Ergo Ergot
The Apothecary Diaries 3.1, episode 3
It’s Maomao Medical Mystery Time again, boys and girls! Meanwhile, Jinshi is dealing with another facet of the same puzzle, and they haven’t (yet) gotten together to compare notes. How she resolves the human side of the mystery is by offering-but-not-recommending that two girls sell themselves to the brothel, which is sure to offend someone.
Kicked Out Into An Internet Café, episode 2
First guy she meets is a city guard, and he says he used to be an adventurer like her, until he took an arrow in the knee. Sigh. The city he’s guarding? “E-phone”. Subtle, this ain’t. By the way, Our Networked Heroine’s name? “Loner Hermit”.
Sadly, while this toe-dip into the shallows of the meta could be amusing, Our Second Cute Gal explodes into over-the-top shouty action the moment she steps on screen, and never stops. I couldn’t finish the episode.
Not particularly exciting, Facebook…
I will not be participating:
Meta worked with Apple to bring Muse Spark Open to the Mac, and the model can run on a desktop setup with two 512GB Apple M5 Ultra Mac Studios running the latest MLX framework.
(never mind that the hardware isn’t even on the market yet…)
Microsoft discovers performance
Windows 11 will soon work on supported computers. Which is nice, since they force-upgraded millions of them.
Adobe sped up their software, too, but only for RTX Spark.
Flat is justice…
I often use Juan’s large archive of LLM-generated artist styles, which I’ve sliced up into various genres, and I’ve generated a few of my own, both for specific artists as well as more general themes (I’m quite fond of the “classic teen adventure novel cover” style), but I decided to try something a little different:
Create a list of 100 distinct image-generation prompts describing unique artistic styles, starting with the basic medium (photograph, painting, illustration, mosaic, drawing, sketch, print, collage, woodcut, lithograph, carving, relief, textile, tapestry, quilt, etc) and then adding detailed descriptions of how the style is executed. Do not use names of real people or places. Do not include image content or composition, just the style. Do not include unrelated details. Describe only the visible elements of each style in concrete terms, as a self-contained paragraph of 50 to 75 words. Do not number the list or include any text formatting. Do not worry about output length or truncated results; longer results are better than shorter ones.
I ran this a few times against Qwen 3.8 Flash Next running under Strata on my big gaming machine, used the output to generate a bunch of images, and then weeded them down to the ones that had at least a little styling to them. The one that I thought had the most potential was a cut-paper style, so I fed that into an uncensored/abliterated version of Gemma 4 12B, using the structured prompt-enhancement system prompt I’ve taken to calling The New Juan Order. (yes, he’s quite prolific in this area, and fun to tease a bit)
The downside of letting an LLM rewrite the prompt is that sometimes it drops certain elements, like “cut-paper style” or, say, “college-age woman”. Fortunately, I didn’t request any nudity, so the occasional minor was only a minor issue.
I mostly just deleted the ones where it applied the style only to the background and made the human figures photorealistic, but sometimes the style mix worked.
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.