BtbN changed the topic of #ffmpeg to: Welcome to the FFmpeg USER support channel | Development channel: #ffmpeg-devel | Bug reports: https://ffmpeg.org/bugreports.html | Wiki: https://trac.ffmpeg.org/ | This channel is publically logged | FFmpeg 7.0 is released
<sonicrules1234>
So I did another test of lossless encoding. This time with rgb input. This time x264rgb beat out libaom-av1
Traneptora_ has quit [Quit: Quit]
Traneptora has joined #ffmpeg
stolen has joined #ffmpeg
<sonicrules1234>
I've been upscaling episodes of Ranma 1/2, and the software I'm using to upscale wants rgb8 pixels
<sonicrules1234>
I'm uspcaling them to 960x720 from 480x360
<Lynne>
what are you targetting?
<Thurmond>
generic question: is the point of upscaling to increase the native resolution in a way that uses more processing time than is reasonable during playback but provides better results (in addition to doing that computation once instead of "every playback")
<Thurmond>
?
<sonicrules1234>
Lynne: What do you mean?
<sonicrules1234>
Thurmond: Upscaling so I can watch it full screen. It's done beforehand. It uses ai
hightower4 has joined #ffmpeg
hightower3 has quit [Ping timeout: 265 seconds]
alexherbo2 has quit [Remote host closed the connection]
alexherbo2 has joined #ffmpeg
hightower4 has quit [Remote host closed the connection]
hightower4 has joined #ffmpeg
<Lynne>
no, with lossless encoding
<sonicrules1234>
Running the ai locally though, so it takes about 6 and a half hours per episode
<sonicrules1234>
As for lossless encoding, trying to get the smallest file, but still lossless. I know I can do visually lossless with a much smaller filesize though
<sonicrules1234>
I mean, when you uspcale, the result is technically lossless, so why not?
ops has quit [Remote host closed the connection]
echelon has joined #ffmpeg
echelon is now known as ops
alexherbo2 has quit [Remote host closed the connection]
<putacho>
sonicrules1234: why upscale and not let the renderer do it? IMO upscale encoding is useless. You can't magically create out of nowhere detail, AI or not
alexherbo2 has quit [Remote host closed the connection]
minimal has quit [Quit: Leaving]
<sonicrules1234>
The ai literally create the details out of nowhere. Well, it does it based on it's model
<sonicrules1234>
*creates
<putacho>
I doubt that
hightower4 has quit [Remote host closed the connection]
<sonicrules1234>
You can try it if you want
hightower4 has joined #ffmpeg
<putacho>
you're basically upscaling twice. 1. during encoding and 2. when watching on TV which pulls it up to native resolution
<sonicrules1234>
I'm watching on my steam deck, which is 800p
wziko has joined #ffmpeg
<putacho>
ah
<sonicrules1234>
:P
<putacho>
still worhtless, AI or not
<putacho>
but it's your content, so go ahead
<sonicrules1234>
Well, I'm just doing it for fun, 1080p ranma already exists
<putacho>
I got an AI upscaler on my high-end TCL TV. Trust me, it's not that smart :)
wziko has quit [Ping timeout: 265 seconds]
<sonicrules1234>
AI can do a lot of cool things, and has come a long way from when it first started
<furq>
you absolutely can magically create detail from nowhere
<furq>
whether it's accurate or horrible artifacts is another question
<sonicrules1234>
Yeah, the detail from nowhere is just it's best guess
<sonicrules1234>
But it can be pretty good at guessing
<furq>
i've seen some really impressive results and some really awful results
<furq>
which is par for the course for this sort of thing
<furq>
but the state of the art is definitely better than what a tv can offer
<sonicrules1234>
Yeah, plus a tv has to do it in realtime
<furq>
well even for realtime upscaling
<furq>
but that also makes a huge difference
<furq>
not just realtime but without the tv going on fire
<sonicrules1234>
xD
<sonicrules1234>
My gpu has been heating up my room during the upscaling
<Thurmond>
<sonicrules1234> I mean, when you uspcale, the result is technically lossless, so why not? <-- this is an assumption
<furq>
it is a bit of a shame all this stuff gets lumped in under "ai" now
<furq>
the machine learning guys must be furious that everyone thinks their job is chatgpt
<sonicrules1234>
Well, since it's going from bitmap to bitmap in my case, it is indeed lossless
zsw has quit [Remote host closed the connection]
<sonicrules1234>
The loss would be any colorspace converting
<Thurmond>
whether it uses ai or machine learning only goes to the quality of the results. i guess what i'm asking is: is the point to reduce the load on the renderer and/or provide better results than the renderer can provide *in real time* ?
<sonicrules1234>
Neither
<Thurmond>
are those things not true, or just don't apply to you?
<Thurmond>
because i'm not asking about you in particular
<furq>
that is broadly what it's for yes
<furq>
the very intensive upscalers will use more resources than encoding the video
<sonicrules1234>
Yeah
<Thurmond>
i have noticed, for example, that VLC sometimes consumes much more battery power when expanding certain streams, but not when they just play at their native resolution, which means that at least in some cases, the real-time processing load is non-trivial
<Thurmond>
i have no way of knowing how generally true that is. it just made logical sense to me in THIS case.
<furq>
that sounds unusual
<furq>
i guess it depends what upscaler vlc is using
<Thurmond>
and it's more noticable on this old machine, which means that the effect is sort of magnified (using the term loosely), which doesn't make it not true, but may be exaggerating the scale (har har) of the cost.
<Thurmond>
i hope that makes sense. i'm not being super precise
StephenLynx has quit [Ping timeout: 272 seconds]
zsw has joined #ffmpeg
StephenLynx has joined #ffmpeg
yans has quit [Remote host closed the connection]
coldfeet has joined #ffmpeg
yans has joined #ffmpeg
Krusher has quit [Quit: Leaving]
billchenchina has joined #ffmpeg
billchenchina has quit [Remote host closed the connection]
Thurmond has quit [Ping timeout: 252 seconds]
fristed has quit [Quit: system restarting]
wziko has joined #ffmpeg
zeromind has quit [Quit: *pooof*]
darkapex has quit [Ping timeout: 252 seconds]
darkapex has joined #ffmpeg
Thurmond has joined #ffmpeg
luva has quit [Ping timeout: 244 seconds]
JanC has quit [Read error: Connection reset by peer]
<welder>
I submitted a ticket #11330 some time ago, about that audio being cut, is it a bug or expected behavior?
iive has joined #ffmpeg
JanC has joined #ffmpeg
fristed has joined #ffmpeg
CruxOfTheB has quit [Ping timeout: 265 seconds]
mw1144 has quit [Quit: Client closed]
luva has joined #ffmpeg
hightower4 has quit [Remote host closed the connection]
hightower4 has joined #ffmpeg
<redeeman>
is it possible I can take 1 input file, run filter to denoise, then encode one output in original resolution, and also scale down the denoised to 720p, and then encode another output of that ?
<ePirat>
you can do this with -filter_complex
<ePirat>
use output labels for the filters respectively and map them for the outputs the way you want
<redeeman>
nice, thanks. do you happen to have a link to some examples or something similar?
hightower4 has quit [Quit: Leaving]
stolen has quit [Quit: Connection closed for inactivity]
relue has joined #ffmpeg
rsx has quit [Quit: rsx]
rsx has joined #ffmpeg
krjst has quit [Ping timeout: 252 seconds]
krjst has joined #ffmpeg
zeromind has joined #ffmpeg
lolok has joined #ffmpeg
Blacker47 has quit [Quit: Life is short. Get a V.90 modem fast!]
coldfeet has quit [Quit: Lost terminal]
Thurmond has quit [Read error: Connection reset by peer]
System_Error has quit [Remote host closed the connection]
Thurmond has joined #ffmpeg
ops has quit [Remote host closed the connection]
echelon has joined #ffmpeg
echelon is now known as ops
anticw has quit [Remote host closed the connection]
anticw has joined #ffmpeg
System_Error has joined #ffmpeg
tomb^ has joined #ffmpeg
<tomb^>
hi, i'm trying to list decklink devices (ffmpeg -f decklink -list_devices 1 -i dummy) however it returns "unknown input format 'decklink'", I understand I need to compile it with decklink support, i'm not sure I know how to do that, is there a precompiled version I could somehow download?
<ePirat>
you need to download the decklink sdk and pass the relevant configure parameters
<ePirat>
there are no prebuilts with ffmpeg enabled
<ePirat>
*with decklink enabled
rex has quit [Ping timeout: 248 seconds]
<tomb^>
oh shoot... copy that, not sure I'll be able to compile it on my own, will look into it, thank you!
Galisma has joined #ffmpeg
Galisma has quit [Client Quit]
<ePirat>
tomb^ , which OS are you on?
<tomb^>
I'd rather not say... windows
<ePirat>
oh yeah thats a bit… non-trivial then
<furq>
yeah it kind of sucks on windows
<furq>
imagine that
<furq>
msys2 is probably the easiest way
<ePirat>
not so much for building ffmpeg even, thats doable, but decklink SDK is not provided as headers for windows but need to be generated from idl files
<tomb^>
my goal is to extract the audio stream from a decklink card and stream it to google for speech to text, thought ffmpeg could help with that..
<tomb^>
https://superuser.com/a/1260130 I'm trying compiling through wsl, I've installed ubuntu 18.04 but when trying to cross compile i'm getting "your meson version is too old 0.45.1 wanted 0.49.2", is there a different ubuntu version you guys recommend trying?
<furq>
ideally one that's not six years old
<furq>
not sure what depends on meson though unless you're using one of those build helper scripts
<tomb^>
I'm using the "/cross_compile_ffmpeg.sh –disable-nonfree=n"