1st Year Round-Up!

It’s been a year since I posted my first Stable Diffusion gallery on 23 Feb 23!

I had been playing with the publicly accessible early versions of Mid Journey whilst deployed overseas in ’22 and was having great fun with it. While somewhere in Eastern Asia, I noticed a good sale online and had my brother pick up a liquid cooled 3090ti on my behalf, so I had an upgrade waiting for me on my return in December.

The card’s 24GB of VRAM opened all kinds of “AI doors” for me, ranging from live voice synthesis to locally run chatbots. My favourite, though, has been Stable Diffusion. From custom checkpoints to LoRAs, embeddings and controlnet, the user’s ability to influence so many aspects of the picture generation process is immensely satisfying and (in my opinion) completely humbles any cloud-based service.

Parameter adjustments aside, the greatest aspect of running something like Stable’ locally is the lack of censorship. It’s not even that I want to generate a bunch of hentai or anything. When seemingly innocuous keywords get you flagged on MJ/NJ, you can’t help but feel stymied. A couple days ago, I tried prompts containing “backless gown” and “swimsuit,” and MJ/NJ rejected both as inappropriate. “Standing on the beach wearing a swimsuit” is now beyond PG-13? Gimme a break.

That being said, I continue to subscribe to MJ/NJ’s paid plan. It is fast, works from anywhere, and can be fun if you know how to operate within its constraints. It just can’t really compete with- or replace- a service running on your own rig.

Okay, that was a lot of hot air. How about some pictures? I’ve trawled through my output folders and chosen ones that I find interesting or indicative of the art style I was working on that month. In each gallery below, the first three are from Stable Diffusion and the second from MJ/NJ.

I present to you, dear reader, a representation of my AI efforts during the past year:

February 2023