Hello everyone! Once again, we're discussing the most popular
topic on my channel: how to use a stable diffuser without a powerful GPU, preferably for free
and with automatic 1111. Unfortunately, it seems that you only want
me to create these types of videos because my latest video, which was quite challenging
to make, to be honest, about samplers, how they work and how to choose the best one,
that video wasn't popular. Probably, it wasn't as good as I might have
hoped. If you'd like to check it out, you're welcome
to do so. As for today's topics, we're going to discuss
possible options for using a stable diffusion, including using automatic 1111 in Google Colab. Unfortunately, all possible methods are currently
banned. However, you can use automatic 1111 with a
Colab Pro subscription. Hold on, adepts of forever free software,
I have something special for you. The 'Very nice UI for stable diffusion Invoke'
was broken but now I fixed it, and it's still available in google colab. So please watch this video until the end and
it's actually the best way to support my channel and show your gratitude for my work. Back to the colab topic. There were some rumors that using stable diffusion
with automatic1111 may result in a ban, and that's actually true. Your Google Colab account could be banned,
but this doesn't apply to users with a Colab Pro subscription. So, if you want to buy a Colab Pro subscription
for $10 in the smallest tier, you can easily use automatic in Google Colab.In fact, I've
created a Colab Notebook for that, available via the link below and in this video. But the first question when you decide to
buy a Colab Pro subscription is: what does it actually mean? "compute unit per month". It's actually quite challenging to explain
because it strongly depends on how you use your GPU for generating art. But in the case of a basic Tesla T4, 100 compute
units equate to about 50 hours of using Automatic 1111, which sounds like a good deal to me.In
addition to the monthly subscription, you can still purchase 100 computational units
as a one-time option, and they will also be valid for three months, which is quite convenient. However, in my opinion, the best option is
to use Colab Pro for $50 per month because you'll not only get 500 units but also an
additional 400 units that expire in 90 days. So, what does that mean? You can just stop your subscription after
buying one month and have nearly 1000 compute units for 90 days, which translates to about
450 hours of using Stable Diffusion, all for $50 – it sounds like a good deal. At the same time, you can utilize more powerful
GPUs for training, run multiple sessions, and even continue background execution with
a closed browser. But it's better to start by checking how many
computational hours you can get for $10 before making such a decision. As I mentioned earlier, the use of units can
vary greatly depending on your appetite. As I said before, I've created a Colab Notebook
with the latest version of the WebUI. It's accessible under this video.That notebook
is quite easy to use. Just open the link, run the first cell to
install the requirements, and provide your link to download your model. In step 2.1, Paeste links with LoRa models
(up to 5), then run step 2.2 to install the control net. In the next step, provide the link for VAE
if needed, run Stable Diffusion in step 3, and then follow the Gradio link. It's that simple SageMaker StudioLab is another method that
allows you to use stable diffusion for free with minimal limitations, offering four hours
per day, which is quite appealing. I've already made a video about it, and it's
still functional. However, it's worth noting that some people
have encountered various installation problems, but most of these issues have been resolved
or explained in the video. You can watch that video and use the method
for free. Nevertheless, there is one problem to be aware
of, and that's the availability of GPU. Since this video became quite popular, the
number of people using SageMaker StudioLab for free has increased, resulting in fewer
available GPUs. As a result, there might be some waiting involved,
and you may need to refresh and wait for GPU access. In my case, it works relatively well, and
I don't have to wait too long. I typically get a GPU within about 15-20 minutes. There is another option for Automatic 11.11,
and it’s Kaggle.com, which provides you with insane free 30 GPU hours per week. Unfortunately, Kaggle bans all solutions for
Stable Diffusion, and surprisingly, my solution has also been banned, along with my account,
for some reason. However, I’ve created a new notebook that
still works perfectly, allowing you to access Automatic 11.11 in Kaggle through this notebook,
available on my Patreon page. To be honest, I would be happy to publish
a notebook for my subscribers here on YouTube, but I know that, in this case, this method
would be blocked very quickly, as they are monitoring and blocking everything, even including
accounts. For those who prefer to use Colab for free
and don't want to purchase anything, I have a solution for you as well. My previous method, InvokeAI, still works. It may have been broken for some reason, but
I've fixed it, and you can continue to use it for free without any limitations. This method is also straightforward. Just open the link, install the requirements,
configure your YAML file, or download any model through the provided link. I added this step because many people encountered
issues with the previous method. Proceed to step 3 for further details, or
refer to the previous video for more in-depth information. For those who may not be aware about Invoke,
I'd like to emphasize that InvokeAI is a valuable tool and a strong competitor to Automatic
11.11. It's worth considering, and in some cases,
it's even better and more convenient. So, I recommend giving it a chance and exploring
what InvokeAI has to offer. Don't forget it's free in Colab, so, in some
sense, you don't even have any other option, unfortunately. However, I'm working on creating my own UI
for stable diffusion, which will likely solve this issue. That’s all for now. Please hit the like button and leave your
comments. I'm very grateful to you, those very 10% who
watch my video to the very end.