so I took these screenshots.

interpolation is a meme, same tier as sigmoid-slope=10

>>62351304
recompile mpv, mplayer and ffmpeg with debugging symbols..

>>62346380
MVTools and SVP actually create new frames where pixels are moved around in an attempt to guess at what an intermediate frame should look like.

That's completely different from what interpolation does and what you are describing.

>>62351280
Dunno but I like it.

>>62348185
lanczos3 isn't that just sinc?

opengl-hq has a lot of other high quality settings. If you just want those things then just enable those things (and don't use opengl-hq).

deband=yes
scale=sinc
interpolation
tscale=oversample

this isn't even scratching the surface of what mpv can do

>>62351653
lanczos3 is sinc-windowed sinc (i.e. scale=sinc scale-window=sinc scale-radius=3)

>>62349717
Probably fine. You might have trouble running meme shaders though.

>>62351072
This is the worst comparison ive ever seen.

>>62351697
>

{{"lanczos", 3, sinc, .resizable = true}, .window = "sinc"},

So just scale=lanczos

How to set scale to bicubic? What bicubic scalers are available within mpv?

>>62351910

» mpv --scale=
Available scalers:
    bilinear
    bicubic_fast
    oversample
    spline16
    spline36
    spline64
    sinc
    lanczos
    ginseng
    jinc
    ewa_lanczos
    ewa_hanning
    ewa_ginseng
    ewa_lanczossharp
    ewa_lanczossoft
    haasnsoft
    bicubic
    bcspline
    catmull_rom
    mitchell
    robidoux
    robidouxsharp
    ewa_robidoux
    ewa_robidouxsharp
    box
    nearest
    triangle
    gaussian

>>62351910

 // Cubic filters
    {{"bicubic",        2,   bicubic}},
    {{"bcspline",       2,   cubic_bc, .params = {0.5, 0.5} }},
    {{"catmull_rom",    2,   cubic_bc, .params = {0.0, 0.5} }},
    {{"mitchell",       2,   cubic_bc, .params = {1.0/3.0, 1.0/3.0} }},
    {{"robidoux",       2,   cubic_bc, .params = {12 / (19 + 9 * M_SQRT2),
                                                  113 / (58 + 216 * M_SQRT2)} }},
    {{"robidouxsharp",  2,   cubic_bc, .params = {6 / (13 + 7 * M_SQRT2),
                                                  7 / (2 + 12 * M_SQRT2)} }},
    {{"ewa_robidoux",   2,   cubic_bc, .params = {12 / (19 + 9 * M_SQRT2),
                                                  113 / (58 + 216 * M_SQRT2)}},
            .polar = true},
    {{"ewa_robidouxsharp", 2,cubic_bc, .params = {6 / (13 + 7 * M_SQRT2),
                                                  7 / (2 + 12 * M_SQRT2)}},
.polar = true},

>>62351881
Exactly

>>62349717
Should be plenty for 1080p unless you want to use the latest meme neural network filters

Reminder that scale=ewa_hanning scale-radius=3.2383154841662362 is better than lanczos/sharp/soft.

>>62352163
How so?

>>62352173
Scientifically proven

Why isn't mov on f-droid?

>>62352173
Faster and more quality, test it.

>>62352196
It's slower though

>>62352188
>>62352196
its just a haasnsoft without the blur option

Is it possible to install MPV on OpenSUSE/Fedora without resorting to a 3rd party repo?

File: on.png (1MB, 1920x1080px) Image search: [Google]

1MB, 1920x1080px

This icc-profile thing is just a meme too?

>>62352347
Wat. My mpv looks much better with the use of icc compared to madvr.

File: meme.png (43KB, 904x187px) Image search: [Google]

43KB, 904x187px

>>62352347
Blame haasn and his copy pasting drones.

File: minterpolate.png (96KB, 1901x484px) Image search: [Google]

96KB, 1901x484px

I will try to post results

>>62352163
i'm sure if i do a screenshot comparison and squint with my eyes an inch away from the screen i'll be able to tell the difference.

Is debanding actually worth it if I'm not watching anime?

>>62352347

[vo/opengl] ICC profile detected contrast very high (>100000), falling back to contrast 1000 for sanity. Set the icc-contrast option to silence this warning.

>>62352376
Better than on.

File: 1504539018049.jpg (147KB, 590x700px) Image search: [Google]

147KB, 590x700px

>>62351794
>>62352128
>>62351834
arigato
will try it out next week then

>>62352655
Depends on if your anime has banding (eg. simulcasts). The degradation isn't really noticeable but it's kind of a waste of gpu resources.

>>62352808
Youre welcome qt

File: on.png (2MB, 1920x1080px) Image search: [Google]

2MB, 1920x1080px

>>62352655
It helps with lower quality encodes that have compression artifacts. Look at the right side of the image. This is default deband settings and opengl-hq.

>>62352655
If you watch low quality streams (anime or not), then hell yes!

File: off.png (1MB, 1920x1080px) Image search: [Google]

1MB, 1920x1080px

>>62352857

>>62352511
>some people want oversaturated garbage
wew, just turn up --contrast and --saturation

>>62352874
>>62352857
These look the same on my tv.

>>62352874
>>62352857
Any difference between these is totally lost on me. I guess with my screen/eyeballs it doesn't matter.

>>62352857
>>62352874
What a terrible example.

>>62352347
You realize ICC profiles are unique for every monitor, right?

If your monitor is different from the one who generated the profile, the image with the ICC profile applied will look wrong.

>>62352655
If you use mpv to watch youtube videos, streams or literally anything with a low bitrate then fuck yes

Just watch an INSIDE speedrun with debanding off vs on :D

Vulkan when?

File: 1487026334609.jpg (132KB, 800x484px) Image search: [Google]

132KB, 800x484px

>>62353109
Saturday

>>62353121
mpv isn't even GPL

>>62353043
I got it from the manufacturer.

File: 1489767782152.gif (260KB, 709x738px) Image search: [Google]

260KB, 709x738px

>>62353005
>>62353019
>>62353021

holy shit you people spend hours jerking off over the minute differences of your meme shader configs as if setting sigmoid slope makes the slightest bit of difference and you can't even fucking tell when debanding is on or off? you're a living meme.

>>62353271
they're using 6-bit TN laptop LCDs

>>62353121
You tricked me once already...

>>62353271
think u might be retarded

>>62353109
October.

>>62353136
parts of mpv are

>>62353271
Don't rope me into this senpai, I'm the guy that asked if it actually matters. It seemed to me like it would mostly effect animations that have smooth gradients. In normal conditions I would never notice what your gif is demonstrating.

File: knife goes in eye.png (7KB, 389x255px)

7KB, 389x255px

>>62353071
>videophile spends time carefully configuring display settings and fine tuning media player configuration to eliminate framedrops, judder, and other shit.
>decides to watch low bitrate encodes of foggy content

Motherfucker, I understand the people who download 720p anime and upscale it using meme shaders (since most anime is drawn at 720p and upscaled by the studio, sometimes badly) but what the fuck is wrong with you? Unless that shit is livestreaming and the streamer is a retard and/or your internet is shit (and the streamer is multiplexing the stream) then you have no excuse.

In other words, yes, debanding is useful if you are watching content that's so poorly encoded that it introduces banding. Especially foggy content at low bitrate.

Pic related, you.

>>62353271
>>62353389
Or just anyone watching on an HDTV with Vivid settings (inaccurate colors/contrast/brightness but punchier image).

P.S. Accuracy is considered more of a personal taste thing like >>62352511 points out (unless you're a graphic designer/artist and you're using the display for work, of course).

>>62353550
There's a huge git issue where they're planning to move away from GPL to LGPL.

>>62353720
>punchier image
pic related?

>P.S. Accuracy is considered more of a personal taste thing
Maybe in america; here in germany there's a definiton of “accurate” that doesn't include “subjective”

Any anons have experience using this on an Optimus laptop? I've been using madvr forever but can't for the life of me get my new laptop to use it's 1070 for my Chinese cartoons. Looking to make the leap into mpv if it plays nice with Optimus.

>>62353720
m8 even well-encoded 8-bit video from good masters can suffer from banding. And all consumer video aside from some 4K streams is 8-bit.

>>62353720
>These may not be accurate as such, but it's more about getting an image which looks and feels more attractive for those uses.
>These may not be accurate
>may not be
learn2read

>>62353664
Sometimes it matters, sometimes it doesn't. If it bothers you to leave it on all the time, set a keybind and turn it on or off at your own discretion.

>>62352655
>>62353071
>>62353720
>Especially foggy content at low bitrate.

Link related. Open this in your browser and notice the obvious banding and artifacts all over the video. Debanding will help with this.
https://www.youtube.com/watch?v=5k17aVeY2rI

It might also be relevant for WebRips, livestreaming, shitty low bitrate encodes, and simulcasts (because CR puts out shitty low bitrate encodes, also their 1080p are just upscaled 720p).

However if you're torrenting high quality encodes made from high quality sources by expert encoders then it will likely just be a waste of resources.

>>62353815
I didn't mean that the definition of accuracy is subjective or up to personal taste. I meant that whether you want your display to be configured to be accurate or not is up to personal taste. Many people like to tweak their settings based on the ambient lighting and just general personal taste.

>>62353918
That was exactly my point, though perhaps my wording was unclear. Vivid settings are not accurate though people sometimes prefer them.

http://www.avsforum.com/forum/139-display-calibration/1188981-vivid-mode-bad.html#post17380771

>>62353856
Do you have some examples?

Just so you know, many high quality encoders out there will actually implement debanding and stuff as part of their process depending on the content (anti-aliasing and stuff as well for downscaling).

Does deband performance depend on screen resolution and video resolution or only video resolution?

MPV drops more frames when I connect my laptop to the TV while playing 720p content opposed to when I play it on my low res laptop screen.

>>62353844
OS?

>>62353815
>>punchier image
>pic related?
http://lifehacker.com/5066858/punch-up-a-photo-in-under-60-seconds

>>62353767
is lgpl the jewish one that lets companies steal your code and put it in botnet software like plex?

>>62353856
>well-encoded
>8-bit
pick one

also note that there's no such thing as a good source for stuff like anime; the blu-rays themselves are shit out of the box

>>62354028
>However if you're torrenting high quality encodes made from high quality sources by expert encoders
Like YIFY and kuchikirukia?

>>62354134
No, like Moozzi2.

>>62354039
>Does deband performance depend on screen resolution and video resolution or only video resolution?
Only video resolution. You can see the time it took for the deband pass by installing the mpv-stats script and pressing shift+i

File: 100.jpg (89KB, 640x480px) Image search: [Google]

89KB, 640x480px

>>62354134
>Like YIFY and kuchikirukia?
pic related

but seriously, I wish someone compiled a list somewhere. Especially with anime it's so hit and miss. PTP has a list of good encoders but I think PTP is more concerned with bit-starvation vs bloat since they only deal with movies.

File: 1504973416994.jpg (902KB, 1131x1600px) Image search: [Google]

902KB, 1131x1600px

>>62333192

too much autism in these threads recently


i want memeVR fags to go back

what shaders are good for low-res video?

>>62355375
scale=nearest

>>62355375
Use an oil painting shader and add grain?

>>62355375
nnedi3, RAVU.

If only majority of these shaders and scalers you guys care about wasn't only for cartoons alas madvr
>>62348653
>DBNL
Bad release
>>62337885
>this looks fine..
>action scenes come up
>>62354149
He makes them look better with filters
>>62354028
>However if you're torrenting high quality encodes made from high quality sources by expert encoders then it will likely just be a waste of resources.
no

>>62356038
>If only majority of these shaders and scalers you guys care about wasn't only for cartoons alas madvr
wat

>>62356102
If only majority of these shaders and scalers you guys cared about wasn't only for cartoons alas madvr

what use is the opengl-swapinterval option?
Does increasing it to say 3 give any kind of performance advantage?

>>62357699
No. Set it to 0 or 1

File: mpv bilinear.png (219KB, 1366x768px) Image search: [Google]

219KB, 1366x768px

What can I do to get better numbers here? This is my config rn

profile=opengl-hq
scale=bilinear
deband=no
video-sync=display-resample

, anything more I can do for performance? I'm on a laptop with integrated graphics btw.

Here's the test file https://imouto.my/download/9th-test-video-angel-beats-trailer/ for reference

>>62358167
What OS are you on?

>>62358202
windows 7

>>62358238
opengl-backend=dxinterop or win

>>62357699
>>62357903
Actually I wonder what would happen if you set it to swapinterval=2 for 30 Hz content; but that would probably not be any different from just swapinterval=1

>>62358167
Here's the opengl-hq profile so you can see what's in it. Personally I would set cscale=bilinear instead of scale (scale carries most of the details so it's more noticeable).
https://github.com/mpv-player/mpv/blob/master/etc/builtin.conf

>>62358167
Show second stats

Also it's possible that the stats for ANGLE are just plain wrong; especially the “upload frame” stat

File: mpv dxinterop.png (199KB, 1366x768px) Image search: [Google]

199KB, 1366x768px

>>62358261
Can't really tell a difference except that the lua script isn't showing timings anymore

File: mpv bilinear2.png (207KB, 1366x768px) Image search: [Google]

207KB, 1366x768px

>>62358293
here's without opengl-backend=dxinterop

File: control panel.png (61KB, 954x780px) Image search: [Google]

61KB, 954x780px

>>62358358
How about the win backend? I have an igpu and get the best performance with win.
Pic might help too.

>>62358555
(He isn't using any compute shaders)

File: mpv win.png (298KB, 1366x768px) Image search: [Google]

298KB, 1366x768px

>>62358555
well, the frame timings dropped by 1/3rd, so win backend certainly helps (although screenshots look fucky now but w/e)
the mistimed/delayed count seems to have gone up a lot though, is there a way to get that under control too?

>>62358279
cscale=bilinear also seems to have helped, thanks for the tip

>>62358636
Try also vd-lavc-dr (if VRAM isn't an issue) or opengl-pbo

Also try cscale=bicubic_fast instead of bilinear. It should also be very fast.

>>62358393
Also it seems like the video file you're playing is higher res than your display. In that case getting a lower res release would help a lot, since currently mpv is upscaling only to then downscale.

I have an experimental patch that would make mpv directly scale to the chroma res in such cases (instead of doing upscale->downscale), but I'm not happy with it. (Trashes scaler config, and doesn't benefit from such options like linear downscaling)

Anyone know what I'm doing wrong to get RAVU working on Linux? Was working fine on Windows 10, but since I moved to Linux it doesn't seem to work anymore. Last thing I need to get working to finish configuring mpv. I think it might be the backend and needing to use a different one for Linux.

opengl-backend=dxinterop
opengl-shaders="~~/shaders/ravu-r4-smoothtest1.hook"

>>62358750
>directx
>linux
????

>>62358750
>opengl-backend=dxinterop
>on linux
nigga what are you doing

>>62358773
Still new and learning for Linux, but yeah I assumed dxinterop wasn't going to work. Tried looking at manual to pick a different one, but had no idea which one to use for Linux systems. Tried removing the backend altogether, and it didnt work either. Decided I asked here if anybody else knew anything.

>>62358800
Remove opengl-backend=dxinterop. Unless you're on Wayland then you don't need to set that option on Linux.

Do the following:
>Run mpv from the command line to check for error messages
>make sure that that file is located in ./config/mpv/shaders/
>enable a log file and check it for messages
>post the rest of your config.

>>62358980
Here is my mpv.conf:
https://pastebin.com/GGqAQpf0

Log.txt is giving me the same error I used to get in Windows 10 before using dxinterop:
0.240][e][vo/opengl] Unrecognized command 'TEXTURE ravu_lut_smoothtest1_4'!

>>62359519
What is your video card and drivers? Are you running a setup with discrete + integrated video card?

>>62359519
>0.240][e][vo/opengl] Unrecognized command 'TEXTURE ravu_lut_smoothtest1_4'!
hurr, update mpv

>>62359519
>the same error I used to get in Windows 10 before using dxinterop:
That has LITERALLY nothing to do with it

>>62354394
AHD and HDB encodes are generally considered the best
PTP has a list on the Wiki, x264 encodes are good too.

>>62354394
Also this exists
pastebin.com/sRpVCjCa