#I Tried Over 40 Keyboards Last Year... (So You Don't Have to.)

[theartofmany]

"I tried over 40 keyboards last year to find out the best keyboard in 2023. Hope this helps you who want to know what keyboard do I buy in 2023?" From Youtube channel Hipyo Tech: I Tried Over 40 Keyboards Last Year... (So You Don't Have to.) If you are interested in buying a new keyboard, this video can give you a few options for that Enjoy... (Brands and links in the description of the video)

[onepawproductions]

Picking the Next Audiobook Podfic!

The Muddy Princess audiobook is done, and will be released over the next few months. Time to pick out the next one! It's a toss-up between Lady of the Lake by Colubrina, Things in Common by Colubrina, and Ring-Maker by Lithos Maitreiya. What do you think?

Expand for further musings!

The plan was to go with Ring-Maker by Lithos Maitreiya, but I've run into problems with creating the chapter art, on that front. Mainly, that I can get perfectly well-rendered images using "a young teen 16-year-old (Claudia Black1.5) with (pale skin 1.5) (a cloud of long curly dark black hair 1.4)(aquiline nose 1.3) (thin lips 1.2) (dark eyes 1.1)", but the results are not really reproducible to what I think Taylor should look like. And if it's not reproducible, it's not art-able.

I have a series of about 40 images from training an embedding using SD1.5, but the embedding won't work with SDXL, and it was heckin' overbaked, anyway.

So I can use the images from SD1.5 as references in the IP-Adaptor, and maybe that will be enough.

Or, even better, would be using the Dreambooth method to alter weights on a checkpoint model. But the last time I tried that, it completely wrecked my webui! I had to uninstall/reinstall! 10 frustrating hours of heartache. Oh, but there were some tears, that day!

But the goal--the ultimate goal--is to craft a LORA for Taylor. For everyone in Worm, really, since they don't have any counterparts that would be recognized by the LAION dataset. The dream is to create a whole Worm Fanart Suite, so other people (people who can see the art in their head, but lack the skill to bring it to life with their hands), can do what I'm doing (ie, bash their heads into their keyboards learning an incredibly complex software and also programming and coding, all for the sake of devotion to fanart). Picture a Worm Checkpoint Model that recognizes all the major characters, and comes with a LORA pack that includes offshoots and AU and Crossover elements. Something so highly trained that you could even specify "Pre-Powers Taylor", "Goddess Taylor", "Tinker Taylor", "Bio-Tinker Taylor", etc. I know that can be done with meticulous data preparation and tag editing before training.

Ambitious, I know. That would take YEARS of development!

So, I know I have alot to learn about the software before I can get steady, reproducible images of Taylor that match what I see in my head. I mean, sure, I can get zero-effort gens (ZEGs), easily. But that's not really *my art*, anymore, is it? That is the much-benighted Ai unArt. The current subject of tarring and feathering.

Aaaaaand...I'm back to having alot to learn.

I could dive into Lithos Maitreiya's Ring-Maker, and really push myself to figure out LORAs or Dreambooth. Which means, since the Dreambooth Extension don't work with my A1111, that I'll need to learn *yet another software*, ie, Google Colab or Kohya. And possibly start learning another programming language. Or, at least, getting familiar with another coding language. By NO MEANS am I fully fluent in ANY coding language.

Its a bit daunting, to be honest.

I could take the safe road and do one more HP audiobook. Maybe two? Figure out the intricacies that I don't understand, a little at a time.

Or I could just...dive in like I always do! Lol.

If anyone's actually reading this far down (you go, you!), and has an opinion, feel free to comment, or PM me! I'm always open for discussion. Hearty believer in Star Trek style diplomacy, here :)