Sure.
To save time, what i do is dump both wanikani mnemonics on Gemini 3 pro and tell it to give me a nano banana pro prompt that emphasizes the elements + has the kanji somewhere in the photo + has the kana for the reading somewhere (though I sometimes skip that and just put it later myself on the photo in powerpoint).
This sometimes yields perfect results (that i might have not thought of myself), and sometimes very lackluster results. For example, i had 10 or 20 iterations before i got the “sick” radical closer to how i wanted it.
The more elusive the composition, the more difficult it is.
Here are some recent examples.
解 (solve)
A cinematic photograph taken at twilight on a calm lake. A person in a wooden kayak is balancing precariously alongside a cow. The cow is trapped not by ropes, but by a complex, ancient-looking mechanical puzzle lock made of interlocking bronze gears, dials, and glowing blue runes that encircle its body. The person is holding a sword with intense focus, holding it at a very specific, acute angle to insert the tip into a hidden slot on the mechanism. A beam of light is bursting from the lock where the sword connects, indicating the precise angle has solved the riddle and the mechanism is unlocking. The black kanji character 解 is painted clearly on the hull of the kayak.
求 (request)
inside a high-end, luxury restaurant with polished marble floors. A wealthy, snobbish customer in a tuxedo is standing and pointing arrogantly at a spilled glass of water on the floor and stares at the waiter, demanding that he cleans it. A waiter looks humiliated, kneeling on the ground and attempting to clean the water spill using a fresh green cucumber instead of a towel. The cucumber has a produce sticker on it that clearly displays the black kanji character 求.
Regarding the prompt for gemini itself - it pretty much “got the task” almost immediately just by me asking it to help me visualizing the mnemonic by providing me with the prompt. Then I had to nudge it in certain ways e.g. “put more emphasis on the roof” or “this is not good, give me 3 alternatives in completely different settings”.
here’s an example of such a prompt to gemini:
ok, let’s make slight modifications to emphasize the mnemonic.
for (1) - we want to emphasize the MOUTHs. so think of ways to put emphasis on the mouth. we don’t necessarily need many guests. it’s enough that we have one (or a few) guests for whom we can emphasize their mouth in some way. maybe offer several alternatives (a,b,c…) from the most realistic to the most outlandish.
for (2) - the overall setup is good, but we want to emphasize the GUEST meaning, rather than just a “traveler”. offer several alternatives here too. one option is to have it as sort of a hotel (and then we must remember to emphasize ROOF). the other is to lean into the “customer” meaning somehow (again, roof should play a major part of the scene).
once i have a good enough prompt, i just do the iterations/modifications directly in nano-banana, e.g. changing angles, genders, adding or removing elements, etc.
