Thread 106025382

FAQs

Q: What are you talking about?
A: A few days ago we figured out how to store images inside VP9 Webms and trick video decoders into decoding images instead of videos.

Q: What's so special about a hardware accelerated images?
A: Normal images are processed by a CPU but these are processed by your GPU, which is more powerful and more efficient.

Q: Can't you just use JPG?
A: Yeah but AFAIK it's not hardware accelerated. Also JPG images tend to look like dogshit at low file sizes.

Q: Are there any limitations?
A: Resolution is caped at 4096x2048 but 4chan caps it even further to 2048x2048 for whatever retarded reason.

for %f in (*.png) do ffmpeg -i "%f" -quality 100 "%~nf.webp" & ffmpeg -framerate 1 -stream_loop 1 -i "%~nf.webp" -c:v libvpx-vp9 -quality best -crf 10 -qmin 10 -qmax 10 "%~nf.webm"

Also I'll be adopting a tripcode because my ego is bigger than my cock and I'll make an effort to either clean up existing JPGs or ensure my source PNGs are not created from JPGs.

>his CPU cant process images
Grim

>>106025382 (OP)
SEXO WITH ANIME FEMALE TEENS AND GIVE NAKAD4SHIS AND PREGNANCY, SPECIALLY THE YURI ONES, FIX THEM WITH NAKAD4SHIS

>he needs a gpu for a static image
okay

>retards still haven't realized that decoding an image in software is faster than creating a hardware surface
typical /g/eets lol

>tracking webm
fuck

>>106025382 (OP)
bloat files instead
waste data and compute

>>106025865
trippy

>>106025957
bro you need a better phone. There is only one keyframe at the very start but it shouldn't corrupt like that.

This file should be 3.77 MB, 120000 fps but 4chan truncates it to 2 MB. At least it did so the last time I posted it.

>>106025464
>>106025570
>>106025677
I still think it's pretty cool that you have the option to decode images via a GPU instead of the traditional CPU. Even if you had say a high end CPU and a low end GPU there's still the efficiency aspect which I find alluring.

>>106026671
nigga, you need a media player to display an image
you need literally play the single frame instead just display it once

>>106026723
Ye, having to open them in video players isn't ideal but at least 4chan auto-hides the player controls. Maybe this bullshit thing might get popular enough to get the attention of video player devs, i dunno.

>>106025382 (OP)
repeat after me
STILL IMAGES DON'T NEED HARDWARE ACCELERATION

>>106027120
rather, a doctor will pay attention to your schizophrenia

>>106025382 (OP)
>A few days ago we figured out how to store images inside VP9 Webms and trick video decoders into decoding images instead of videos.
wow what an incredible discovery... ffmpeg -i input.jpeg output.webm, how did you figure that one out!!!?
...and it also works for h264, literally nothing special about this, has been possible forever
in fact there is NOTHING special in the command you shared, all of that stuff is already the default that ffmpeg applied when transcoding images to video, what the fuck are you on about

>>106025865
nigger

>>106027277
No they don't especially with high end CPUs but it's still a nice thing to have if you want it.

>>106027314
>ffmpeg -i input.jpeg output.webm
Nope, for some reason it must be converted to a webp image first for this to work and then you have to stream loop twice the output webp and a bunch of other bullshit. IF you want the VP9 hardware decoder to actually do its thing else your CPU gets raped. Also H264 isn't as efficient as VP9.

On a related note should I bother hardcoding the required encoding parameters to make these picture webms? I mean it's in my OP and maybe I'm wasting my time...

>>106027507
>it's still a nice thing to have
no it's not, the overhead that comes from initializing the hardware decoding and streaming the data to the hwdec is already more expensive than just doing it on the cpu like computers have been able to do since the 80's, it's an overcomplication that doesn't make any sense, plus it locks you into 4:2:0 which isn't the standard for image formats like it is for video, not to mention all sorts of incompatibilities and other shit (half the replies in this thread are complaints that your shit doesn't work in case you didn't notice)
>Nope, for some reason it must be converted to a webp image first for this to work and then you have to stream loop twice the output webp and a bunch of other bullshit.
...no, it just works on my machine, no need to transcode to webp, what is most likely happening is that you are a complete total fucking retard and really cannot just -pix_fmt yuv420p your shit and are letting lossy webp handle it for you
>else your CPU gets raped
oh yeah i'm sure my cpu is getting raped from decoding a single 1920x1080 vp9 keyframe... sure, whatever you say daiz.
>Also H264 isn't as efficient as VP9.
wow the retard that is transcoding his images into lossy webp to then transcode them to lossy vp9 is talking about efficiency here... interesting.
>On a related note should I bother hardcoding the required encoding parameters to make these picture webms?
no, you should just take some meds like you've been advised hundreds of time for the past few years, or kill yourself.

>>106027810
>what is most likely happening is that you are a complete total fucking retard and really cannot just -pix_fmt yuv420p your shit and are letting lossy webp handle it for you
Interesting, can you elaborate?

>>106025382 (OP)
>implying my system supports hardware video decoding

>>106027857
what the fuck is there to elaborate?
what you are saying here is that you literally don't even know what chroma subsampling (the very first thing a lossy image/video codec does) is... that's concerning.
hardware decoders only support 4:2:0 (4:2:2 is now a thing in some newer nvidia/intel gpus but that's beside the point, and usually only for video production work with h264/h265 rather than vp9), what is happening is that you are most likely transcoding some 4:4:4 images to vp9, 4:4:4 vp9 isn't supported by any hardware decoder, that's why it doesn't work
just subsample the chroma to 4:2:0 and you're done, no need to do the same thing by first transcoding to lossy webp, the fact that I even have to explain this is completely absurd.
you shoud stop trying to emulate a schizo faggot (daiz) on topics you are clueless about, what do you get from this? actually deranged behaviour, kys.

>>106025382 (OP)
>A: Normal images are processed by a CPU but these are processed by your GPU, which is more powerful and more efficient.
This literally doesn't fucking matter. You're a schizo, take your meds.

>>106027967
it does, there isn't a single computer built in the last 20 years that doesn't support h264 hardware decoding.
you are just running your average dogshit troonix distro that has basic functionality missing out of the box.

>>106028006
first thing a normal person does on loonix is install non-free codecs

>>106028054
correct, and this retard did not do that.

>>106027977
Is fun, that's all. Chroma subsampling being a requirement for most hardware decoders does suck, can't deny that. Although if the source if a video then does it really matter?

>>106028006
>>106028077
I have all the codecs, it's just that firefox refuses to use them
it's not a problem, even 4k videos play well

>>106025382 (OP)
>>106028001
gpu acceleration can end up using more power thats the funny thing. all it does is kill your phone/laptop faster.

>>106028109
i bet they fucked up something in their "utility process" sandboxing properties or w/e somewhere along the way. you can probably toggle something off in about:config

>>106028101
>Chroma subsampling being a requirement for most hardware decoders does suck
it doesn't, there's no use case outside of professional production work for higher chroma than 4:2:0 when it comes to video, and almost no codec supports higher than that in its main profile, it only sucks if you are a complete fucking schizoid like you are.
>Although if the source if a video then does it really matter?
of course it doesn't, unless your video is for some reason more than 4:2:0, but you are discussing images here
>>106028174
>gpu acceleration can end up using more power thats the funny thing
for still images and very old codecs such as mpeg-2 or vp8 (which was specifically designed to be as fast to decode on cpu as possible), especially at lower resolutions, it definitely can be more inefficient than just using the cpu.
>all it does is kill your phone/laptop faster.
no idea what you mean by this, hwdec is supposed to reduce power draw, not increase it, it's the entire point, it's not "all it does", it's what it can do in some limited scenarios.
>>106028109
>using 80% of your cpu to play 4k videos is not a problem
yes it is retard, fix your shit.

can you explain why vp9 lossless is better than ffv1

>>106028233
hmm, with media.hardware-video-decoding.force-enabled set to true it's now all supported
>>106028302
>yes it is retard, fix your shit.
I have a good cpu

>>106028327
>it's now all supported
but does it actually work or does "force" actually just mean "show it as supported even if it doesn't really work"
>I have a good cpu
which doesn't change the fact that you are burning lots of electricity for no reason and slowing down any other task that is also using the cpu

the OP such a schizo it is unreal
you are literal autist holly shit

>>106028174
A VP9 ASIC LITERAlLY consumes less than 1 watt of power. The only way a CPU would consumes less power is it had a TDP of less than 1 watt you slobbering baboon.

>>106028302
>>106028382
browsers force your gpu into higher power states. the only way you're getting "low power consumption" is if you're viewing this outside your browser.

>>106028376
Who is he anyway? Is he Daiz? You know, that bald butt hurt buffoon!

>>106028388
Okay cool so which CPU with a TDP of less than 1 watt should I buy?

>>106028382
>A VP9 ASIC LITERAlLY consumes less than 1 watt of power.
...on that specific intel laptop cpu, not on your average dedicated desktop gpu, which are way less efficient (the asic itself might use very little power, but the vram, bus, and all that other shit adds a lot of consumption, my rx 7600 uses 10 watts extra compared to idle to playback 4k30 vp9/h265/av1 video), even with video playback you coud have an edge case where cpu decoding is tied, or actually slightly more efficient, but that's beside the point, we are talking about single keyframe images here.
initializing hwdec for an image that could be decoded in less than 10 milliseconds by the cpu is likely to be less efficient than doing it on the cpu.
>>106028388
no they don't, or at least they don't on my machine, there's no difference in power consumption between mpv and a browser for me, there's no reason your browser should force your gpu into an higher power state.

>>106028302
The images come from 4:2:0 video in OP's case. There's a sea of people out there saving video screenshots as PNG images instead of quality 90-100 4:2:0 JPGs.

These ""people"" get to vote...

>>106028370
>but does it actually work or does "force" actually just mean "show it as supported even if it doesn't really work"
huh it's actually doesn't work
>libva error: /usr/lib/dri/nvidia_drv_video.so init failed
but with
>export MOZ_DISABLE_RDD_SANDBOX=1
>export LIBVA_DRIVER_NAME=nvidia
>export NVD_BACKEND=direct
it does, at the small cost of having a non-sandboxed browser which is massively retarded
>which doesn't change the fact that you are burning lots of electricity for no reason
with hw encoding properly working playing the same video my computer burns 740 watts, without - 700 watts lol
>slowing down any other task that is also using the cpu
maybe if you have 1 core without hyperthreading

>>106028488
>it does, at the small cost of having a non-sandboxed browser which is massively retarded
it's not massively retarded you are just a schizoid
>with hw encoding properly working playing the same video my computer burns 740 watts, without - 700 watts lol
...what the fuck? where are you getting those numbers even? is your computer mining crypto with 4+ gpus or something?
>maybe if you have 1 core without hyperthreading
...no?
let's say you are installing a trannygirl repack, which is using 100% of your cpu
playing back a 4k video takes let's say, only 10% of your cpu
the game installation will now be (roughly) 10% slower, it's an objective mathematical fact, you can't just cope with "muuh i have a good cpu" retard.

>>106028543
>the game installation will now be (roughly) 10% slower, it's an objective mathematical fact, you can't just cope with "muuh i have a good cpu" retard.
actually what's also likely to happen is that the video playback will start to drop frames
>but muuuh my cpu is very very stronk
doesn't fucking matter, if a task is using all of it, there's not much left for video decode unless your scheduer is really smart and efficient or you manually set usage limits/priority for each task

>>106028543
>...what the fuck? where are you getting those numbers even? is your computer mining crypto with 4+ gpus or something?
from kill-a-watt
it's the whole home lab including the computer, nevertheless it's 40w overhead for hw video decoding for some reason
>you can't just cope with "muuh i have a good cpu" retard.
what if I can

>>106028606
>nevertheless it's 40w overhead for hw video decoding for some reason
that reason is probably "running a ngreedia gpu on troonix"

>>106025382 (OP)
I've done this before doe and I literally generate my ffmpeg commands with ChatGPT so I don't have intricate knowledge of encoding fuckery
On that note, I'd be happier if there was a way to encode vp9 video with hardware acceleration, but alas, it is impossible, so it's a dead codec in my eyes.

>>106028694
>he needs chatgpt to figure out how to "ffmpeg -i input.jpg -pix_fmt yuv420p output.webm", complete fucking retard, kill yourself.
>I'd be happier if there was a way to encode vp9 video with hardware acceleration, but alas, it is impossible
not impossible, you just need an intel iGPU or intel arc dGPU that supports it (pretty much all of them since the last 5-10 years I think.
but I don't know why you would do that, sure it's slow as fuck to encode, but you probably need it for 4chins webms every now and then which are usually short and low res due to the filesize constraint, plus using hardware encoding at lower bitrates decreases the efficiency even more, not a use case where hardware encoding makes sense, slow/veryslow preset x264 would be a good compromise aready if you realy gotta go fast (probably just 20-30% less efficient than hw encoded vp9, while still much faster than libvpx-vp9.

>>106028694
I doubt VP9 is going away anytime soon. The sheer amount of devices that have hardware acceleration support for it is mind boggling. Which is partly what inspired me to make this thread. AV1 might reach that same level in another decade maybe but until then VP9 will be the kind of codecs especially since unlike kiked codecs like HEVC, you don't have to pay any money to use it.

>Tesla actually rolled back support for YouTube’s newer AV1 video encoding. Instead, vehicles now default to the older VP9 encoding standard.

>While AV1 is highly efficient in terms of bandwidth, it requires considerably more processing power to decode and display videos. VP9, on the other hand, is less computationally demanding but uses more bandwidth to achieve the same video quality. This trade-off means smoother playback and better overall performance, even if it comes at the cost of slightly higher data usage.

>>106028757
forgot source.

https://www.notateslaapp.com/news/2490/tesla-improves-youtube-app-with-smoother-playback

>>106028757
>While AV1 is highly efficient in terms of bandwidth, it requires considerably more processing power to decode and display videos.
this is always a very misleading claim as it only applies in instances where hw decoding is not present.
on a device that has hw decode for both codecs, the difference in power consumption between the two is most likely absolutely 0, plus needing to transmit less data means less power consumption in mobile devices (see picrel, avif already uses less power than jpeg in phones because of this)

>>106028793
Source on your pic?

>>106028755
Yeah I ain't signing up for the humiliation ritual of learning the ffmpeg interface by heart, I am not mentally ill enough.
>but I don't know why you would do that
I just think it would be neat to use it in web services, but there's no way to get a cheap VPS which supports encoding VP9. Of course, if it's for a one-off video for 4channel, I can simply wait it out, though most new video stuff I post comes from twitter directly, and fortunately you no longer need to reencode the shitty h264 to even shittier vp9.
Actually, it would be nice if there was a webp version with vp9 instead of vp8 too, but that is also fucked forever.
>>106028793
You know how many fucks I give about decoding? Literally 0 as long as it works on the client. Encoding AV1, however, it fucking retarded.

OP here, I updated the CLI input and even made it a script you can save and launch as a file. I don't really notice any quality improvements doing this vs saving from the webp file but idk, this is more "correct" I guess?

for %%f in (*.png) do (
ffmpeg -framerate 1 -stream_loop 1 -i "%%f" -c:v libvpx-vp9 -quality best -crf 10 -qmin 10 -qmax 10 -pix_fmt yuv420p "%%~nf.webm"
)

>>106028969
Just copy and paste this text onto a text editor and save it as something.bat and then double click that bat to commence the transformation...

>>106028969
>-c:v libvpx-vp9
completely pointless, it's already the default for .webm files
>-framerate 1 -stream_loop 1
i really have no idea why you do this, all it does is turn it from an actua image which will display indefinitely in mpv to a video that closes after one second, i'll attach an image encoded with just "ffmpeg -i input.jpg -i "input.jpg" -quality best -crf 10 -qmin 10 -qmax 10 -pix_fmt yuv420p output.webm", if it works, you are a massive fag.

>>106029114
>if it works, you are a massive fag.
...yup, works just fine, and hwdec also works, even inside the browser, what was the point of turning the pics into a 2 second video again?
daiz did that shit as well... interesting...

>>106029114
Your webm causes high cpu usage so it doesn't fool my hardware vp9 decoder. Might work work on some VP9 decoders but not all so I'll stick to the added baggage.

>>106028819
>"-i <input> -pix_fmt <chroma sub> <output>" is learning ffmpeg by heart
...is this retard serious
>but there's no way to get a cheap VPS which supports encoding VP9
I don't know how vps exactly work... but surely if they offer hw encoding for h264/h265/av1 there are some that offer an intel gpu which also does vp9, right? why would that not be the case?
>Actually, it would be nice if there was a webp version with vp9 instead of vp8 too
webp2 with vp9 was a thing but then it died, yes, but what would be the point when avif exists?
>You know how many fucks I give about decoding? Literally 0
then why are you replying to a completely unrelated reply to a post talking about decoding performance? I don't get it
>Encoding AV1, however, it fucking retarded
what does this even mean

>>106029185
>Your webm causes high cpu usage
...how the fuck does decoding a single keyframe cause ANY cpu usage at all? are you running a 386? why are you relying on cpu usage at all as an indication on wheter it works or not instead of relying on the actual gpu video decoding utilization? you really get more and more retarded as you go on
>Might work work on some VP9 decoders but not all
as opposed to... your shit that failed to decode for half the people that interacted with this post? jesus fucking christ lol

>>106025382 (OP)
don't mind me
i'm just here to say i liked teppen!!! a lot

>>106029210
>webp2 with vp9 was a thing but then it died, yes, but what would be the point when avif exists?
On a CPU, AVIF is even more resource-intensive to encode than a VP9 frame.

>>106029228
>your shit that failed to decode for half the people that interacted with this post?
and actually I forgot to mention, your webms also have problems on my machine.
after one second, the colors shift, then they stay like that indefinitely, and only happens like half the time, doesn't happen with my example, and it's actually more difficult to tell wheter it's using hwdec or not because it's clearly using less of it, I can only see it spike to 1% sometimes (and no, that doesn't mean it only works half the time, i know you're going to pull that shit).
I'm willing to bet my example is actually more compatible, as it gets actually detected as a fucking still frame image instead of being an odd 1 fps looping video, there's a reason youtube transcodes videos less than 6 fps to 6 fps, nothing below that is standard and expected to work properly on all decoders.
but mine isn't even really a video at all, these codecs do support encoding single images, and i'm just doing that instead of doing it in this weird roundabout way, all of this discussion is so fucking stupid and insane, once again, rope yourself.

>>106029259
on mobile devices avif is already less power hungry than webp, and would definitely still be less power hungry than vp9
on desktop computers, performance is the only constraint and even a second gen i3 is already more than enough to decode a bunch of avifs in a webpage. this "decoding complexity" argument just doesn't make any sense when talking about images.

>>106029289
Which part of the word "encode" eluded you?

>>106027369
>his browser doesn't support 12-bit webms
Works fine on Firefox

>>106029315
my brain didn't even process it as "encode" rather than "decode"... because how on earth is that a problem
in what use case is the encoding speed for a single still image a problem to you, i really don't get it, you don't have to encode hundreds of pictures per second, it really doesn't matter
and besides
>AVIF is even more resource-intensive to encode than a VP9 frame
...it's literally not
you spent all this time complaining about libvpx-vp9 being slow as fuck, that would also apply to a still image
svt-av1 can surely encode an avif WAAAY faster than the time it takes libvpx-vp9 to encode a single frame, while achieving higher efficiency, the same is true for video, i really don't know why you would presume that svt-av1 being way faster than libvpx-vp9 wouldn't also apply when it comes to still images.

>>106029228
>>106029280
I don't know man, this is just a dumb hack afterall. MPC/VLC both spaz out and rape my CPU in addition to the chrome web browser video player with your 40 millisecond video webm. I have no idea why though.

I'm just assuming that in order for this to work at all we have to trick the VP9 decoder into believing it's decoding real VP9 video somehow but maybe there's something else missing if my webms are causing playback/color shifting problems for other anons.

Then again this could be a software/browser issue?

Does this solve playback problems for some of you anons? I've increased the framerate and stream loop both to a value of 10. It takes 10X longer to encode but my hope is that this creates a more "correct" VP9 "video" stream...

Okay I think I figured something out. Framerate seems to affect the VP9 decoder and 6 might be a magic number for that. So then all that we have to figure out is how many stream loops we need.

No loops = 166.666 milliseconds

5 loops = 1 second

Here's 11 loops at 6 fps totaling 2 seconds. Could this be the holy grail?

Anons does this webm give you perfect playback on mobile/desktop web browsers and video players? No weird color glitching or CPU raping?

>>106029445
>>106029882
>>106029945
none of this works
>CPU raping?
mine doesn't have this problem, you literaly made that up to dismiss the fact that you are overcomplicating all of this shit, otherwise, prove it.

>>106025382 (OP)
>which is more powerful
Yes but now your app hooks into special OS APIs that are dependent on a long chain of some of the most retarded and uncooperative people on the planet all cooperating with each other and that's *after* a bunch of bus traffic and waiting on context switches.

>and more efficient.
Nope.

>>106030594
What browser? I can't replicate this.

>>106028543
>...what the fuck? where are you getting those numbers even? is your computer mining crypto with 4+ gpus or something?
Those are pretty normal for a modern desktop with one GPU. They're not "efficient" like you retards keep saying they are.

>>106025382 (OP)
>Yeah but AFAIK it's not hardware accelerated
you don't need hardware acceleration for still images
the overhead from logistics involved in hardware decoding a still image will dwarf the cost of software decoding it
you're an idiot

>>106030650
chromium.
>>106030660
bait used to be believable
5090+14900k under full load could barely hit those power figures, how would that be considered normal for the average modern system during regular use?

>>106030685
>you don't need hardware acceleration for still images
no but it's still nice
>the overhead from logistics involved in hardware decoding a still image will dwarf the cost of software decoding it
I find this hard to believe otherwise why is Apple using this for their HEIF images?

>All iOS devices loaded with iOS 11 and later are able to software-decode HEIF content and HEIC images. However, only devices with Apple’s A9 chip and later are able to hardware-decode HEIF content. To hardware-decode HEIC images and 8-bit video content on a Mac, you will need to use Intel’s 6th Generation and newer CPUs, while 10-bit video content will require 7th Generation and newer Intel CPUs.

https://photographylife.com/definition/heif

>>106030811
>no but it's still nice
how is wasting additional energy "nice"
>I find this hard to believe otherwise why is Apple using this for their HEIF images?
there's a big fucking difference between using a video decoder for something that it wasn't supposed to and using an image decoder... for the thing it was designed to, most android phones also have heif hardware decode but it's either a separate decoder chip, or the hevc decoder chip but that gets initialized differently so it doesn't for example, fire up the parts dedicated to decoding b and p frames.
now i get the suspect that you are actually daiz back, because this has been explained to you 200 times already and as always you never fucking learn anything
and by the way, you understand the concept that the higher the resolution, the more efficiency is gained by switching to hardware decoding, right?
heif image decoding on iphones is used for high megapixel images, while you are using vp9 for 720p/1080p pictures which are like at least 20 times less resolution, and that is also something daiz has always ignored... very interesting.

>pixdaiz is now on his 5th tripcode
this man needs to kill himself asap

>>106030811
>why is Apple using this for their HEIF images?
because they contain HEVC encoded images that take less space but more time to decode
JPEGs don't have that problem

>>106030690
>average modern system
one stick of average modern workstation ram has more ram that an average modern system

>>106025382 (OP)
>Q: Can't you just use JPG?
>A: Yeah but AFAIK it's not hardware accelerated
is there not? amd gpus literally have jpeg hardware block
https://github.com/torvalds/linux/blob/master/drivers/gpu/drm/amd/amdgpu/jpeg_v5_0_1.c

>>106030948
...ok, and?
???

>>106030980
pretty much all gpus have jpeg decoding, yes, but that's for motion jpeg (e.g. webcam or ip camera video streams), it's never used to decode static jpegs, or you have to try really hard to do it.
same thing with avif and webp, stuff that has vp8 and av1 hardware decode could theoretically decode those in hardware as well, but no software will do it because it's retarded.

>>106031063
that's a solution to that 700w power consumption mystery - it's not an average system, and its 24 channels of ecc ddr5 at 6400 mhz need more idling power to stay refreshed than an average system needs to run under an average load

>>106031124
>an average modern desktop system uses 750 watts easily
>me: no it's impossible that's bullshit
>yeah it's possible because my non average system with two epyc processors consumes that much
...again... ?????
literally what the fuck is your point.

>>106031143
you're right, I misread the post you were replying to kek

>>106031174
you deserve an award for the least intelligent comment desu

>>106025382 (OP)
>Normal images are processed by a CPU but these are processed by your GPU, which is more powerful and more efficient
So there is no benefit at all and the player looping the image at a certain FPS actually wastes millions of clock cycles. Got it.

>>106027369
Finally a way to anger chromekeks

>>106031174
>reddit soýjak
grim

>>106025382 (OP)
>A: Normal images are processed by a CPU but these are processed by your GPU, which is more powerful and more efficient.
/g/ - technology everybody!
I suggest you do some basic reading about how computers work before you make any more threads

>>106031900
this, I'm fairly sure the entire viewport is GPU accelerated, but that the CPU feed the data to render and decodes it is not wrong, but it negligible

>>106025382 (OP)
>A few days ago
This was on /a/ few years back idiot.
I distributed berserk manga even someone made a lua script to read those in mpv.
>Normal images are processed by a CPU
If you use Pentium 3.
All CPUs have hardware acceleration now for most image format.
>my ego is bigger than my cock
Anything multiplied by zero is still zero
>>106025957
Use better browser and enable HW decoding.

>>106025382 (OP)
This is ancient and completely useless. Rope. Now. hw accel for images will never be a thing

>>106025382 (OP)
This nigga is using webm to track people. Don't click on his "images"

>>106032198
i fucking kneel anon.

>>106025382 (OP)
Have a (you) since you are so desperate for it.

Yet another codec tripfag? Can't wait for the usual suspects to include them in the templates.

>>106032363
>hw accel for images will never be a thing
It's already a thing for HEIC images on iPhones. There's no way a cellphone CPU can encode 50 megapixels of HEVC image data in 1 second. This is the exact same problem haunting every other modern image format.

>>106033861
Everyone is DAIZ™ until proven.... ughhh DAIZ™?

>>106025382 (OP)
>I'll be adopting a tripcode because my ego is bigger than my cock
that's some Daiz shit right there
>how to store images inside VP9 Webms and trick video decoders into decoding images instead of videos
good to know but what's the point? I see no real use case.
I'll have to write my own hardware accelerated encoder and decoder in the end for it to work seamlessly within my application anyway. This is the only way to avoid bloating my binaries with video decoder's shit.

>thumbnails are still JPG
>only matters if you care opening the link
>no one does in 99% of cases
>but now, saving the picture gives you a Webm
>editing the picture means you have to convert it back into an actual image format
>so webm images are useless
>mfw

>>106035279
We now have the ability to use a GPU instead of a CPU to decode an image now, here on 4chan. Apple is already doing this for HEIF like it's been discussed in this thread. Their solution is obviously more refined than a shitty hack that we're trying to punk out but it's there if you want it.

>>106035334
Good point but I still find this amusing. HEIC support would have been better but it only works on Apple devices which is bullshit.

https://caniuse.com/?search=HEIC


Since this thread is still up maybe some of you anons can help me improve this hack somehow. Currently there are playback issues and they might have something to do with this:

https://trac.ffmpeg.org/wiki/colorspace

When you open my webms in MPC you'll find this in mediainfo tab:

>Color primaries : BT.709
>Transfer characteristics : sRGB/sYCC

Anyway maybe we're supposed to mirror HEIC images regarding this info? I curiously opened a few HEIC images in MPC and I found this in their mediainfo tab:

>Color primaries : Display P3
>Transfer characteristics : BT.709
>Matrix coefficients : BT.601

SOOOOO the playback issues might not be because of -framerate 1 -stream_loop 1 but instead they're due to that missing matrix coefficient information and wrong color primaries/transfer characteristics????

>>106035552
>We now have the ability
yeah, I see that.
I was just hoping for that ability to be readily available and easily adaptable.
Rage is all my shattered hope left me with now.

>>106025382 (OP)
>A few days ago we figured out how to store images inside VP9 Webms and trick video decoders into decoding images instead of videos
Some autists been storing images in webms years ago, to cope with no webp support here. So idk, it's like you're a few years late and out of the loop.

>>106027810
truth nuke
also imagine storing images in google's video format

lmfao

>>106025382 (OP)
Is this how you made those idiotic static webms that are 40% larger than a PNG while having worse quality?

>>106027810
Perfect quick rundown of this thread.

>>106036598
No you don't understand. You must use Google's format. 4:2:0 chroma is good, actually. It's actually good for you because the human eye can't see above 30 FPS and is more sensitive to luminance than chroma so this 720p 4:2:0 lossy image is better than PNG despite taking up more file size and being hugely incompatible with almost every system on the planet because, well, hardware acceleration dude. Obviously.

>>106025382 (OP)
>Q: Can't you just use JPG?
>A: Yeah but AFAIK it's not hardware accelerated.
>Normal images are processed by a CPU but these are processed by your GPU,

Normal images already use YUV subsampling, which has been GPU accelerated since the days of Internet Explorer 11. See: https://jatindersmann.com/2013/09/12/using-hardware-to-decode-and-load-jpg-images-up-to-45-faster-in-internet-explorer-11/. No reason to think that modern browsers 10+ years later don't have it.

Meanwhile opening the video stream, demuxing it, setting up a GPU buffer for it, etcetera requires extra work (the GPU context switching adds extra lag as well, it's why things like skipping in videos is faster with software renderers).
This thread is retarded and it's just a lot of extra work for nothing. I mean I guess it allows you to post webp on 4chan so there's that.

>>106025382 (OP)
What is it with tripfags and being unironic drooling morons

>>106032198
>This was on /a/
/a/ thinks that MPEG-2 DVB-T OTA streams are peak quality and the best of the best.
And they think this, because they are a cargo cult and Japs in Japan, who can't afford anything better, watch their televison animuu over OTA.
It's almost as dumb as cargo culting for random Google form... oh, wait...

>>106028413
Intel N100, see Odroid H4 plus. 12W but if you disable turbo clocks it will use like 3-4 watts for the full system.

>>106028543
>is your computer mining crypto with 4+ gpus or something?

740 watts is just 1 gpu nowadays (assuming the cable does not burn down).

I think there might be another angle to this: AVIF. I used squoosh to make one and then fed it to the script to make webm related. Does this solve playback issues for you anons? Also interesting to note that the only relevant info listed in mediainfo is now

Matrix coefficients : BT.601
BUUT there's been a change to the color range...
Color range : Full

WTF is "color range"?


>>106036598
kek, that does make sense given how some image data won't compress well into 4:2:0 especially text.

>>106036688
Interesting, hitler-chan. I'm too far gone to stop now and aren't there anons saying that software decoding for JPG images is now used instead now? Still even if JPG is being hardware accelerated, it looks like dogshit at low file sizes so this thread is COMPLETELY bonkers I think.

>>106036698
my ego would kill you traveler

>>106036878
Whoah there's something weird about this bt 601 matrix shit, the webm no longer loops on 4chan. What could that mean?

>>106025382 (OP)
>Q: Can't you just use JPG?
>A: Yeah but AFAIK it's not hardware accelerated.

JPG has been hardware accelerated for over a decade.
https://chromium.googlesource.com/chromium/src/+/master/cc/tiles/gpu_image_decode_cache.h

Can't find the source for Firefox, but google shows plenty of results where disabling hardware acceleration changes how image decoding works, which would imply that image decoding is hardware accelerated.

OP is a stupid roodypoo.

>>106036925
I was 1 minute too late but even so VP9 webm quality > JPG quality at low file sizes.

>>106036748
>/a/ thinks that MPEG-2 DVB-T OTA streams are peak quality and the best of the best.

what, and not 1125i MUSE laserdisc transfers? fer shame.

>>106036748
>/a/ thinks that MPEG-2 DVB-T OTA streams are peak quality and the best of the best.
No, they don't.
Stop starting board wars.
We all know /g/ is /a/'s secret identity

>>106036938
So increase the bitrate then? Far easier than encoding your image into a fucking webm.

>>106028793
>energy
>measured in mAh
Holy shit you can't make this up. Actual retarded graph, how can you even take this seriously with such a fucking ridiculous label?
This graph would unironically be better without labeling the horizontal axis. It's that retarded

>>106025382 (OP)
I do enough hardware accelerated video encode/decode to know your idea is dumb. The CPU can decode images faster than the encoded packets can be uploaded to the GPU for processing. Not only that, the video decode pathway is not conducive to final display in a web browser so there's very likely to be another round trip as the decoded video frame is turned into a texture object for display and this very VERY likely isn't done on the GPU in any browser.

>>106037246
I don't believe you, see >>106033896

Or is there something really special about HEIC hardware decoding?

>>106037637
He is talking about sending VIDEO data into the gpu av decoder, which requires lots of state changes and initializing the decoder stack and setting up extra framebuffers etc etc etc.
The pic you quoted measures IMAGE decoding hardware acceleration, which is basically using the YUV to RGB conversion offloaded to GPUs because it can be parallelised heavily so GPUs are ideally suited for it (and may even have dedicated silicon to do as AV decoding also requires lots of YUV to RGB conversion). However this has been available for over a decade by now, even fucking Internet Explorer supports it, so you know it is ancient tech.

Eureka!

None of the colorspace parameters would fucking stick but by editing the output webm now they do. Unfortunately this creates a temp webm file that you have to delete later but I think I have successfully committed VP9 video stream FRAUD. Updated batch script:

for %%f in (*.png) do (
ffmpeg -framerate 1 -stream_loop 1 -i "%%f" -c:v libvpx-vp9 -quality best -crf 10 -qmin 10 -qmax 10 -pix_fmt yuv420p -y temp.webm && ffmpeg -i temp.webm -color_range 1 -colorspace 1 -color_primaries 1 -color_trc 1 -c:v copy "%%~nf.webm"
)

Not completely sure what these things do but all 4 seem necessary to mimic actual VP9 video streams. If anyone has any playback problems, I don't know what to tell you, your software is borked beyond reason.
-color_range 1
-colorspace 1
-color_primaries 1
-color_trc 1

Hmmmm, I think 4chan is stripping one of these. The uploaded webm mediainfo reads:

Color range : Limited
Color primaries : BT.709
Transfer characteristics : BT.709
Matrix coefficients : BT.709

>>106037944
When I download this it now reads:

Color range : Limited
Color primaries : BT.709
Transfer characteristics : BT.709

So the matrix coefficients are fucking gone now. Maybe this is a new security feature against sharty? Uploading a REAL moving video to verify this. Mediainfo of this webm reads:

Color range : Limited
Color primaries : BT.709
Transfer characteristics : BT.709
Matrix coefficients : BT.709

>>106037982
When I download this webm, mediainfo reads:

Color range : Limited
Color primaries : BT.709
Transfer characteristics : BT.709
Matrix coefficients : BT.709

GOD FUCKING DAMNIT. It's not 4chan then, something else is missing in the temp.webm colorspare parameters.

>>106037982
>this trigger femcels

>>106037830
just fucking use jpg/png you mong
nobody cares if they are 500kb larger when it means they work on everything going as far back as the sega dreamcast.

God dam this is way more difficult than I thought it would be. The only other thing I can think of is converting the PNG to a Y4M file.

for %%f in (*.png) do (
ffmpeg -framerate 1 -stream_loop 1 -i "%%f" -pix_fmt yuv420p -y temp.y4m && ffmpeg -color_range tv -colorspace bt709 -color_primaries bt709 -color_trc bt709 -i temp.y4m -c:v libvpx-vp9 -quality best -crf 10 -qmin 10 -qmax 10 -pix_fmt yuv420p "%%~nf.webm"
)

Uploaded webm mediainfo reads:

Color range : Limited
Color primaries : BT.709
Transfer characteristics : BT.709
Matrix coefficients : BT.709

>>106038437
SWEET BABY JESUS IT FUCKING WORKED

Downloaded webm now reads:

Color range : Limited
Color primaries : BT.709
Transfer characteristics : BT.709
Matrix coefficients : BT.709

VP9 video stream FRAUD has been 100% successfully achieved!

>>106038437
>>106038462
just
use
fucking
jpg/png

Nobody gives a shit about your ffmpeg command line jigsaw that can save 1kb space at huge costs to convenience and compatibility, especially on a forum that serves cloudflare lossy recompressed images to begin with. You are basically pissing in the wind.

>>106025382 (OP)
That's cool and all, but what about fucking with the jannies and doing this instead?

>>106039300
>dub

>>106039072
nah, this is fun.

>>106039425
don't mind me, just undoing all the data saving you've done so far

what is the point of this
just upload PNG like a normal person

>>106039986
PNG and JPG DO NOT natively support BT 709, they default to BT 601. With extra metadata (that 4chan destroys LOL) you can sorta get it to support BT 709 but only newer image formats like Jpeg XL and AVIF actually have native support for BT 709 (and BT 2020).

As stupid as it sounds, OP just uncovered a method of sharing BT 709 images on 4chan. Which is kind of a big deal because blu rays have been using BT 709 for like the past 20 years or so. YUP, like half the images on the web have the wrong colors.

https://blog.maxofs2d.net/post/148346073513/bt601-vs-bt709

>>106039986
The OP is one of the bots of Daiz who is behind every WebPiss and Tel Avif shill post.

These tranny codecs get obliterated by JPG, JPGXL and PNG, so he spams the website with this retarded nonsense that WebPiss and Tel Avif have hardware decode support, are faster than JPG and drain the battery less.

>>106040324
this

https://desuarchive.org/g/thread/105959070/#105968430

>>106040324
>>106040344
nigger
nigger
nigger
nigger

>>106025382 (OP)
>A: A few days ago we figured out how to store images inside VP9 Webms and trick video decoders into decoding images instead of videos.
I swear we had this same fucking thread 5 years ago. "We" didn't find anything, you just discovered ffmpeg.

>>106040390
VP9 was added recently like a few years ago so OP isn't incorrect I think. Before that we could only use VP8 which fucking blows and is only like 20% better than mozjpeg.

VP9 should fare better especially since it supports 10-bit but at the same time many hardware VP9 hardware decoders do not support 10-bit so maybe it's a wash...

>>106040431
put your trip back on instead of samefagging, daiz

>>106040460
Nope not him. I'm not praising webp at all, it's barely better than mozjpeg so only a fucking moron would use it.

WEBP2 on the otherhand might not be so bad but we already have Jpeg XL and AVIF so I doubt many people would want to support it.

>>106040390
lol

daiz spams these posts 24/7

>MFW this objectively shitty thread still hasn't been deleted yet!
Tsk tsk!

>>106040431
>>106040490
>T. Daiz
You're a bald butt hurt buffoon who knows nothing about technology!

bump