---
title: 'SeedVR 2.5 ComfyUI Tutorial | FREE AI Video Upscaler to 4K'
source: 'https://youtube.com/watch?v=9kEux1X_I-U'
video_id: '9kEux1X_I-U'
date: 2026-07-28
duration_sec: 603
---

# SeedVR 2.5 ComfyUI Tutorial | FREE AI Video Upscaler to 4K

> Source: [SeedVR 2.5 ComfyUI Tutorial | FREE AI Video Upscaler to 4K](https://youtube.com/watch?v=9kEux1X_I-U)

## Summary

This tutorial demonstrates how to use SeedVR 2.5, a free AI model, to upscale low-resolution videos to 1080p or 4K within ComfyUI. It covers installation, model selection, settings, and compares results with paid tools like Topaz Video.

### Key Points

- **Introduction to free video upscaling** [0:00] — The video presents a free method using SeedVR 2.5 to upscale low-res videos to 1080p or 4K.
- **Hardware requirements** [0:39] — Hardware impacts speed; GPUs with 8GB VRAM can work but results vary. Online cloud options are available for better performance.
- **Installing ComfyUI and workflow** [1:42] — Download the SeedVR workflow JSON from GitHub, drag into ComfyUI, and install missing nodes via the manager.
- **Sponsor segment: OpenArt.ai** [2:22] — OpenArt.ai centralizes AI models for video, image, and voiceover generation; offers a discount code.
- **Loading video and model selection** [3:28] — Import low-quality video; choose SeedVR2 model variant: 7B for quality (requires >24GB VRAM), 3B for lower VRAM. Avoid 'sharp' models for natural results.
- **Target resolution and settings** [5:40] — Default 1080p recommended; 4K may need high VRAM. Batch size should match total frames for temporal consistency.
- **Running the upscale** [7:06] — Click run; first time downloads models. Preview results in the output node.
- **Quality assessment and comparison** [8:04] — SeedVR performs well on faces, low-light, and black-and-white footage. Compared to Topaz Video (paid), SeedVR is free but less intuitive.

### Conclusion

SeedVR 2.5 is a powerful free alternative for video upscaling, especially when used within ComfyUI, though paid tools like Topaz Video offer more polish and speed.

## Transcript

In today's video, I'm going to show you
a completely free way to use AI for
upscaling low-resolution videos, old
clips, and movies, turning them into
1080p or even 4K videos with sharp
details and high-quality results. Now,
it's worth mentioning that there are
already plenty of other video upscaling
tools [music] out there. However, most
of those tools come at a high price or
require subscription. So, instead, we're
going to use a completely free upscaling
model called SeedVR 2.5. I'm going to
show you exactly how to use it, test its
output quality and upscaling speed, and
see how it compares to other paid tools
[music] in the market. Before getting
started, it's worth mentioning that the
hardware you use will directly impact
upscaling speed and your ability to
tweak certain [music] settings. I will
put my own specs on screen for your
reference. Keep in mind that you can
still use SeedVR on GPUs with as little
as 8 GB of VRAM, but results will vary,
and I will explain this in more detail
later in the process. Now, if you do not
meet the minimum hardware requirements,
or if you want to use SeedVR at its full
potential, you can run it entirely
online. There are dedicated websites
that let you run ComfyUI in the cloud
with access to high-end GPUs at
different hourly rates, and I will leave
direct links to those websites in the
description box. With those options, you
can skip all the installation steps, and
you can still follow the exact same
process as running ComfyUI locally. And
if you want to run the workflow on your
computer, you will first need to install
ComfyUI. If you haven't done that yet,
make sure you watch this video first. It
will take about 5 minutes, then you can
come back and continue this tutorial.
Now that you have ComfyUI installed, go
to the GitHub page that I linked in the
description and download the SeedVR
video upscaling workflow. Make sure you
get the JSON file, then drag and drop it
into ComfyUI to open it. As soon as you
open the workflow, you will likely see a
pop-up window telling you that you're
missing some [music] required nodes. To
fix this, open the manager, go to all
extensions, and search [music] for this
node, then click install. It only takes
a couple of seconds, and once it's done,
click apply changes. Wait until it says
extensions successfully installed, then
close the manager, and the workflow
should now be ready to go. But before we
go through the settings, I want to share
something with you. If you are someone
who creates AI content, [music]
you probably already know how scattered
everything can get. Different websites,
new model drops, it's all over the
internet, and you're just sitting there
trying to figure out where to even
[music] use it. And that's exactly why I
want to tell you about today's sponsor,
openart.ai.
It brings everything into one place. You
get access to powerful video models like
One, Kling, LTX, and everything runs
online, so your GPU can finally relax.
I've been using it a lot over the past
couple of years, especially for
generating images and [music] thumbnail
ideas using models like Flocks or Nano
Banana Pro. And using OpenArt, you don't
burn through a ton of credits, which is
huge when you're just trying things out.
You can also use OpenArt to replace
characters in videos, generate
voiceovers, and creating accurate lip
sync. Wow. It's honestly become a solid
part of my workflow, and [music] I
definitely recommend trying it out. I
will leave a link in the description,
and make sure you use my code [music]
MDMZ20 to get 20% off when you choose
any monthly subscription. And now, let's
get back to the tutorial. First, let's
go to the load video node and click here
to import the low-quality video you want
to upscale. For this demo, I'm going to
use [music] this clip of a parrot. I
specifically chose this to see how well
SeedVR recovers feather details and
whether it keeps the blurred background
[music] soft or tries to sharpen it. The
resolution of this video is 360p, which
is pretty low and pixelated, as you can
see, but you can still use the same
workflow to upscale even
lower-resolution videos. Here, you can
choose from a list of model variations.
They are all SeedVR2 models, but they
come in different sizes, they have
different hardware requirements, and
they are suited to slightly different
use cases. So, yes, your choice here
matters, [music] and it will affect
generation speed and even quality. But
to simplify things, the 7B models
generally produce better quality, but
they also require more VRAM. I recommend
using them only if you have more than 24
GB of VRAM. The 3B models, however,
should work well on lower VRAM GPUs such
as 8 GB or 12 GB. Quality might be
affected. In some cases, you will notice
a big difference. Other times, the
results will be very close to what you
get with 7B models. Still, use the 3B
models if you have low VRAM or if you
want to upscale long videos. You will
also notice that some models are labeled
sharp. I do not recommend using these,
since they tend to produce less
natural-looking results. That said, you
can use them if you want extra crisp
edges and more detail, and they can be a
good fit for cartoons or animated
videos. Choosing the right model depends
on many factors, so I recommend
experimenting based on your input and
computer specs. If a model is too heavy
for your machine, you will get an
out-of-memory error when running the
workflow. That said, if you want the
best results, go with the 7B FP16
version. It is the most demanding model
of the bunch, but if your machine can
handle it, it's the one I recommend.
Personally, I found that it hits its
limit on my machine when videos exceed
10 seconds. So, for longer videos, I
switched to the FP8 version. Now, let's
move over to this node. [music] Here,
you can choose the target resolution you
want to upscale your video to. By
default, it is set to 1080p, and I
recommend sticking with this resolution
for a good balance of quality and
upscaling speed. Personally, I did not
have any issues upscaling to 1080p using
the 7B models. I did try pushing it to
4K, and unfortunately, that did not work
on my machine, but it worked just fine
with the same video on RunComfy using a
GPU with 48 GB of VRAM. There are still
a bunch of other settings here that
directly affect output quality and speed
that you might want to adjust based on
your input and needs. For example,
[music] the batch size should ideally
match the total number of frames in your
video. Not only will you achieve better
quality, but you will also get better
temporal consistency. With a small batch
size, you might notice some flickering
in the upscaled video. However, a bigger
batch size requires more VRAM, so there
is a limit to how high you can push it.
This is where understanding the settings
and experimenting comes in. You can read
more about what each of these inputs
does by simply hovering over it. For
simple upscaling tasks, you do not need
to worry too much about the other
settings. But let me know if you would
like me to make another video where I
dive deeper into the settings, and I
will also link some useful resources in
the description box. Okay, at this
stage, we can go ahead and click run to
start upscaling the video. Keep in mind
that if you're running this for the
first time, the workflow will use some
time to download the models before it
starts processing. [music] And real
quick, if you're finding this tutorial
useful so far, please support me by
liking the video and subscribing to the
channel. I make a lot of ComfyUI and AI
tutorials that you don't want to miss.
Once processing is done, you can preview
the upscaled video in this node, and as
you can see, SeedVR did an impressive
job enhancing our low-quality video and
restoring detail. I also like that it
retained the softness of the blurred
background. Some upscaling models overdo
it and try to sharpen out-of-focus
background, which would look terrible.
You might argue that it looks a little
over-sharpened and unnatural, especially
around the parrot's eye, as you can see,
but let's not forget that the input
quality was very low, which makes the
upscaling process more challenging than
working with a medium-quality video.
That being said, I'm confident this is
one of the best, if not the best,
open-source [music]
upscaling model out there. During my
tests, it performed really well with
human faces. It has preserved the
original colors and contrast. I also
found it very useful for enhancing
low-light footage like this, as well as
black and white videos, which is truly
impressive. [music]
And by the way, if you want to learn how
to colorize black and white videos using
AI, check this video out. Another great
use case of SeedVR is AI-generated
videos. You can generate videos at a
lower resolution to save time and money,
then run them through SeedVR to upscale
them for free. Now, I also tried to see
how SeedVR performs against the video
upscaling giant, Topaz [music] Video.
This has always been my favorite
upscaling tool, and I use it all the
time. It's fast, works really well on
all sorts of footage, and gives you
access to a bunch of upscaling models,
both local and on the cloud. It also has
frame interpolation, advanced exporting
settings, and much more. Personally, I
find Topaz much more intuitive. The UI
makes the process easy and simple. You
can preview short segments before
upscaling, and with the right
combination of settings, I'm able to get
more natural-looking results. However,
when you put the price into perspective,
things change. Topaz Video requires a
subscription, and prices start at $25 a
month. For a tool that's absolutely
free, SeedVR stands really strong.
Personally, Topaz Video has become a
crucial part of my workflow, and I will
keep using it, but I also like having
the option of using SeedVR inside
ComfyUI. And for a free tool, I find it
very capable. [music] Let me know what
you think in the comments below. Is
SeedVR good enough for what you need? I
would love to hear your thoughts. Stay
creative, and I'll catch you in the next
video. Peace.
>> [music]