We're using DALL-E 3 to generate mnemonic images for every item on WK!

For the last 14 months we’ve been steadily working to create mnemonic images for every radical, kanji, and vocabulary on WaniKani. So far we’ve created 855 of them, covering 9.3% of all items, and we’re going to keep going! With the release of DALL-E 3 we’re now able to create more complex and better mnemonic images, more easily than ever! Original post:

Install the userscript to see our mnemonic images: (userscript)

Join our Discord to help us create mnemonic images: (Discord invite)

Here is a sample of some of our latest creations made with DALL-E 3:

乳 (milk): “Giving cleats to a child carrying an umbrella is the best way to gather milk. You don’t have any containers, so the child has to hold milk in the upside-down umbrella. The cleats are so the child can run back and forth to transport the milk around.”

濡 (wet): “If a tsunami hits, it’s raining, and you’re outside trying to use a rake, you’re going to be very wet and damp.”

撮 (photograph): “You put your fingers and an ear on a stool in the sun so you can get your photograph taken. This is a glamour photo, so you have to be in one of these weird, cheesy poses for it to come out jusssst right.”

飾 (decorate): “When you eat you want to see the shape of your gun from under a towel, so you decorate your house with them.”

19 Likes

I think that’s a fantastic idea! Will definitely add to my user-scripts! Thanks!
Visual mnemonics really help me!

sailor-moon-feels

1 Like

That’s a very cool idea. I don’t use the wanikani mnemonics myself but for those who do it seems like it would be a valuable addition.

Actually maybe I should force myself to use them with this extension to see if they work better for me.

2 Likes

Love this idea, well done!

Guess the kanji? I tried to get Dall-e 3 to put the kanji in the picture, with the main meaning, and the onyomi kunyomi, but really couldnt get it to. Anyways, fun stuff, good luck with the project.

2 Likes

That’s a great one! That’s definitely 検 (examine) and you covered both meaning and reading quite well. We don’t have an image for that one yet, could I add yours? Also if you want to contribute more images directly to the project, please feel free to join our Discord server. :slight_smile:

Absolutely, all yours!! Ill check out the discord. Ive always wanted to have imagry for the mnemonics and glad this is in motion!

Could someone remove this one for 構 please? It’s unplesent to look at and doesn’t suit the mnemonic well

Imagine こういち setting up a care station. You hope you don’t get in trouble in the water so you don’t have to see how good he is at caring for people.

Thanks!

I’ve updated the images for meaning and reading for 構 with new ones that I generated with DALL-E 3. See https://www.wanikani.com/kanji/構 - you may have to hit CTRL+F5 or SHIFT+F5.

If anyone else spots an image they don’t like, feel free to ask me to update it here, or on the Discord server (Discord server invite link: WaniKani Image Mnemonics).

1 Like

Ahhhh SO much nicer. Thank you!

I am absolutely loving the idea behind this as I am very much a visual learner so this is very welcome. Certainly looking forward to whatever nightmare fuel my eyes may gaze upon!!

I’m glad you’re enjoying it! There’s still a lot that isn’t covered but we’re working on it every day! Also I’ve never seen that one for 園 and it is both perfect and also horrifying, I love it :smiley:

1 Like

I love this script. As a visual learner it has totally helped me with WK. But… it’s really suffering from being ahead of it’s time. A lot of the images have that heavy early-AI scrapbook look. Also tbh the ones with AI people in them are kinda unpleasant to look at. Maybe it’s because I do WK first thing in the morning but it really puts me off.

As I said, I really love this script and I hate to give negative feedback. But my suggestion would be rather than pushing to get every item in WK finished, take a beat and replace some of the people images with modern AI imagery.

Thank you for the script regardless! :pray:t2:

.

1 Like

You are absolutely right, these were all generated with older AI systems and I think they should have avoided photographic style in light of its limitations with anatomy. I think revisiting these with DALL-E 3 and (where necessary) going for more cartoony styles will help avoid a lot of the unpleasantness. I’ll plan to comb through them and identify candidates for replacement and get them fixed. Apologies for the unpleasantness and thank you for highlighting these examples!

3 Likes

Gotta go anime-style. We’re learning Japanese here, after all. :slightly_smiling_face:

3 Likes

Thanks a mill, much appreciated!

1 Like

Imo I find “drawing” style generated images involving fantasy imagery to be a whole lot more intuitive over the more realistic images oh i just love the squid one sticking to the tree it really makes an impression on me

Can you guess the kanji ? :smiley:

2 Likes

降! Nice :slight_smile: I’ve been using your images and they are hilarious and really helpful

1 Like

Can I add this one to our database? :slight_smile: We don’t have one for this kanji yet.

Btw if you have more to contribute feel free to join us and add them directly on the Discord server:

2 Likes