[0:00] Hello once again audience. Welcome back
[0:03] to another video. And today we're going
[0:06] to do something different. I'm going to
[0:07] show you guys how to upscale a video and
[0:10] interpolate a video to 60 frames per
[0:13] second. But first, please subscribe to
[0:16] the channel if you haven't already. I
[0:18] would greatly appreciate it. This video
[0:20] costs a lot of time and effort to make
[0:22] and I would really appreciate if you
[0:24] could just subscribe to the channel to
[0:26] get me to a th00and subscribers. I mean,
[0:29] we got a bunch of views already and a
[0:30] bunch of watch time. We just need the
[0:33] subscriber count. So, I'm going to
[0:36] explain real quick what upscaling means
[0:39] and what video interpolation also means.
[0:42] Upscaling is the process of increasing
[0:44] the resolution of an image or a video
[0:47] without reducing the quality or creating
[0:51] ugly stretches.
[0:53] And
[0:54] uh in this case you can also add more
[0:57] details by adding an AI model to improve
[1:00] the quality or add more details or more
[1:03] sharpness whatever you want.
[1:05] And video interpolation is the process
[1:07] of adding more frames per second to a
[1:11] video. So basically generating in
[1:12] between frames uh for your video so you
[1:16] can make them buttery smooth.
[1:19] And uh basically what I tried to do
[1:22] there was uh run a compui workflow to
[1:26] upscale a video
[1:29] uh which is getting split into separate
[1:31] images and that's the the video right
[1:34] there which I showed in the intro and
[1:36] those images are getting upskilled
[1:38] separately. However, due to my PC being
[1:41] kind of poopy and not that great uh it
[1:45] crashed. So, what I'm going to do is I'm
[1:47] going to manually
[1:50] um split the images. I'm pretty sure
[1:52] there is a compi workflow node somewhere
[1:56] which allows you to do parts of the
[1:59] batch.
[2:01] Uh so what I did first I was splitting
[2:04] the audio from the video because if you
[2:07] put images uh together you don't have
[2:10] the audio. So, you need to keep the
[2:11] audio from the original video uh
[2:14] somewhere safe, which is what I did
[2:16] first. And next up, I uh I use a little
[2:19] command using ffmpeg uh which is the
[2:22] program I use for video and audio
[2:26] uh processing, editing,
[2:29] uh just trying to combine videos and
[2:31] whatever if I get stuck uh grabbing
[2:34] images from videos, etc., etc. That's
[2:36] what FFmpeg It's open source library and
[2:40] what I'm doing now is I'm running a
[2:42] command to uh extract all the frames of
[2:46] the video and create PGs from them. And
[2:51] uh you can see it takes a little bit of
[2:53] time and it should be around 300 images
[2:56] because it's 300 frames obviously.
[2:59] Uh, don't worry. I'll put all the
[3:01] commands in the description because uh,
[3:04] you're not going to remember everything
[3:05] I type. And I guess typing it well,
[3:07] pausing the video is also kind of
[3:08] annoying. So, don't worry. I put
[3:09] everything into the description. Um, so
[3:12] now we have the little uh, images here.
[3:15] I'm just going to put them in a separate
[3:17] folder for safekeeping
[3:20] so I can easily use that folder in a
[3:22] comfy UI.
[3:24] And besides Scumview, I also have
[3:26] Lossless Scaling for frame
[3:28] interpolation, which is like a Steam
[3:31] program which you can use for screen uh
[3:35] applications or games to create 60 fps
[3:38] without actually having to do any video
[3:39] editing. Uh, and I'm also going to show
[3:42] ART video enhancer, which is a paid AI
[3:45] tool. I got a free trial. I'm going to
[3:46] show you guys how it works. Uh, I'm not
[3:49] sponsored, by the way. I have to say I'm
[3:51] not sponsored. Uh, I hope they sponsor
[3:54] me. So, if you're watching the ARD, yo,
[3:56] sponsor your boy, please.
[3:59] So, what I did here is, uh, basically I
[4:01] had a little conf workflow and I grabbed
[4:06] a batch of images and I upskilled them
[4:09] to, uh, 4K resolution uh, without the
[4:13] use of the model. I do have a model in
[4:15] this little preview, but uh actually
[4:20] using that model kind of tenfolds the
[4:23] time. This already took over an hour to
[4:25] do in total. So, I'm just using like
[4:27] little batches here to to upskill them.
[4:32] And uh you can see that it takes about a
[4:36] minute for like 50 I want to say.
[4:40] uh maybe a bit longer but the process of
[4:43] like generating the images and doing all
[4:45] that stuff setting up confiu is such a
[4:48] drag uh I'm using Pinocchio
[4:52] and I think it has like 18 gigs of
[4:54] storage usage just for nodes and other
[4:58] libraries and whatnot. It's huge and uh
[5:02] it really eats up my PC.
[5:05] Uh, I I had a previous version of this
[5:08] video and there was just static noise
[5:10] and just the viewing experience wasn't
[5:13] that great. Honestly, I'm sorry. It's
[5:16] just cuz um because of my PC was kind of
[5:19] dying. So, that's why I'm re-recording.
[5:22] If you saw the other video, that's why
[5:24] I'm re-recording
[5:26] for uh for better audio because I I
[5:29] checked it out. I'm like, dude, this is
[5:30] awful. I cannot just put this out. And
[5:33] also I added some some nice new tweaks
[5:36] during the results section. So I I'm
[5:39] really happy how that turned out and
[5:41] might be a bit easier to spot the
[5:43] differences now.
[5:45] So uh yeah, you can see I think this
[5:48] takes like 2 minutes for 70 uh and then
[5:51] it's sourced to disk. So let's say it
[5:54] takes about two minutes for 60 and it's
[5:57] uh 300 frames.
[5:59] That's about uh uh 10 minutes for
[6:03] upscaling.
[6:05] Uh which isn't too bad uh considering
[6:11] ARD also takes about 10 minutes.
[6:14] Uh but there is a quality difference
[6:16] because ARD is using a
[6:20] AI model to increase the details and I'm
[6:23] not doing that here because that would
[6:25] probably make the runtime close to one
[6:28] hour maybe two hours
[6:31] uh just for for the image upscaling and
[6:34] then I'd still have to do 4K and
[6:38] I combining it back into a video. So,
[6:42] and for the sake of the video, I'm not
[6:44] using a model, but you could use a model
[6:46] if you have more RAM. Uh, I have a GTX
[6:50] 1070
[6:52] and it's doing fine. It's got 8 gigs of
[6:55] VRAM and it's it's holding up fine. The
[6:57] problem is,
[6:59] uh, mostly the RAM. I'm running out of
[7:01] RAM. I have 16 gigs of RAM, which is
[7:03] fine like like
[7:05] uh like eight years ago when I got the
[7:07] PC, but nowadays it's not. And and Comy
[7:12] UI just eats everything up and
[7:16] uh like I'm not sponsored by by ARD, but
[7:19] ARD has better um system resource
[7:24] management basically. So I'm just
[7:26] showing off the um the image there. it.
[7:30] It's upscaled to 4K now. It looks maybe
[7:33] a little bit sharper. I don't know.
[7:34] We're going to find out at the results
[7:37] section later in the video. And what I'm
[7:40] doing now is I'm uh combining
[7:44] the
[7:46] uh frames back into a video. And I'm
[7:49] specifying using the minus r uh
[7:53] parameter. I'm I'm specifying the frame
[7:55] rate basically. And I'm also setting up
[7:58] a empty audio clip. Uh because when you
[8:03] want to merge like concatenate videos or
[8:05] if you want to add audio to an existing
[8:08] video, it's sometimes it kind of
[8:11] uh like doesn't want to because you
[8:14] don't have an audio track on the video.
[8:17] So it's like just just a GIF basically.
[8:20] It treats it like a GIF and a GIF can't
[8:22] have sound. So if you have an empty
[8:23] audio track uh by using the 128k thingy
[8:28] um you ensure that you are not going to
[8:31] run into issues later
[8:33] uh adding audio or combining videos
[8:35] which do have audio and videos with jon
[8:37] of audio. So I had like um a banner
[8:42] project before. I had like a banner and
[8:44] I tried to merge it uh with like an
[8:46] actual video but the banner didn't have
[8:48] any sound. It was like just one frame of
[8:50] banner. um and um that caused issues. So
[8:55] that's why I know you you find out stuff
[8:57] like that. Uh again, don't worry. I'm
[8:59] just putting everything into the
[9:01] comments.
[9:02] And you can see this takes quite a long
[9:04] time. Uh probably because each of the
[9:06] images is like five megs big, five
[9:08] megabytes, over 300 images. That's um do
[9:13] the math, 1.5 gigs of images.
[9:17] uh it's it's adding into a video and
[9:21] honestly the video turns out to be like
[9:24] 55 megabytes or something. It's not
[9:26] really that big. Uh and ARD basically
[9:30] has the same thing. You know it it takes
[9:31] a 2 gig raw video and it converts it
[9:34] into like 55 or 20 megabytes video 4K.
[9:38] So it does do some very good
[9:40] optimization there. uh which I think
[9:43] it's H 264 265 something.
[9:48] So
[9:50] uh now I'm going to be
[9:55] uh let me check.
[10:00] I
[10:03] grab the video and I grab the audio.
[10:06] Yeah. And I combine them back to one. So
[10:09] now we are finished with our 4K video.
[10:13] We we've upscaled it perfectly and we're
[10:16] going to combine it with the other. This
[10:18] is a very fast process cuz basically
[10:21] you're just adding the track to to it.
[10:25] And this is how it looks like now. And
[10:27] at the end of the video, I'll be making
[10:29] comparisons so you can compare for
[10:31] yourself. And let me know in the
[10:33] comments if you think it looks any
[10:34] different. Uh and if so, which version
[10:37] do you think looks better? I'm really
[10:39] excited to hear you guys opinion. So
[10:41] this again is the um
[10:44] load video note which is the first one
[10:47] on the left there. Uh you can load a
[10:49] video in and then you can kind of
[10:52] interpolate it using another node. Uh
[10:55] basically I'm multiplying it to two
[10:57] meaning it will get uh 60 frames instead
[11:00] of 30. And then uh like the first node
[11:03] is actually splitting the video into
[11:05] images which uh basically I was doing
[11:08] myself before as well. Uh and then the
[11:11] the middle node is taking those images
[11:14] and then adding in between frames. And
[11:17] then the last node is actually combining
[11:22] the images back into a video. So
[11:26] basically
[11:28] I could work with only getting the in
[11:31] between frames and then generating the
[11:33] file myself. But the last node is not
[11:35] the issue. The first node is the issue
[11:38] which again crashed my PC because I
[11:41] don't have enough RAM. So I had to
[11:45] create smaller bits of video. So I I cut
[11:50] all the 300 frames into little parts.
[11:55] and like 50 frames per video, which is
[11:58] like 1.8 seconds, I guess, on a 30 frame
[12:02] video, 30 frames per second video. And I
[12:05] cut them in six parts cuz it crashed.
[12:09] And uh this this worked. And it's
[12:13] basically the same command as before.
[12:14] We're just trying to to create another
[12:16] video with the um 30 frames per second
[12:21] parameter, the minus or 30. empty audio
[12:24] track just cuz. And uh uh yeah, we
[12:28] create six of them. And I have a very
[12:30] bad naming schemes, you know, B1, A1,
[12:33] C1. Uh I just want to like get this done
[12:36] and I remove all the files afterwards.
[12:38] So like when I'm doing stuff like this,
[12:40] I usually have very default commands in
[12:42] a in a notepad or something
[12:45] uh with like a b c uh and they are
[12:48] sequenced. So I know like oh I'm at C
[12:51] step which is the third step. So I'll
[12:53] try to do the same in the comments as
[12:55] well for you guys. So you can easily
[12:57] copy paste it in sequence for one
[12:59] project and then uh you have like a a
[13:02] nice finished end result.
[13:05] So uh this like 50 frames not too long
[13:08] takes uh almost 30 seconds. Yay.
[13:12] And now I put it in into the um load
[13:16] video again. Uh I do notice that you can
[13:19] kind of limit the frames. Uh maybe you
[13:22] can set a different starting frame. Uh
[13:24] and you can kind of batch it like that.
[13:29] That might work. But I'm also scared
[13:31] that it's because it's a larger video,
[13:33] it's trying to still load some stuff
[13:35] into memory, which is going to cause it
[13:38] to die again, and I don't want it to die
[13:40] again because just sucks so much.
[13:44] Um like I didn't get a blue screen. My
[13:47] entire PC just completely froze. But you
[13:49] can see that the first step was finished
[13:51] way faster now, which again, less RAM
[13:54] needed. So, it all fits into my little
[13:57] 16 gigs memory. Uh, if you have more
[14:00] RAM, I I guess you could do this a lot
[14:03] easier with bigger videos. Let's look at
[14:06] the the the usage there. It is already
[14:10] for 50 frames already almost capping my
[14:13] my memory. It's taking 30 gigs of my
[14:15] swap file on my hard disk. like you can
[14:18] set up a certain section of your hard
[14:20] disk for extra RAM, which is really
[14:23] slow, but you can see that's also taking
[14:25] like 30 gigs, 28 to 33. Um, and that's
[14:30] that's besides my 15 to 16 gigs actual
[14:33] RAM usage. So, it's almost 50 gigs of
[14:36] RAM usage for a 1 second video for
[14:40] interpolation. So, that's you need like
[14:43] 64 gigs at least to do this stuff.
[14:46] Uh, so I finished all the uh little
[14:48] parts there
[14:50] and I'm combining them in order. You can
[14:52] have a little text file with ffmpeg
[14:55] and it will combine those videos in
[14:58] order and it will just concatenate those
[15:00] videos uh without any transitions. You
[15:03] might be able to do a transition with
[15:05] some fancy stuff maybe. I don't know.
[15:08] You would have to look it up. But if you
[15:10] don't have a video editor, but you want
[15:12] to have like a transition between clips,
[15:15] uh you could maybe do something like
[15:17] that. And also
[15:19] um you can easily combine clips using
[15:22] ffmpeg. You don't need a fancy video
[15:24] editor. You don't need Adobe Premiere
[15:27] for just combining clips. Uh also Adobe
[15:31] Premiere sometimes just crashes and
[15:32] doesn't want to combine videos and and
[15:34] just creates corrupted stuff. Uh, Fmpe
[15:38] is really stable for video creation, but
[15:40] if you want to do animations and like
[15:44] fancy text and whatnot, you might need a
[15:46] video editor.
[15:48] So, I combined the 60 fps video now as
[15:51] well. There we go. It's now 4K 60 fps.
[15:55] There is no sound again because we do
[15:57] have to add the sound to the video
[16:01] again, which as we know from before is a
[16:04] very fast process. So,
[16:07] almost done with the 4K 60 fps uh AI
[16:10] version.
[16:12] I am also going to create a
[16:16] nonupskilled 60fps version. See how that
[16:19] runs and I'll show the comparison later.
[16:24] That was me. Sorry, I pressed the
[16:26] recording button. Uh that's not your
[16:28] Windows doing a little plink sound.
[16:29] That's just me. I'm sorry. I hate it
[16:31] when YouTubers make a sound like Discord
[16:33] notification or whatnot and they don't
[16:35] mention it. This was me, okay? So, I'm
[16:37] sorry. Uh, don't don't be scared. So,
[16:40] again, we're we're combining the audio
[16:42] with the um with the video. And like I
[16:47] said before, I was going to try to do a
[16:49] 60 fps version with out upscaling. And
[16:53] that one took maybe like 20 to 30
[16:58] minutes. Uh, but it didn't crash my RAM.
[17:01] It just took a very long time. Like I
[17:02] was still able to watch some YouTube and
[17:05] and open Discord. I could I could do
[17:06] stuff while it was running. So,
[17:10] it's really good. Uh, that's the wrong
[17:13] version. Here we go. This is the uh the
[17:16] sound version again.
[17:18] 60 fps with sound.
[17:21] It looks It looks decent. I guess I'll
[17:23] have to compare it with uh with the
[17:25] original one later.
[17:28] Uh, but that's basically it. It took us
[17:29] about 15 minutes and 70 minutes even.
[17:33] And now I'm going to show you ARD video
[17:36] enhancer. I'll put the link in the
[17:38] description. Again, I'm not sponsored.
[17:40] Uh, but it's very easy. It's very
[17:42] optimized and it checks your hardware
[17:44] and everything. You can upscale using a
[17:47] model. You have three models of choice.
[17:49] I'm not sure if you can get other models
[17:50] somehow. You can upscale to 4K. You can
[17:53] use step mode for higher quality. You
[17:56] can use turbo for faster processing. Uh
[18:01] but it is uh uh it is lower quality. So
[18:05] here you can see like I have a little
[18:07] slider. You can see the edge of the uh
[18:10] ridge basically gets a little sharper.
[18:12] Let's look at Maido for a sec. Um yeah,
[18:16] he does get sharper. Like clearly
[18:17] sharper. Let's zoom in a little bit.
[18:19] There you go.
[18:21] Uh so you can see he does get a little
[18:23] sharper. You do have to be careful with
[18:25] the strength though, especially when you
[18:27] have a very low res image. Uh sometimes
[18:30] AI is adding some weird details like
[18:33] with text. I've had instances where uh
[18:37] the C turns into a D or a O. And uh as
[18:42] you can see, you can also increase the
[18:43] frame rate very easily. Uh you can
[18:46] change the color. Uh you can use HDR uh
[18:50] audio D noiseis like whatever you want.
[18:53] and you just press the the uh export
[18:56] button. You can even do a batch export
[18:58] of multiple videos using the same
[19:00] settings. So, you can see it takes about
[19:02] 10 minutes for a 10-second video. Uh
[19:06] it's a lot less setup. It's very
[19:09] much faster. So, I hope you enjoyed this
[19:12] and I hope you enjoy the results in a
[19:14] little bit. Uh because yeah, I think
[19:18] they look pretty cool. So, please
[19:20] consider subscribing to the channel if
[19:21] you have not already. I would greatly
[19:23] appreciate it and I hope to see you all
[19:24] next time. Let me know in the comments
[19:26] which versions you like the most and if
[19:29] you notice any big differences or not.
[19:32] So, thank you for watching. I hope you
[19:35] found this useful or interesting at
[19:37] least. And see you tomorrow with Maple
[19:40] Story again. I'm sorry. Okay, bye-bye
[19:43] everyone.
[19:46] [music]
[19:56] >> [music]
[20:08] >> Don't
[20:36] >> [music]
[21:06] [music]
[21:43] >> And now a shout out to our members. On
[21:46] the extreme tier we have Dustin H. On
[21:49] the normal tier we have Daniel Chavez.
[21:53] Thank you all so much.
[21:55] See you all next