/mpv/ - The Open Source and Cross Platform Video Player

Last thread Installation:
mpv.io/installation/

Wiki:
github.com/mpv-player/mpv/wiki

Manual:
mpv.io/manual/stable/

User Scripts(including opengl shaders):
github.com/mpv-player/mpv/wiki/User-Scripts

Migrating from MPC-HC?
github.com/dragons4life/MPC-HC-config-for-MPV/blob/master/input.conf

input.conf:
github.com/mpv-player/mpv/blob/master/etc/input.conf


Vulkan(Linux only for now):
github.com/atomnuker/mpv

Test vulkan and post logs if it gives you any kind of problems.

For better playback quality paste this in your mpv.conf file:
profile=opengl-hq
cscale=ewa_lanczos
scale=ewa_lanczossharp


Check your settings for compatibility errors by running mpv in command line or with

log-file=log.txt

. Search the log for "dumb" anything or for [e].

REMINDER →

Other urls found in this thread:

github.com/mpv-android/mpv-android
diff.pics/g813R3To2Mn5/1
imagemagick.org/discourse-server/viewtopic.php?t=15742
stackoverflow.com/questions/384991/what-is-the-best-image-downscaling-algorithm-quality-wise
mpv.io/manual/master/#options-save-position-on-quit
0x0.st/92
0x0.st/9_
apt.izzysoft.de/fdroid/index/apk/is.xyz.mpv
diff.pics/S8efhYJdVivg/1
en.wikipedia.org/wiki/Squash_and_stretch
animationsmears.tumblr.com/
youtube.com/watch?v=haa7n3UGyDc
github.com/nagadomi/waifu2x/issues/92
mpv.srsfckn.biz/stats.html
github.com/rossy/mpv-install/archive/master.zip
twitter.com/NSFWRedditVideo

Bump

Since when there's an android version, and why isn't it on f-droid?

this media player is the ultimate mega garbage.
fuck you for this piece of shit.

Compile and upload it yourself?
Or create an issue:
github.com/mpv-android/mpv-android

Linked mkv when?

I think subtitles should automatically go there unless you have blend-subtitles enabled.

I'll check it out, thanks.

Are the numbers normal?
specs= i5 4690k @4.5GHz + GTX1070 G1

are you using swdec?

swdec?

software decoding
what is your config?

#Video#
profile=opengl-hq
scale=ewa_lanczossharp
cscale=ewa_lanczos
dscale=mitchell
correct-downscaling=yes
dither-depth=auto

#Deband
deband=yes
deband-iterations=2
deband-range=12
deband-threshold=48

#Interpolation
interpolation=yes
video-sync=display-resample
tscale=oversample

#Other
opengl-backend=dxinterop
hwdec=no
vd-lavc-threads=4

#upscale
sigmoid-upscaling=yes
sigmoid-slope=10.0
opengl-shaders="~~/shaders/SSimSuperRes.glsl"

Try hwdec=yes or hwdec=auto-copy

Red frames on scale=ewa_lanczossharp both of them

Try opengl-backend=angle

Meant for

Red frame goes over 60k total and also get Mistimed.

Erase
#Interpolation
interpolation=yes
video-sync=display-resample
tscale=oversample

Yes, but I wanted interpolation on.

p.s. still red frame on scale w/e interpolation :(

(Obviously take note that styled subtitles, ie. ASS will always be within the video bounds, even if it's a plain line)

Bump

Good night bump

Current state of FSRCNN
diff.pics/g813R3To2Mn5/1

Why does it seem everywhere I look, people say mitchell is good at upscaling, and lanczos is better for downscaling, while the mpv manual states mitchell is best for downscaling and ewa_lanczossharp is the best for upscaling.

imagemagick.org/discourse-server/viewtopic.php?t=15742
(ImageMagick defaults to mitchell for upscaling and lanczos for downscaling)

stackoverflow.com/questions/384991/what-is-the-best-image-downscaling-algorithm-quality-wise
The top answer claims a survey of professional cinematographers claims that mitchell was the top rated for upscaling and "sinc" (assumed to be lanczos) was rated the best for downscaling.

What about compute shaders v2 I need that shit in my life

report this issue and hope dev could give an explanation. Or admit they messed things up

tech illiterate here, i'm on mac os and really happy with mpv since it handles really high res video way better than vlc and any other program i've tried. my question is how can i get it to resume videos from the last played position? google around a bit but no luck so far...

mpv.io/manual/master/#options-save-position-on-quit

1. When you implement it yourself.
2. When you find a example of actual Matroska hard segment linking (NextUID/PrevUID) in the wild. (This is not the same thing as ordered chapters.)

It seems like that line about mitchell was originally suggested to be put in by lachs0r 0x0.st/92

When it was made the default for opengl-hq, the commit note says "several people agree that it's a good filter for downscaling", despite what general opinion seems to be
0x0.st/9_

Guess we would need an mpv dev or somebody who knows better to tell which one is correct.

also, Sup Forums's spam filter is fucking horrible, so short links.

thanks bud, i'm retarded. i kept looking for a command or something lol. sometimes thinks are easier than you think

>diff.pics/g813R3To2Mn5/1
looks reasonably good, but it's somewhat hard to judge ringing artifacts with this image. which fsrcnn kernel configuration is this?

There's an option to force ass into black bars as well. It's just image based subs (DVD/BD) where it's not possible.

2x ratio, 56,12,4, MS-SSIM (I fixed it), 120 epochs on expanded data (took 4 hours to train).

Wait, lanczos is not ewa_lanczos! Completely different.

Normal lanczos is even worse than spline36, which is used in opengl-hq. Ewa is better than both by far.

so the "official" fssrcn kernel? the "big"/slow one? is there any hope to run that in real time? thx.

Add apt.izzysoft.de/fdroid/index/apk/is.xyz.mpv to your Fdroid repos.

With compute shaders.

Why is it showing my scale bilinear eventhough it is set to ewa_lanczossharp?

profile=opengl-hq
opengl-backend=dxinterop
#hwdec=auto-copy

video-sync=display-resample
interpolation=yes
tscale=oversample
tscale-clamp=yes

vf=format=gamma=gamma2.2
opengl-gamma=1.1218765935747068

scale=ewa_lanczossharp
cscale=ewa_lanczossharp
dscale=ewa_lanczossharp

volume-max=200

screenshot-format=jpg
screenshot-jpeg-quality=100
screenshot-template="%F-%p"

Probably because the source is the same size as your monitor (1920x1080), so the scaler is disabled (see --scaler-resizes-only.)

Re-asking
>What about direct rendering?
>And what about people with old GPUs (no compute shaders)? Will we use functionality...?

What's your monitor resolution and what's the source resolution?

So is ewa_lanczos better than mitchell when it comes to downscaling?
What about mitchell+SSimDownscaler.glsl vs ewa_lanczos?

>What about mitchell+SSimDownscaler.glsl vs ewa_lanczos?
diff.pics/S8efhYJdVivg/1

>.jpg

>idiot

My monitor and the source is 1080p I thought it was supposed to work all the time. Thanks.

>Comparing quality with a lossy format
Use png next time.

I keep seeing stuff about using neural nets trained on imagenet data but I never hear about anyone training on a data full of anime images in particular. Has no one tried it?

isn't that what waifu2x did?

For people that ask which downscaler is better. This image was made on avsforums by some experts and approved by madshi few years ago. As far as i remember people always suggested to use catmull_rom (bicubic 50) for downscaling (in madVR). Its sharper than mitchell but is somewhat less natural and has more aliasing and doesnt hide source artifacts. In theory mitchell is better for downscaling. Plus SSIM is designed by igv to work with mitchell.

Another "guide" created by lots of experts and also approved by madshi in madVR thread on doom9 few years ago.

Man, it really isn't a madVR algorithm unless it has a meaningless acronym.

There is also this image about dxva scalers used by different GPU manufactures.

Why does it say scale=billinear even though I have set scale=ewa_lanczossharp?

Not quite, waifu2x has a few different models. The ones corresponding to anime aren't anime itself but rather "anime-style artwork". This includes stuff like fan-art, manga covers and promo-art, doujin stuff, etc...

Real anime (and 2D animation in general) has all sorts of things that never show up in artwork like squash and stretch:
en.wikipedia.org/wiki/Squash_and_stretch
animationsmears.tumblr.com/
youtube.com/watch?v=haa7n3UGyDc

tl;dr: waiffu2x is trained on anime-style art which is analogous to comics. This is different from training it on cartoon still-frames.

More info on waiffu2x's models:
github.com/nagadomi/waifu2x/issues/92

See

*
#Video
profile=opengl-hq
hwdec=d3d11va-copy
opengl-backend=angle
opengl-fbo-format=auto

# Video profiles #
[4K]
profile-desc=cond:get('width', -math.huge) >= 3840 and get('container-fps', 0) >= 59
interpolation=no

[Default]
profile-desc=cond:get('width', math.huge) < 3840
opengl-shader="~~/shaders/FineSharp.glsl"
video-sync=display-resample
interpolation
scale=ewa_lanczossharp
dscale=lanczos
cscale=ewa_lanczossoft
sigmoid-upscaling=yes
deband-grain=0
deband-iterations=2
deband-range=12
deband-threshold=48
tscale=oversample
scale-blur=0

Dat intel. Its not possible to use dxva as upscaler in mpv, right? Could be cool for people with older intel iGPU's.

Got a question about my config (see )

Does it make sense to use lanczos on 4k videos if I have a 1080p screen and like sharp images or will the image get "too sharp"?

Literally erase your whole disgusting config and use
profile=opengl-hq
cscale=ewa_lanczos
scale=ewa_lanczossharp

video-sync=display-resample
interpolation
tscale=oversample
You can also add SSSR shader.

Why though? By the way I didnt just copy&paste it. It was a long time ago and I might have made some mistakes but I remember that I had to get rid of the smooth motion which triggered me.

>but I remember that I had to get rid of the smooth motion which triggered me
But you have tscale=oversample ???

Bro, are you watching cartoons?

ain't you too old to be on that disney?

Nice squat, good form.

Ass to grass.

Okay, seems you're right. Got two more questions:

What speaks against
hwdec=d3d11va-copy
opengl-backend=angle

It helps with performance and if I understood correctly, there shouldnt be a quality loss if I use d3d11va-COPY

Second question: I don't know much about shaders but why SSSR over FineSharp?

What CPU do you have?

Hey, it's, "Linked mkv support, industry standard right-click context menu, volume control, non-retarded keybinds, ability to properly count, and Doom9 unbanning when?"

Xeon 1230v2

HOLY SHIT. I just noticed that downloads on shinchiro's mpv build skyrocketed after mpc-hc death! Amazing.

lachs0r's also did, check it out:
mpv.srsfckn.biz/stats.html

You dont need any hardware acceleration. It might only be necessary for some super heavy 4K test video. Hardware acceleration often fucks shit up. If you can its better avoid it.
>I don't know much about shaders but why SSSR over FineSharp?
FineSharp is better used with higher quality sources or after doublers (like nnedi3) only because it sharpens artifact. SSSR is less picky about source and in theory provides image close to the ground truth original. Plus SSSR has built in antiringing.

How the sweet fuck to I get vp9 hw acceleration on Windows?

I have 1080p vp9 60fps webm video from youtube, if I play it in Chrome or on youtube it is hw accelerated but if I open in in mpv it isn't.

>Hardware acceleration often fucks shit up. If you can its better avoid it.
I keep hearing this meme and I think it's just the same user posting it over and over again.

It may be anecdotal but I've never had any problems with hardware acceleration. I mean it's possible that on some hardware configurations, on some operating systems, with some drivers, some users may experience some problems on some files.

As far as quality degradation, I think that depends on what is being decoded and with what hardware. From looking at the github issue tracker a bit it looks like it's limited mostly to corner cases and unsupported stuff that should no longer even be possible to trigger.

You have a shit CPU? Put hwdec=auto into config.

>Computer with Wangblows
>Download any mpv version
>Right click video and try to set mpv as default program
>After selecting mpv in its directory just reprompts for default program
Why?

>You have a shit CPU?
No but it's a matter of principle, I should be able to use hwdec when the machine I spent good money on does it well.

>Put hwdec=auto into config.
Doesn't do fuck all unfortunately, I've tried auto, dxva* (all crap), d3d11va (works best for me with everything but vp9) as well as various backends win (works but no improvement), dxinterop (completely not working) and cuda (only copy works and also no vp9 hwdec).

There was some batch script that I used that set mpv as the default program and everything. No idea where I got that though.

>I keep hearing this meme and I think it's just the same user posting it over and over again.
Its not a meme. Its been known for ages. Its always suggested (even by madVRfags on doom9) to be turned off if possible.
Use shinchiro's build and run install.bat. Right click and choosing mpv as preferred program works for me though.

github.com/rossy/mpv-install/archive/master.zip

Thank god I haven't used Wangblows in years. Linux software is soooo much easier to use on Linux. Everything just werks.

>madVR
>doom9
Aren't those just wangblows fags who still live in the codec pack era of video playback?

Yeah they had the script right next to all the mpv builds, thanks. Still don't know why mpv's the only player I've tried that does that though, it's not even applicable through the control panel (on this computer at least)

Why did you put the name in forward-slashes?

I'm using a nvidia graphic card, should I use dxva2-copy or cuda-copy?

Because that's how Sup Forums works you dipshit.

I think FineSharp wins the placebo battle

Cuda-copy has best performance on my 970

I haven't updated mpv in just about a year (v0.20.0)
What should I expect? Were there many major changes? Will all my configs break?

Try using that script that gives you performance stats and see which one performs better.

>Hardware acceleration often fucks shit up. If you can its better avoid it.
I have just been trying out different shaders to find out which one suits me the best and hardware acceleration adds some artifacts to the image lol. Maybe I can record it. Seems like you're right

Cuda seems way faster under windows: 3.8 ms to render a frame with cuda and more than 10 with dxva2.

That probably has less to do with windows and more to do with your GPU and video card drivers.

That's because with cuda your GPU works at highest frequency always.

What type of file are you playing? What is your setup and your mpv.conf? Also, do you get this with all supported hardware decoders including copy-back decoders?

Wtf are those artifacts. First I tried disabling hardware acceleration, which didn't work. But if I remove profile=opengl-hq only, those artifacts do not appear.

profile=opengl-hq
hwdec=dxva2-copy
opengl-backend=dxinterop
opengl-fbo-format=auto
opengl-shader="~~/shaders/FineSharp.glsl"
video-sync=display-resample
interpolation
scale=ewa_lanczossharp
dscale=lanczos
cscale=ewa_lanczossoft
tscale=oversample

Right, I thought gpu load was way higher using dxva2. But as it uses a really low clock speed I guess it's more power saving.

see Yes, I tried switching to dxva2 with dxinterop etc.. but those artifacts only disappeared when I if I remove profile=opengl-hq