TL;DR
Just want to run Invoke UI to generate some pics?
Create Account https://runpod.io?ref=c71blwtm
Throw in some cash https://www.runpod.io/console/user/billing
Go to https://runpod.io/console/deploy?template=elr3w646vn&ref=c71blwtm
Choose GPU (e.g. A40)
Deploy
Wait for start
Connect
Create
Delete pod
The long story:
RunPod
If you want to run your own "local" models but don't have the suitable GPU, there are a few sites offering to run "pods", which are mostly containers, i.e. light weight virtual machines, equipped with the necessary GPU hard ware for your GenAI Inference or Training. One such site is RunPod.
The Advantage of running your own pod is full control over the running software. You can completely control whats happening, run whatever UI you like (provided you find a suitable container), train/generate/... And pay per time, not per generation.
Create an Account
If you want to do me a favor, use that referral link: https://runpod.io?ref=c71blwtm to create your account. Makes no difference for you, but I get a bit of free GPU time. ;)
Deploy a pod
Go to https://www.runpod.io/console/pods and hammer that button
Then you end up in a big menu to choose all kinds of things, most notably the
GPU
The A40 used to be really good value for money. Still my favorite one, but you could also start with a cheaper RTX of course.
The next important thing is choosing a
Template
"Template" on RunPod means mostly choosing the container image that you want to run. There is a lot of templates available. Just search e.g. your favorite UI
But make sure to have a look at the README before starting a pod. I prefer to run pods which
Run the tool I want without a lot of other unnecessary stuff like jupiter
Automatically start and expose a web UI for my tool without the need to go here and type that to actually get the tool running
One Template that I like particularly is the "Invoke AI Web". Click here https://runpod.io/console/deploy?template=elr3w646vn&ref=c71blwtm to go to the pod creation page with that template already selected.
Deploy
After choosing a GPU nad pod template, push the button
to get a nice waiting animation...
...well thanks for that.
On the pod overview you will find your pod starting up.
Klicking on the blue arrow to see the actual logs, which will mostly consist of downloading and extracting the container image.
Connect
Once finished, your running pod will look something like this:
Where you can hit the "Connect: Button
To see which HTTP Services are published by your pod. Depending on your template, you can now hopefully just click on that service to get to your WebUI.
This article is not about the invoke UI, which I really like, especially for in-painting. If there is interest in this, leave a comment and I might write another article on invoke. Or just check their page, those guys get paid for explaining their paid SaaS variant. ;) https://www.invoke.com/
I have just written a basic getting started guide for Invoke at
Those two give sime details about inpainting/outpainting for more advanced AI image editing:
A1111
The UX depends a lot on the Template. Another Example:
If you are trying the A1111 template by AI-Dock you can click on any of the service endpoints, enter the credentials "user" / "password" and end up at
go to the A1111 web ui, wait for the container to finish setup
and eventually get your GenAI UI:
Well... if you want that I guess you can run it locally on your CPU. ;) But A1111 is still probably the reference UI. ;)
Terminat0r
The clock is ticking. As long as your pod is up and running you pay every minute, even if those are just cent fractions. So, after you are done with your work, do not forget to
First download your results (make sure to wait until the download is finished)
Then hit those stop/delete buttons on https://www.runpod.io/console/pods
The next Level
Tired of downloading all the models before you can start generating with your pod? I started to use a Persistent Volume which gets mounted into your pod to store my data (models, DB for the GenAI tool, outputs...). This speeds up the startup considerably, although it means a constant money drain. If you want to try i its worth it, just create a Network Volume at the top of the pod generation menu. make sure to select that volume along with the GPU and template, and keep your state, although your pod is re-created.
Disclaimer:
I am not affiliated with any of the mentioned tools/sites, do not get any money from them (unfortunately). I might (hopefully) get some money if you use a referral link (thank you!). The content of this article is not endorsed by anyone and only represents my own limited understanding. I do not own any logos and use those just as references to the original tools/sites.