HeyThereGuy said:
post #8559250 and post #8582542 by roribitanx are tagged as ai-generated.
Earliest post on their Pixiv was posted in March 2020 and their recent art is stylistically consistent with their art posted before AI became common. (Late 2022.)
Pixiv posts have variants with an alternate background and a transparent background as well as the chibi characters on some images posted separately. Indicating that the images have layers.
They post WIPs on their Bluesky.
Several of their images feature danmaku patterns directly based on ones from the Touhou games which I severely doubt AI could recreate.
This feels like a classic case of someone tagging based solely on the art style looking “AI-ish”.
These posts have been brought up here before and tagged based not on the style but an egregious degree of AI artefacing. Marissa's broom has two sides, her hair is merging with the hat left and right, the clothing folds are non-Euclidean (just try to parse what's going on with her right sleeve). All of the artist's post-2023 works are like this. You're right that the general style resembles their 2020 upload (worth noting that if you check the "i" button next to their upload dates, most have actually been reuploaded in 2023-24, but still look human so I'll give them the benefit of the doubt) and they might be using AI to render their own compositions. But it can't be the bluesky WIPs - in the Marisa WIP, for example, all the issues are already present. You could make a case for ai-assisted, but something this messy likely isn't going to go over well in the queue.
Also the artist has a known history of no-ai stance with commercial free-lance work.
There is no history to speak of, the guy popped out of nowhere a month ago and lies out of their ass every second post they make. At first they were "fast because they used to work at a black company", then they "had experience as a lecturer in the Kyoto University of the Arts". Unfortunately, after people pointed out both backstories as not very convincing and shared by a noncoincidental amount of AI fraud accounts, they've now had to settle for plain old designer at an unspecified vtuber company.
To wit, yeah it's pretty obviously painted over AI and the timelapse introduces more questions than it answers, like the already mentioned fully rendered tummy spawning into existence, and why a purported professional illustrator would lie about Krita's timelapse feature not letting you hide layers. They've done some cleanup, but the signs are still there:
post #8769926 - window bars suddenly MIA under her hair
post #8740173 - hair lock on the right gives up on being a lock and becomes a single strand blending with the wall halfway through
post #8831888 - stomach line under the nightgown breaks under her left arm
post #8739930 - ear area, hair blends with shirt in places, blurry glowing window breaks under right arm
post #8739933 - window curtains break under head
post #8724998 - whatever the pillow shape/shadow is doing in the top left, area between her chin and shoulder
I bet those images would've been decently well received on both twitter and here if they were transparent about their AI usage, as plenty of AI-assisted accounts are, but grifters gonna grift.
Edit: fixed web archive links
