WEBVTT

00:00.870 --> 00:03.480
-: All right, let me walk you through Comfy UI.

00:03.480 --> 00:06.210
So, we're using Google CoLab to run this,

00:06.210 --> 00:08.250
because to be honest,

00:08.250 --> 00:10.680
I've struggled to get it to work on my Mac,

00:10.680 --> 00:12.360
there are some issues there,

00:12.360 --> 00:15.390
and it's built really to work on Windows.

00:15.390 --> 00:18.990
Using Google CoLab you can run it on its own separate system

00:18.990 --> 00:20.700
with an Nvidia GPU.

00:20.700 --> 00:21.840
I would recommend, by the way,

00:21.840 --> 00:23.820
you upgrade to CoLab Pro.

00:23.820 --> 00:25.890
I haven't for this case,

00:25.890 --> 00:28.620
just because I'm just running the most basic model,

00:28.620 --> 00:31.050
the Stable Diffusion 1.5,

00:31.050 --> 00:33.660
but if you wanna run Flux, or some of the other stuff,

00:33.660 --> 00:35.940
it's a Stable Diffusion Excel,

00:35.940 --> 00:37.620
then you're probably gonna run out of memory,

00:37.620 --> 00:39.780
and will need to upgrade.

00:39.780 --> 00:42.360
You go into this CoLab and it's pretty straightforward,

00:42.360 --> 00:46.863
you just run this thing here to set up the environment,

00:48.360 --> 00:52.380
and just create some space, and all these things.

00:52.380 --> 00:56.370
So, if you look here, it's ComfyUI folder,

00:56.370 --> 00:58.680
and it's got a different,

00:58.680 --> 01:01.410
it's gonna download the models and stuff into here.

01:01.410 --> 01:02.790
So, this is what you would have

01:02.790 --> 01:04.380
locally on your computer as well

01:04.380 --> 01:06.690
if you were running it locally.

01:06.690 --> 01:08.970
And that seems to be all done.

01:08.970 --> 01:11.280
And then you wanna choose which models you download.

01:11.280 --> 01:13.560
So, in this case, for example,

01:13.560 --> 01:17.010
it's downloading the Stable Diffusion v1.5,

01:17.010 --> 01:20.010
and then it's gonna save that in the models/checkpoints.

01:20.010 --> 01:24.990
You could uncomment this, so. (voice clips out)

01:24.990 --> 01:28.740
Press Command + /, if you're on a Mac,

01:28.740 --> 01:31.200
and then it would download that model as well.

01:31.200 --> 01:34.170
I'm just gonna get Stable Fusion 1.5.

01:34.170 --> 01:36.570
And then we also just need the VAE,

01:36.570 --> 01:39.450
which is another kind of model that goes into models/vae/,

01:39.450 --> 01:41.430
and that's to decode the image afterwards.

01:41.430 --> 01:43.050
And there's a bunch of them in here there,

01:43.050 --> 01:45.870
and you could add whatever it is you want as well,

01:45.870 --> 01:47.910
like you could add Flux into it.

01:47.910 --> 01:51.717
But just gonna run this, just gonna download the model.

01:51.717 --> 01:53.417
You can see it's downloading here.

02:01.020 --> 02:03.240
And if you were doing this locally, by the way,

02:03.240 --> 02:07.140
you could download it into your ComfyUI folder,

02:07.140 --> 02:09.657
into models, and then checkpoints.

02:09.657 --> 02:11.910
And you can see it's downloaded it.

02:11.910 --> 02:15.270
There's also the vae thing here,

02:15.270 --> 02:17.420
and you can see that's been downloaded too.

02:18.540 --> 02:23.040
Okay, now I'm gonna run this with Cloudflared,

02:23.040 --> 02:26.280
I haven't been able to get the other way to run it working.

02:26.280 --> 02:29.913
So, the other way is with a local tunnel, or CoLab iframe,

02:30.870 --> 02:32.940
and this one worked just fine,

02:32.940 --> 02:34.900
this Cloudflared example.

02:34.900 --> 02:37.263
So, just going to run this.

02:37.263 --> 02:40.710
What this does is it runs it locally on Google CoLab,

02:40.710 --> 02:42.180
but then it creates a URL

02:42.180 --> 02:46.000
where you can actually try this out on cloudflared.

02:46.000 --> 02:47.820
So, that's pretty cool.

02:47.820 --> 02:50.043
You're just gonna see that comes up with it.

02:52.650 --> 02:54.453
Yeah, here we go, this.

02:57.810 --> 02:59.590
And it's still loading.

02:59.590 --> 03:00.923
Just gonna wait.

03:03.060 --> 03:03.960
And what this is doing

03:03.960 --> 03:07.860
is this is running the ComfyUI interface,

03:07.860 --> 03:12.090
but then getting it to load on this URL,

03:12.090 --> 03:14.627
and it takes a little bit of time to spin up,

03:14.627 --> 03:16.823
you can see it's blank right now, it's still loading.

03:17.790 --> 03:18.900
And while that's loading,

03:18.900 --> 03:20.047
let me just talk to you a little bit

03:20.047 --> 03:22.785
about how to get it running on your own local computer.

03:22.785 --> 03:25.463
So, I recommend this guide,

03:26.344 --> 03:28.590
"How to install Flux AI model on ComfyUI."

03:28.590 --> 03:30.360
This is a stable diffusion art website,

03:30.360 --> 03:31.590
so it should work great,

03:31.590 --> 03:33.630
and they walk you through how to do it.

03:33.630 --> 03:35.610
There's a few different things you need

03:35.610 --> 03:38.910
when you're running the ComfyUI.

03:38.910 --> 03:41.430
One is that you need to download the model,

03:41.430 --> 03:42.720
which is what I already showed you,

03:42.720 --> 03:44.550
and that goes into models and checkpoints.

03:44.550 --> 03:47.340
And so, in this case it's this model.

03:47.340 --> 03:52.340
And then you also need to load the interface,

03:53.130 --> 03:54.180
like the workflow,

03:54.180 --> 03:56.160
which is gonna make a lot more sense to you in a second.

03:56.160 --> 03:56.993
Let me see.

03:56.993 --> 03:57.826
Okay, here we go.

03:57.826 --> 03:59.520
So, now this is loaded,

03:59.520 --> 04:01.950
and this is an example of a workflow.

04:01.950 --> 04:03.630
First of all, it loads the model.

04:03.630 --> 04:05.550
We can see, there's just this one model here.

04:05.550 --> 04:08.430
Any model that you add into your models folder

04:08.430 --> 04:09.330
is gonna show up here.

04:09.330 --> 04:13.350
Once you hit refresh, then it will update,

04:13.350 --> 04:16.860
and if you had another model, it would show it in here.

04:16.860 --> 04:17.970
And what these things are,

04:17.970 --> 04:19.710
these are different nodes.

04:19.710 --> 04:23.190
And the first node just loads the model.

04:23.190 --> 04:25.350
The second node processes the prompt.

04:25.350 --> 04:28.200
And then this one processes the negative prompt.

04:28.200 --> 04:29.033
So, this would create

04:29.033 --> 04:32.070
a beautiful scenery nature glass bottle landscape,

04:32.070 --> 04:33.390
purple galaxy bottle.

04:33.390 --> 04:34.287
That's the prompt.

04:34.287 --> 04:37.860
And then we don't want any text or watermarks on there.

04:37.860 --> 04:40.290
And then it initializes the image down here,

04:40.290 --> 04:42.960
and then feeds all that stuff up into the KSampler,

04:42.960 --> 04:46.590
which is where the actual kind of work happens

04:46.590 --> 04:48.630
with StableDiffusion.

04:48.630 --> 04:50.610
Then we have the VAE decoding.

04:50.610 --> 04:53.670
You can see that this is like a little bit more complicated,

04:53.670 --> 04:56.040
it shows the guts of the operation period

04:56.040 --> 04:58.110
compared to automatic fill of 11,

04:58.110 --> 05:00.930
where you can't really see a lot of this happening,

05:00.930 --> 05:03.390
this actually gives you all the different steps.

05:03.390 --> 05:05.130
And then it saves the image at the end,

05:05.130 --> 05:07.590
so I'm just gonna click Queue Prompt,

05:07.590 --> 05:08.850
and hopefully it works.

05:08.850 --> 05:09.690
So, you can see,

05:09.690 --> 05:13.260
it's lit in green where it's currently processing.

05:13.260 --> 05:14.220
Cool, it did all that,

05:14.220 --> 05:16.200
and now it's loading this.

05:16.200 --> 05:18.063
Decodes, and then saves the image.

05:19.889 --> 05:21.330
And here we go. There's the image.

05:21.330 --> 05:23.130
So, we have that saved now.

05:23.130 --> 05:25.740
And if we go back to our Google CoLab,

05:25.740 --> 05:30.440
we should be able to see where the images are as well.

05:56.850 --> 05:58.713
So, if we download this image,

06:01.560 --> 06:04.260
then you can see it's the same one that we saw

06:04.260 --> 06:05.370
in the interface.

06:05.370 --> 06:07.650
So, that's loaded locally.

06:07.650 --> 06:10.440
Now, the benefit of using ComfyUI

06:10.440 --> 06:13.200
is that you do get a lot more control over this,

06:13.200 --> 06:16.099
and this work is saveable and loadable.

06:16.099 --> 06:17.940
So you could go and save this workflow,

06:17.940 --> 06:22.209
or you could add new nodes, or whatever it is you wanna do.

06:22.209 --> 06:24.780
You can kind of load, reset the view,

06:24.780 --> 06:28.320
or kind of load in a bigger set of nodes,

06:28.320 --> 06:31.500
and get to the point where you have a full workflow.

06:31.500 --> 06:33.180
Maybe you wanna generate four images

06:33.180 --> 06:34.403
instead of something else,

06:34.403 --> 06:35.820
or maybe you want a change,

06:35.820 --> 06:38.580
run some tests where you sample two different times,

06:38.580 --> 06:40.140
or maybe for some workflows,

06:40.140 --> 06:43.320
like for some models, you wanna add like, additional steps.

06:43.320 --> 06:44.700
I mean, you can do all of that,

06:44.700 --> 06:46.500
and then whenever you hit Queue Prompt,

06:46.500 --> 06:47.520
then it's gonna run through

06:47.520 --> 06:49.230
all the different steps in order.

06:49.230 --> 06:50.820
And the nice thing about this

06:50.820 --> 06:55.290
is you can go and load these workflows from other people.

06:55.290 --> 06:59.397
So, if we go to top AI,

06:59.397 --> 07:03.427
you can see an example of this where, you can see,

07:06.477 --> 07:07.980
there's a bunch of different things in here.

07:07.980 --> 07:09.540
These are different prompts.

07:09.540 --> 07:10.680
We have the models.

07:10.680 --> 07:14.043
But then you can also find ComfyUI workflows as well.

07:16.260 --> 07:21.260
So, say for example, you wanted to make images like this,

07:21.360 --> 07:24.570
then it has all the code diagram,

07:24.570 --> 07:27.573
and you can see this one's a much more complicated one.

07:28.800 --> 07:32.700
And you could go and download these models,

07:32.700 --> 07:34.350
like they have a custom model here,

07:34.350 --> 07:37.110
like a custom anime model that they created,

07:37.110 --> 07:39.750
and then you can download the nodes.

07:39.750 --> 07:41.100
You can download, here we go,

07:41.100 --> 07:43.140
this version here, download that,

07:43.140 --> 07:47.190
and then you could go back into ComfyUI,

07:47.190 --> 07:48.190
and then click Load.

07:49.118 --> 07:51.090
And then you can choose that from it.

07:51.090 --> 07:54.660
That's quite useful, and adds to a lot of creativity.

07:54.660 --> 07:56.670
People can find recipes of things that work,

07:56.670 --> 08:00.003
and then they can share that with other people.

08:00.003 --> 08:02.760
So, I don't use ComfyUI that much,

08:02.760 --> 08:05.670
because it's a little bit overkill for my use case,

08:05.670 --> 08:08.130
and I think the main people that use it

08:08.130 --> 08:11.010
are people who are doing a heavy creative work,

08:11.010 --> 08:14.790
people who are doing text to video,

08:14.790 --> 08:17.670
or people who are doing other types of like,

08:17.670 --> 08:21.060
hardcore image generation stuff with a lot of custom models.

08:21.060 --> 08:23.640
So, it's worth checking out if you are going deep into that,

08:23.640 --> 08:25.380
it's what most of the pros use.

08:25.380 --> 08:28.530
But I wouldn't say I use it on a day-to-day basis.

08:28.530 --> 08:31.140
I also found that it doesn't work quite well on Mac yet,

08:31.140 --> 08:33.570
which I mentioned, so, hopefully, they'll figure that out,

08:33.570 --> 08:35.883
but it's still fairly early days.
