Since tomorrow is Valentine's Day, thought I would get myself all nice and proper and looking like a regular Vin Diesel to celebrate. On the 18th is my birthday so that's two reasons to celebrate for me this month:

I find myself spending way too much time playing with AI generated art just at the moment (see current avatar) for illustrations -- it has been described as a gacha game, where all you pay is time, and you still get those dopamine hits when you get close to what you were after.

At the current state of the art, it's fine for things where the mind's eye is vague -- some character art, or incidental scenery



but if there's a firm idea in mind, it can be frustrating; especially when minor changes to art style, or description of clothing or background can both fail to materialise, and cause unexpected side effects, like sudden NSFW failures, and, in making further changes to recover, changing a previously stable subject appearance



or pose -- this set shows various results of trying to force some background scenery into the avatar image



In sum, my recommendation is that it's probably best not to go there.

Over a year and no posts? I guess everyone uses dedicated social media for their pictures these days. Anyhow, last autumn I took a Japanese Language and Culture evening class, which was interesting, not just for the time when I and another "volunteer" got dressed up in kimono by Hiroko-sensei



The course was mostly minimal tourist Japanese - buying tickets, ordering food, hailing a taxi sort of thing - so more vocabulary, but less grammar than I had hoped for. It also showed quite how much effect 20-ish years of subbed anime can have - making it easy to echo back much of the common formality, for example, in the spoken parts of class, when everyone else was stumbling over romanizations like "onegai shimasu". So when people say you can't learn Japanese watching anime, the correct response is that you don't learn much Japanese grammar or vocabulary, but you can learn how it sounds.



To set the record straight and update on my pontifications in the previous post above - I had been prompting for under a week when I wrote that, and things have changed greatly in the past year.

For anyone living under a rock, the Bing image creator is a (currently) free access to the DALL-E 3 model, which can do interesting things like


Prompt: "a girl in ragged clothes, with an umbrella in the rain, looking at an ivory tower, drawn in the style of iridescence/opalescence style"

with the limitations that your prompts get filtered on the way in, and the images filtered on the way out, and you only get a few goes each day.

For the enthusiast with a modest graphics card (NVIDIA much preferred), there is also the option of running with local models, which while perhaps not quite as broadly capable as the web ones, have the advantage of permitting much more individual control. Purely by prompting, images can look so much better than the embarrassing ones I posted above



but it is also possible with local models to take existing images (in this case one I generated late last October) and give them a finishing coat of paint or two, either directly at the image level, or through the generation process



including providing reproducible faces for specific characters.

I'll shut up now, so if you're wanting for more, find me on 𝕏

---

Also, for the record, I'm rather grateful that (despite some stirring on my part) we've not seen the fanart threads drowned by AI Asuka posting, even though she is one of the most well known (to the models, including Bing/DALL-E) characters. Alas, Ritsuko is not so known, and cannot be prompted by name.