//=time() ?>
"an archmage walks through a library, hyperrealistic"
"radioactive tyrannosaurus kaiju, ruins of downtown, destroyed city, glowing green fog, panic"
A sea-change in 9 weeks of AI improvement.
"redhead sorceress, torch singer, art nouveau, Mucha"
So, the improvements aren't just in efforts to match reality, other styles are strengthening too.
I suspect I could do a whole line of covers like the new images.
"a knight fights the grim reaper, inkblock print"
@LizardSF I used "A giant brown rabbit with the head and shoulders of a buffalo, buffalo hybrid, front hooves, monster, --v 4" to get these.
In one case, it looks like #Midjounrey semi-copied an artists signature, too.
"military stealth submarine, underwater, unreal engine, sunken ruins"
Again, both Midjourney, 9 weeks ago on the left, today on the right.
Against, testing capacities: "fat man in a suit sitting on a couch, on a lawn, as a political cartoon."
On the left, 9.5 weeks ago with #Midjourney. On the right, today.
One of my test cases on AI image comprehension. "an angry drunk computer tears apart a cabbage, hyperrealistic, covered in gravy, in a kitchen"
The new one is definitely "better," but still very wrong.
"descent into hell, fiery rock, the gods are dead, doom of mankind, apocalypse"
Again, one option is for framed art in the new version.