No.75400
>>75399What you say makes sense. But I'm still confused by what I am looking at.
Perhaps it's a weird filter that they attacked some of their shots with. The image you just posted has way cleaner lines than mine (though one of the hair lines has a hiccup as it passes through the eyebrow).
No.75403
>>75401That episode in general was a quality dip from the usual. It wasn't constant, but it cropped up now and then. AI backgrounds then are a symptom of a greater problem, unfortunately.
Kind of interesting that I never noticed that terrible AI text, though.
No.75404
>>75403Probably because filtered photographs were already used as a shortcut and they also tend to have blurry, unreadable blob text. And then there's also the actual human-drawn background text that is also complete nonsense because the guy drawing it was just told to put some latin characters there.
No.75463
I highly, highly doubt it. For things like backgrounds and the like it is plausible, but animating a distinct character, without glaring mistakes, is way beyond current capabilities (or, at the very least, would be more effort than its worth).
If AI generated imagery, beyond just isolated cases of laziness, is to be integrated into anime production, it's going to require the development of specialized programs for it, rather than a simple application of current software. For instance, current models are only trained on static images, not on animations, and likewise all images are generated individually: there's simply no realistic way to get natural-looking motion out of that. Maintaining hairstyles, clothing, colours, artstyles and so forth between frames and between shots would similarly be borderline-impossible (each individually would be doable, but all at once, while also having to actually get the precise positioning you want... yeah, nah).