Did the test with these, seems it’s not easy to reproduce
The Japanese image above does verify with SynthID, but detection takes larger content then just a close crop of the text. Note though in the right bottom stroke and the center strokes where they join, unnecessary noise that is not solid white nor like a brush hair:
So: Append to my prompt to see any improvement in the “printing” to be coaxed or guarded against by language (plus vary the font colors)
Text is solid color strokes of antialiased smooth fonts, where each glyph and word is one color choice but each style of face can also have a color change, and has no artifacts that look like ink stamping or grain in the font strokes.
It seems the AI does have control of the noisy starry grainy symptom often seen (quality:medium):
Enhance!
But even early today I note this spoiling of infographic background colors also:
wonder if that is because of that or the 粗、ぶれ、ボケ that is playing a role in the image.
That’s cold, wishing well to others has nothing to do with prayers. It brings awareness. It brings humans in unity. Guess also the “Sending Support” was missing, guess need to read more what is written and not assume. 
