[0:00] You can now upscale your footage locally
[0:02] to HD, 2K, or even 4K resolution. Our
[0:05] workflow will break down your video into
[0:07] smaller segments and upscale them tile
[0:09] by tile, making it easier for your
[0:11] computer to handle. It uses generative
[0:13] upscaling, meaning it will actually
[0:15] generate new information into your
[0:17] video, making even pixelated footage
[0:19] like this usable.
[0:22] We built this workflow to switch over
[0:24] from subscriptionbased paid tools, which
[0:26] this workflow also outperforms in many
[0:28] cases. and now we're giving it to you.
[0:30] So, here's how you can upscale your
[0:31] footage free and locally.
[0:37] This workflow is based on the one 2.2
[0:39] video model that lets you generate 720p
[0:41] videos for a maximum of 81 frames, which
[0:44] is by far not enough resolution or time.
[0:46] So, to fix this, we built a custom
[0:48] workflow for Comf UI, a free nodebased
[0:51] interface for AI models. And to install
[0:53] it, you can just follow the guide on our
[0:55] website links below. Once you have it
[0:57] installed, you just need to download the
[0:58] JSON file and drag and drop it into the
[1:01] Comfy UI interface. You'll need to
[1:03] install these missing custom nodes. Do
[1:06] that by going to the Confui manager,
[1:08] install missing custom nodes, select all
[1:10] of them, and click install. Once it's
[1:13] done, restart Confui and the full
[1:15] workflow is here. Now, we need to
[1:16] download the actual models for this
[1:18] workflow to work. And you can find all
[1:20] of them to the left here in these yellow
[1:23] notes. First, we need to decide which
[1:24] kind of base model we want to use for
[1:26] this workflow. If you have a good GPU
[1:28] with like 24 GB of VRAM or more, you can
[1:31] use this FP8 version of the model, which
[1:33] will be just a tiny bit faster. To use
[1:35] it, you need to download it from this
[1:37] link and put it inside of CompuI models,
[1:40] diffusion models. Then load it here.
[1:42] Connect this model to this node right
[1:45] here. But in most cases, I recommend
[1:47] using the GGUF version of this model.
[1:49] You can find all of the GGF models right
[1:52] here. GGF is basically a way to compress
[1:54] model size so that it can run on GPUs
[1:57] with lower VRAM. This will cost you some
[1:59] quality, but usually it's not that bad.
[2:01] You can ignore all these high- noise
[2:03] models here. We just need one low-noise
[2:06] version. Check how much VRAM your GPU
[2:08] has, and then you select one of these
[2:10] versions that comfortably fits onto your
[2:12] GPU. If you only have 12 GB of VRAM,
[2:14] check out the Q5 versions. But I'm going
[2:17] to use the Q8 version. Just download
[2:20] that and put it in confusi models unit.
[2:23] Make sure it's selected right here.
[2:25] Next, we're going to use the light x2
[2:27] vora which allows us to run this
[2:29] workflow at a lower step count,
[2:31] increasing performance by so much. You
[2:34] just need to download the low-noise
[2:35] Laura. Download that, put it inside of
[2:37] Confui models and Loras, refresh, and
[2:40] load it right here. The next two Las
[2:42] here are completely optional, but if
[2:44] you're upscaling a lot of realistic
[2:46] footage, I recommend you get at least
[2:48] this one right here, the stock
[2:50] photography. What this one does is
[2:52] basically it just pushes the style
[2:54] towards more realistic video. Next, you
[2:57] need the clip model. You can get that
[2:58] right here, the VAE. Put it in VA and
[3:01] load it here. And next, we need an
[3:02] upscaling model. You can use any
[3:04] upscaling model here that you can
[3:06] install via the CompuI manager. I
[3:08] recommend really going with this one. So
[3:10] to install it, go to manager, model
[3:12] manager, search for X2, then download it
[3:16] right here. I already have it installed.
[3:18] That's all you need to do to set up this
[3:20] workflow. If you want to skip all these
[3:21] installation steps, or maybe you don't
[3:23] have the most powerful computer, you can
[3:25] also run this workflow on RunPod. Runpot
[3:28] is a cloud GPU platform that we know a
[3:30] lot of you guys use. So we built this
[3:32] template for you that just starts up the
[3:34] workflow with everything installed for
[3:36] you. You just need to follow the link in
[3:38] the description. Set up your account on
[3:39] Runport, add some credits to rent a GPU.
[3:42] I usually run this on a 5090, which
[3:44] costs around 89 cents an hour. Now,
[3:47] let's upscale a video. The cool thing
[3:48] about this workflow is that we made it
[3:50] as plugandplay as possible. There is
[3:52] actually a lot of stuff happening in
[3:54] these groups here in the background, in
[3:56] these subgraphs that you never have to
[3:57] check out or open. There's a bunch of
[3:59] math happening here, but you don't need
[4:01] to care about that at all. You can just
[4:03] load the workflow and choose a video
[4:05] right here. Let's take for example this
[4:07] here. This should be a very good
[4:09] challenge. You can see the quality is
[4:11] just horrible and we're just trying to
[4:12] be able to use it. Next, you can set the
[4:14] frame load cap and this should
[4:16] theoretically work. You can upscale
[4:18] videos with any length, but this is
[4:19] really intense. So, I'm just going to
[4:21] cap that at 81 frames. Next, you can
[4:24] come down here and this is where you set
[4:25] up the whole workflow. Put your final
[4:27] output resolution. And I'm just sticking
[4:29] with 2K here. I feel like 2K is a very
[4:32] good starting point where it adds a lot
[4:34] of detail, but it also doesn't take too
[4:36] long. And then you can choose if you
[4:38] want to enable this here for higher
[4:40] quality. I would say this adds about
[4:41] like maybe 30% of quality, but the
[4:44] problem is it takes like nearly twice as
[4:46] long. So I usually do this for smaller
[4:48] upscaling sizes like this one right
[4:50] here, 2K or HD. But when I go to 4K, it
[4:53] really is not worth the time. So I
[4:55] usually switch that off for 4K video. So
[4:58] here we have this creativity slider
[5:00] which just tells the workflow how much
[5:01] it can change in the image. And I
[5:03] usually start pretty low, something like
[5:05] this right here. Finally, these are the
[5:07] iterations that the workflow will
[5:08] automatically create. So if you have a
[5:10] video that's longer than 81 frames, it
[5:13] will create multiple 81 frame batches
[5:16] and then blend them together so you
[5:18] don't realize that there are any seams.
[5:20] For me, it's 81 frames. Don't go above
[5:23] that, but you can go below that if you
[5:25] don't have the best GPU. 41 frames is
[5:27] also pretty good. And next are the
[5:29] overlap frames. So, how many frames does
[5:31] this workflow have to blend these
[5:33] iterations together so you don't notice
[5:35] that there are any seams? We realized if
[5:37] you want to upscale really long videos
[5:39] to like 4K resolution, in the end, it
[5:42] can actually crash compi when it's
[5:44] stitching everything together. So, if
[5:46] you activate this setting right here,
[5:47] you always have all the images that it
[5:49] already upscaled as a fallback. So you
[5:51] can import them into After Effects or
[5:53] your video editing software of choice
[5:55] and then just stitch the video together
[5:56] there. So now we can just click run and
[5:58] as you can see it will automatically
[6:00] create a very detailed prompt for this
[6:02] video. If it's not perfect or you want
[6:04] to add more detail, you can come up here
[6:06] and manually input something here. You
[6:09] can also input a negative prompt, but
[6:11] this is usually enough. So all I need to
[6:13] do now is wait for this workflow to
[6:15] finish. Now, while this is running, this
[6:16] is a good time to mention that this
[6:18] workflow and video are sponsored by our
[6:20] amazing Patreon community. If you want
[6:22] access to exclusive example files,
[6:24] advanced versions, and our amazing
[6:25] Discord community, consider supporting
[6:27] us on Patreon. Your support makes
[6:30] creating these workflows and sharing
[6:31] them with you all for free possible. And
[6:33] it also makes it possible that we don't
[6:35] have to rely on sponsors in every single
[6:37] video, which is just amazing. Thank you
[6:38] so much for that. 10 minutes later and
[6:41] the video is done. So, let's look at the
[6:43] result. And you can see this was the
[6:45] original video and this is after. Before
[6:49] after. You can see how much detail it
[6:52] was able to generate, especially in the
[6:54] hair here. But it's still not perfect.
[6:56] We could improve this result even
[6:58] further if we allowed a bit more
[7:00] creativity. You can see the hair is now
[7:03] more detailed and the overall quality is
[7:06] higher. If you use a high creativity
[7:08] value, it can be more likely that you
[7:09] see these tiles here. And sometimes you
[7:11] just have to experiment what works best.
[7:13] But here, for example, I tried the
[7:14] highest possible value and I did not
[7:16] have the problem with the tiles at all.
[7:18] We also upscaled the same video using
[7:20] different upscaling models by Topaz.
[7:21] Even the 4K output from Topaz doesn't
[7:23] compare to the better 2K output from our
[7:26] workflow. Now, let's talk about the
[7:28] creativity value. The higher you set the
[7:30] value, the more it will modify your
[7:32] video content, and this might change
[7:33] your subject a bit too much. So, if you
[7:35] need a super precise upscale that stays
[7:37] faithful to every detail in your
[7:39] footage, keep that creativity low. The
[7:41] trade-off is that you'll get less
[7:43] dramatic quality improvements. Overall,
[7:45] we also tested this against Flash VSSR,
[7:47] another local upscaler. There is a
[7:49] version for Chromei, but it does not
[7:51] produce the best results yet. We
[7:53] recommend installing it via Pinocchio.
[7:55] Pinocchio is this one-click installer
[7:57] for AI tools, and it makes the
[7:58] installation super easy. Flashvsr is
[8:01] great at precision. It sticks very
[8:03] closely to your source material.
[8:05] However, it really struggles with
[8:07] extremely lowquality videos. But when we
[8:09] used 720p footage as input, it actually
[8:11] produced sharper results than our
[8:13] workflow could produce yet. But as you
[8:15] can see in this shot, for example,
[8:16] especially with the grass, it's also not
[8:18] perfect. So when should you use our
[8:20] workflow? It shines in situations where
[8:22] perfect accuracy isn't your top
[8:24] priority. For example, when you're
[8:26] upscaling AI generated videos or videos
[8:28] with very poor quality. In these
[8:30] scenarios, our workflow outperforms
[8:32] FlashVSR because it can actually
[8:34] generate new plausible detail instead of
[8:36] just trying to preserve what's barely
[8:38] there. I hope you enjoyed this video and
[8:40] found these workflows useful. If you
[8:42] create something with it, feel free to
[8:43] send it to me or tag me in your work. I
[8:45] always love to see what you come up
[8:47] with. Thanks for watching and thank you
[8:48] to our lovely Patreon supporters for
[8:50] making these videos possible. See you
[8:52] next time.
[9:07] Hey.