I've been meaning to get into AI image generation for a long while now but the ones on the main AI platforms like OpenAI, MidJourney, while appealing, isn't something that makes me feel like I had any part in it.
What I want was something running on my local machine, on my own hardware, something that I can tweak around the edges.
To do so, I installed and configured InvokeAI. I'll probably do a more detailed instruction post later as the official one seems to be a little outdated.
There are some things that you need to be aware of - like having the right version of Python, the right versions of certain packages and also equally important, the right version of your GPU drivers.
Here's my first image which took forever to generate:
Why did it take so long? Because my GPU drivers were outdated and so InvokeAI defaulted to using the CPU. Once I had my GPU configured properly, here's what I generated:
The prompt for both were 'a man walking in Switzerland'. I'm looking to install new models and especially ones that have been blocked for being 'unsafe'. That's probably the true potential of running this on your own.