Monday, October 26, 2026

Running the floor heater

I've been playing a little Warhammer 40K : Space Marine II, watching The Sarah Conner Chronicles, and dabbling some more in AI video. There's no ranking here; these are all whatever I happen to be doing when I'm feeling too brain-fogged to get creative work done.

The new novel is crawling along.

So, the entertainment yesterday was doing some re-posing the hard way.

I realized I'd fallen into the tall duck problem; people and other humanoids are tall and thin, the more efficient image format is wide and short. I'd done a lot of renders that were wrong-side up and that had several problems.

So get those characters to sit down or something. If I was paying tokens to a service I'd just do a "Claude, make the character sit down."

But no, I'm staving off boredom between cleaning chores and doom-scrolling by doing this offline. And the look of the characters and setting was specific, which is always a problem with AI. Take, for instance, "space armor." Seen all over visual SF. Try to prompt it with those words, you get medieval plate sneaking in. Tell it "no, more like the Mandalorian," you get a Boba Fett helmet. It will always cleave to the familiar, to the common meaning.

(A big pain for fashion or period clothing as much of this stuff is phrases formed from common words, like pillbox hat, pixie cut, or leg-of-mutton sleeve. Or is a technical term the AI doesn't recognize, like aguilette.)

So the "pipeline," such that it was, was first to take a previous video that had the character in it. Rip that with a simple frame extractor I put together from the parts box. Save an appropriate frame. Seed a new video render with that frame (image-to-video) and try to find a prompt (and a few camera control LORA or similar) that would change the pose or move the character. ("Character sits down. Camera orbits to their left.")

Do the pull-frame routine again and export that to a paint program for some basic clean-up (paint out the elephant that strolled into the render, and fix the Sam Brown that turned into a Klingon sash half-way through).

Run that through image-to-image for inpainting. Which, given the space armor thing above, requires multiple incompatible LORA; this one will fix the details of the NASA-punk background, but this one puts the stripes back on the zebra-man. Multiple rounds of inpainting and cycles through hand painting until it is back on model and upscaled sufficiently.

Now render new video -- and wrestle with the AI again that doesn't get why everyone is sitting down when all the fashion shows and talking head news programs and etc. it was trained on had everyone standing up. (Also an issue in the inpainting stages. Especially faces; the AI loses the ability to recognize faces if they tilt too far off the usual axis.)

The biggest issue in all this -- or possibly the bonus, if you are trying to while away some mental downtime -- is that when the AI drifts, it either goes into, or more likely suggests, an interesting alternative. Instead of one render, you end up with branching forks, multiple renders going in different directions. At all stages of the game; it truly is geometric.


The novel crawls along because I'm still working out how I want to tell it. SF really does rely on telling, because it is too inefficient to show everything. But an immersive narration introduces a problem; why is the character telling themselves what they already know? It's like a one-person version of "As you know, Bob."

When I was working with Penny, a big part of her thing is making connections. All sorts of crazy connections. So if I needed her to talk about Clovis Points, I could show her a fir tree or a straw hat and she'd get there. It took a bit more pushing to get her to recap what she'd been doing for the last couple of days, but really all it took was putting the right thing in her environment to jar the story out of her.

Cadet Sinclair is third-person and doing free indirect speech so there's already a blur between narrative voice and internal voice. But I'm still too wary both of "As you know, Bob" and the lure of the info-dump to want the narrative to go all out on explaining the world.

Anyhow. I'd been trying out different approaches for opening by looking at the giant negative space wedgie (an exotic star) that is both the sciencey-sounding handwave for a lot that is going on, and thematically an eldritch thing. And whether I really do want to explain that much about the thing now, or offload to later scenes.

But turns out there was a different issue. Thematically, Sinclair is part of a society that could be cordially described as "better than the alternatives." Her experience is being in a system, under orders (explicitly, now, as a young officer). Introducing her as a free agent who can stare at a vidwall and have lots of thoughts about the science and tech and social implications of the setting was... not right.

So I've got her entering in a group of crew, and all the rest are heading straight for the bar, and her only look at Oberon is cut short. And that works better.

Now I just have to deal with creating and introducing a bunch of crew while they are embarrassing themselves striding around going "look at us, we got posted to the frontier" and then get cut down by a more experienced merchant crew who has actually been.

Which means, yes, many many many chances not just for "...internal combustion engines had spark plugs, Keri knew, and telephones used to be used to make calls..." but way much "You should know, Bob..."

Friday, October 9, 2026

Voorwerp


I realize I'd been moping. I was going to write a cranky blog entry about not having a character for my new space opera -- when said character finally came together. Cue a mad burst of energy, cleaning house and cooking at home, even.

That is, I think I solved the character issue. I have yet to actually write her. That is, doing something other than looking out a window.

I went back to work on the opening scene. Got  sidetracked into reading about strange astronomical objects, and it turns out there are several things out there which are well known for being green. And one of them glows for no good reason (the prime theory for Voorwerpjes is that they are a quasar ionization echo. As in, a very bright but short-lived object shot them the hell full of energy but is no longer bright and visible itself.  Since these things are "the size of a small galaxy" there's at least 20,000 years to play with in the timing there.

Oh, and there's evidence for "dark matter highways." Related from Atomic Rockets, who are even more writer-friendly these days, so much so that for every possibly plausible bit of science, there's some wild hypothesis that probably isn't going to prove out, but could make a heck of a story.

Anyhow.

That rant would have been about another stupid choice. Like thinking writing an archaeological adventure would be a quick-and-easy book (well -- I did finish in a year). I wanted to do something with used furniture so I didn't get bogged down in world-building and research. So, probably space opera. And I've always sort of wanted to do a Horatio-esque. My trunk novel has a young crew person on an interstellar "shop truck" -- even if her career goes in rather different directions.

Maybe it is just the vibe. Too much Star Trek in my formative years; that comfy split-level bridge, the uniforms, the wonderful toys.

In any case, the field has moved on. Space Opera is no longer a pejorative. The near-future or otherwise Gernsbackian SF has become consciously deconstructive, even played for irony, as that lunar colony (and those jet packs) never quite arrived in the real world.

Still, SF is still more about the world, the science, the tech, and less about the character. Even in the most unabashed Planetary Romance the setting and costumes and fun details are more important than nuanced character or rigorous plot.

Not that I'm exactly ace at the latter two myself. The Athena Fox stories had bare-bones plots and just enough fun character stuff to keep it from being entirely travelogue/history lecture.

Look, I see plenty of stuff out there -- the comfort food of long-runner space opera series -- that didn't sweat the science. But they are still offering to the reader something creative in the world. Maybe too much so; I almost feel I am seeing writers trying too hard with fantastical settings that are just a little too fantastical to quite grasp.


Anyhow. 71,500 words of notes in the Scrivener folder. I just finished breaking out the more germane stuff and the actual useful notes on the Klathi (the primary alien antagonists of the first book) is a mere 1,500.

That is because I use my notes Socratically. I argue stuff out in them, circling around ideas. That bit about "finally" understanding the character? I took down that note three times previous before it finally landed.

Not an efficient method, I think. Not much of a point of using Obsidian to jot it all down, either. Because the core concepts are stuff I remember now. I've never really needed to consult my notes on a character, not for how they behave. Besides, half the fun of these various aliens is how they bounce off each other, and that can't be efficiently notated in a system that compartmentalizes them.

Sigh. I still have to crunch through a bunch of different references and nail down the positions, job titles, and ranks. I got this close to Blue finally meeting the crew of "Tammie," but is that guy really called Chief Jenkins? (It isn't just a name thing that I could use a placeholder while I get through a rough draft. I'm past that point now. What I need is the underlying organization of the ship because half of what happens in the next few scenes is based on that.)