Learning the Hard Way: Why I Switched from AWS to RunPod for AI
A creator’s perspective on finding the right infrastructure for image and video generation.
Dear friends,
I’m wrapping up some open website projects these weeks, and I’m looking forward to diving back into AI. Yay!
Currently, I’m developing a custom AI image and video generation tool and learning a lot along the way. Some of this, I’d love to share with you here.
At home, I have a powerful computer that can handle almost all my experiments. But the tool I’m building requires a computer that’s always online and equipped with a strong graphics card (GPU).
For those unfamiliar: Amazon makes most of its profit not by selling things online but by providing online infrastructure through AWS (Amazon Web Services). AWS offers everything from massive cloud storage and AI computing to controlling satellites. Basically, 31% of the internet runs on Amazon (source).
AWS provides everything needed to run AI models for image generation online. However, I found their system overly complex. I spent weeks trying to set up Stable Diffusion on AWS in a way that’s performant and supports multiple users simultaneously—sadly, without success. I believe this setup could work for large-scale projects where money isn’t an issue, but for me, AWS turned out to be overkill for my AI image and video generator.
BUT, there are alternatives that suit my needs much better. Services like RunPod or Baseten allow you to run various AI models online without the need for a complicated setup. You can use them for image or video generation, as well as for running large language models (LLMs).
Their interfaces are surprisingly easy to use—especially compared to AWS. The pricing is transparent, and you can effortlessly switch between strong, medium, and weaker GPUs depending on your use case.
One feature I absolutely love is the ability to keep your online computer on standby. This way, it doesn’t consume resources or cost anything when idle. When I send a request to generate an image, the computer wakes up, does the job, and then goes back to sleep.
If you wanted to program this functionality in AWS... good luck! It’s totally possible, but not for me.
Well, that was a little dive into the technical world of AI—and we actually did not even get much technical.
Thank you for reading!
Much love,
Marius
