Lack of interest
Frieren 2 starts on ThursdayFriday. Next Japan trip booked, with a very
healthy yen-to-the-dollar rate. Honestly, it’s reaching the point
where it’s cheaper to fly there and fill your suitcases than it is to
pay the duty on overseas purchases. Speaking of finance…
“I’m so old…”
“…I remember when a 10% interest rate was high for a credit card, and it meant you had a poor credit rating.”
Activist mindset, unpacked
There’s yet another video of the aftermath of the Good shooting in Minnesota, in which it is alleged you can hear her wife shouting “why did you have real bullets?!?”. I won’t link it, because you cannot clearly see who’s speaking, but it’s certainly plausible, because only an idiot activist completely divorced from reality would ask this question, and she established those credentials in the earlier videos where it was clear she enjoyed confronting the officer documenting the scene.
Right up until she realized they were about to experience a felony arrest for deliberately interfering with law enforcement, tried and failed to get into the car, and yelled, “drive, baby, drive!”.
(and for those who haven’t been keeping up, the released video evidence clearly shows Good blocking traffic on a narrow one-way street for several minutes, having a good time as she dances around in the driver’s seat banging on her horn like a drum, then both she and her wife cheerfully taunting the officer; it also clearly shows her smiling as she shifts into gear and accelerates)
Split-random prompting
I want to inject more background variety into my pinup wallpaper, and the most common method is to pass your prompt through an LLM and give it instructions to enhance.
This has not produced satisfactory results, because every LLM I’ve tried has ended up removing significant details from the prompt. Things like having the women be young and pretty with lively expressions, which kinda defeats the purpose.
My latest test is to generate two separate dynamic prompts, one for the background and one for the subject, pass just the background output through the LLM, then merge them back together. The LLM model is an uncensored variant of Gemma3-12B, recommended by someone on the SwamUI Discord. I’d been using an older variant of the same model, and this one seems to produce cleaner results. I need to tweak the system prompt, though, because it’s adding men to the picture a bit too often. Sometimes that works, but only if she’s still the focus.
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.