---
title: 'STOP Paying for AI Upscaling - This FREE Tool is BETTER [ComfyUI Workflow + Tutorial]'
source: 'https://youtube.com/watch?v=NpNagmQI4yg'
video_id: 'NpNagmQI4yg'
date: 2026-06-14
duration_sec: 555
---

# STOP Paying for AI Upscaling - This FREE Tool is BETTER [ComfyUI Workflow + Tutorial]

> Source: [STOP Paying for AI Upscaling - This FREE Tool is BETTER (ComfyUI Workflow + Tutorial)](https://youtube.com/watch?v=NpNagmQI4yg)

## Summary

This video presents a free, local AI video upscaling workflow using ComfyUI that outperforms paid tools like Topaz. It uses generative upscaling to add detail to low-resolution footage, supporting up to 4K output. The tutorial covers installation, model selection, and settings for optimal results.

### Key Points

- **Free Local Upscaling** [0:00] — Upscale footage to HD, 2K, or 4K locally using generative AI that adds new information to pixelated videos.
- **Workflow Based on Wan 2.2** [0:37] — Uses the Wan 2.2 video model (max 720p, 81 frames) extended via a custom ComfyUI workflow.
- **Installation Steps** [0:53] — Download JSON file, drag into ComfyUI, install missing custom nodes via ComfyUI Manager, then download models (base model, LoRAs, CLIP, VAE, upscaling model).
- **Model Selection by VRAM** [1:23] — For GPUs with 24GB+ VRAM use FP8 model; for lower VRAM use GGUF compressed models (e.g., Q5 for 12GB, Q8 recommended).
- **Performance LoRA** [2:27] — Use light_x2 LoRA to reduce step count and increase performance; download low-noise version.
- **Optional Realism LoRA** [2:42] — Stock photography LoRA pushes style towards realistic video, recommended for realistic footage.
- **RunPod Cloud Option** [3:18] — For less powerful computers, use RunPod template with pre-installed workflow; costs ~$0.89/hour for a 5090 GPU.
- **Upscaling Settings** [3:47] — Set frame load cap (max 81), output resolution (e.g., 2K), enable higher quality for smaller upscales, adjust creativity slider (low for precision, high for more detail).
- **Batch Processing** [5:07] — Workflow splits long videos into 81-frame batches and blends them; overlap frames prevent seams. Enable fallback setting to save individual frames if stitching crashes.
- **Results Comparison** [6:41] — Generative upscaling adds significant detail (e.g., hair). Higher creativity improves quality but may introduce tiles. Outperforms Topaz and FlashVSR for low-quality inputs.

### Conclusion

This free ComfyUI workflow offers a powerful alternative to paid upscaling tools, especially for low-quality or AI-generated videos. It requires some setup but delivers impressive results with customizable settings.

## Transcript

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