Cover girls
Because it’s a day ending in “Y”, Github is flaky again.
(I originally posted this on xTwitter to see how many people see my mostly-reply-only account: two weeks, 5 views; this was a pretty trivial K2T render, since it knew the character and costume)
A Canticle For Degurechaff 2, fin
Tanya did indeed get to fuck some people up, but not the specific one she set out to, and she got pretty fucked up in return. Then of course she just had to go and trip another flag. The biggest surprise was that she had let Visha in on the secret plan.
Verdict: really, really hoping it doesn’t take as long to get another season going.
Bumpkin & Harem 2, fin
“Dear Amazon, why does it take five minutes to get a Fire TV box to cough up the latest episode of a show that’s on Prime, on my watchlist, and that I’ve been watching for 12 weeks in a row?”
This episode felt like it was at least double-length, it was so packed with material; seriously, I had to check the runtime. Just the final battle with the Bad Pope felt like an entire episode, and then they had an episode’s worth of wrap-up, with just the faintest hint that Our Bumpkin is starting to think of His White Haremette as the front-runner in The Game Of Waifu.
Also, season 3 announced.
Verdict: pretty darn good. The pope fight was much too talky, but I guess zombie-summoning clergy was just the theme of the season, so what can ya do?
(perhaps if Our Age-Appropriate Magical Schoolteacher dressed like this, Our Hero would notice her obvious interest…)
Well, that was pointless…
I wanted to customize my iPad lock screen with the GenAI book covers below, picturing a kind of “this is what I’m reading right now” look, with the image changing each time. TL/DR, you can’t do that.
First, if you select an album as the source, then you also turn on automatic zoom-to-detected-subject, so it won’t display the entire image. This is not optional.
Second, the lock screen must display the date and time, and it must be displayed at a large size, centered, in the upper third of the display. You can try to make it mostly transparent, but you cannot make it go away or move it to a less-intrusive location.
Third, it keeps a very aggressive cache of the images you’ve tried to use in previous attempts, including zoom and blur effects that it applied without asking. You have to completely delete the images from the device to revert these hidden changes.
Apple knows what’s best for you. Always.
Series silliness with Qwen 2.1
I really liked how the not-Nancy-Drew book cover came out, so I took
the prompt template and applied it to a batch of my GenAI fantasy
battlemaidens, with titles generated from the images by gemma-4-31b,
using a simple prompt template:
Analyze this image and use its significant elements and themes to compose 8 distinct titles for a genre fiction novel. Output only the titles, one per line, without any number, formatting, or Markdown.
I thought 90% of the titles I got out of it were uninspired and
repetitive, but some of them at least drew on elements present in the
images, so I was able to combine them into a “series formula”. The
author name came from one of my really old scripts that does weighted
random selection based on US Census data. Also, I got frustrated with
my current prompt script’s habit of locking up and knocked together a
title-specific one using the openai Python SDK instead of LM
Studio’s native one.
(generating each image took ~13 seconds with Krea 2 Turbo; adding the titling with Qwen 2.1 (which also often adjusted the position of the subject and generated new background around the edges) took ~40 seconds)
Comments via Isso
Markdown formatting and simple HTML accepted.
Sometimes you have to double-click to enter text in the form (interaction between Isso and Bootstrap?). Tab is more reliable.