Won’t users remember better if there are added images to each vocabulary word? Maybe image that ties with mnemonic somehow?
We have extensions where you can see AI generated images users have submitted, but those change and depends on extensions not breaking. Would be better if images were built into WaniKani.
Shouldn’t be too hard to implement with all the AI image tools available.
I am by zero means an expert on this topic, but I have worked with photos quite a bit in the past and have some sparing experience with photo implementation/layout in a technical context like this. Undeniably an interesting concept and I think mnemonic options are a good thing- but I could and do see quite a few issues with this at large.
The two biggest flaws come from the under-the-hood nonsense most (quality) photo implementation requires. Even when not doing drastic and visible changes, usually pictures need individual editing for proper pixel density, cropping, formatting, etc. If not done correctly it can easily lead to issues with display, upscaling, etc. Photos also take up a pretty considerable amount of both hosting space and processing power. The user experience could suffer loading deadtime, the hosts could have to adjust a lot behind the scenes to account for it, etc.
If this was, for example, a singular web article- sure, it’s not a difficult thing to circumvent. In the context of WK, though, all of that process is being multiplied by nearly 7000 to account for the vast vocabulary. I think those issues would compound pretty fast and just on a technical level it would be quite a bit more difficult than it seems, to the point of impracticality.
I also think that the vast majority of the WK userbase (myself included) would be pretty disappointed in AI generated images being used. Any chance to avoid the environmental toll, creative dilution, and job scarcity they bring is a plus- so generating quadruple digits isn’t particularly favorable to most of us here- and I think the alternative of high quality illustration and/or photography at this scale is not practical either.
Besides the technical complications- I do also have to wonder what exactly would be gained from this?
I’m having a hard time seeing the value added to “橋” having a picture of a bridge under its description- at least that just reading the definition couldn’t express already? How would abstract concepts like grammar elements (形容動詞, as an example) get a visual representation- let alone one that feels additive and descriptive enough to help reinforce it? ((Not to mention I doubt WK would be thrilled to supply us with visual representations of its more raunchy, violent, or taboo vocabulary)). Ultimately I feel like that brain power spent memorizing and creating associations is more aptly spent reinforcing the kanji itself, truthfully, rather than making a secondary association to bounce between if the goal is comfortable reading.
Again, I always will see potential and advocate for user options and learning variety as to accommodate as many as possible- but in this instance I do think it is a pretty huge undertaking for an ultimately small tangible reward. In my opinion (as someone who does work and learn fairly visually), I would prefer if we could upload our own photos in the notes tab- as an example- so we could have an easy way to refer to our own handcrafted mnemonics via illustration, calligraphy, or however it manifests if so desired.
Of course my perspective and ideas aren’t one-size-fits-all, but everything considered I don’t see this idea really coming to fruition beyond userscripts like you mentioned, especially since WK seems to be pretty conservative with its feature implementation at large.
Seconding this. I would NOT be impressed.
Also, WaniKani/Tofugu already has an artist on staff. We just gonna kick her out on the street?
I think the gain is that many (if not most) people recall images better than words alone. Indeed, having an image of a bridge made from hashish would be more memorable and allow more efficient recall.
Yes, bandwidth is an issue. Maybe vector .svg images can be used somehow to save on that? Though honestly AI generations sound like the best solution. Hiring artist for 7000 entries would be too much.
((Not to mention I doubt WK would be thrilled to supply us with visual representations of its more raunchy, violent, or taboo vocabulary)).
They should. You memorize the raunchy and violent ones the most. The things that stand out are easier to remember than bland content.
No, make her work overtime 
Putting aside that I personally don’t want to see sexually graphic imagery, who’s implementing the age restrictions? I can think of at least a few under 18s on the forums, does wanikani really want to be responsible when (not if) a minor’s parents has issues with the images?
what a massive eyesore that would be
If WK went the easy route of using AI, instead of having their artist make images slowly over time for words where images make sense, I for one would welcome the ability to turn off this feature in WK and the third party apps.
This is probably the best compromise for the mass undertaking. They could have their artist prioritize images (where it makes sense) for the first few levels, where new Japanese will benefit the most, and let us upload pictures (with size restrictions and/or automatic resizing and compression) so those who wish to have a vetted image can have their own even for cards that have not received an official image yet.
Though I would be concerned that the cost of storage and bandwidth, even if size limits and compression limits are used, may be too much of a problem. There are a lot of people on lifetime, so they won’t be helping to offset costs. This could result in having to increase for people not on lifetime and increase the cost of lifetime itself even while on sale.
This is more or less what is behind https://www.kanjidamage.com/ (minus the AI part) you can use it if you find it useful. No need to change WK.
I wrote a program that generates flash cards for words using AI to produce the audio and images. From experience, I can say that it takes a bit of intelligent effort to get a prompt to produce something useful. Ideally you’d want to reinforce the mnemonics - and probably therefore use the same characters and memes introduced across multiple words. That’s harder to automate - doable, but requires some thought.
As an experiment, I pasted in the WK descriptions and mnemonics for 申す and then 申 (because the former really builds on the kanji and reading mnemonics of the latter.
This was my prompt and the result. (The reason for the anime styling is because I’ve found its a good way to get consistency across sets of images). This was using OpenAI’s GPT Image 1.
Generate an image in an anime style with no text that would accompany the text below to help me to remember the vocabulary definition described.
申す
To Say Humbly
transitive verb, godan verbThis word consists of kanji with hiragana attached. Because the hiragana ends with an う sound, you know this word is a verb. The kanji itself means say humbly, and this is just the verb form, to say humbly. You can also call this to say, just remember that it’s humble!
This is a very humble way to say “to say” so it’s used in especially formal situations, and only to talk about yourself or your associates. Because of its formality, it’s almost always in the ます form, which is 申します.
Also, the kanji definition used a mnemonic it would be probably useful to work in to the image:
Mnemonic for the kanji, which is a cross radical and a mouth radical: Mnemonic
There’s a cross in your mouth, so the things you say are very pious. Everything you say, you say humbly.Hints
Imagine yourself with a cross in your mouth, saying all kinds of humble things. “Oh no, I’m not that good.” “Oh no, my mustache doesn’t look that great.”And the mneumonic for the reading of the kanji:
Mnemonic
As you’re saying all these humble things, you’re also mowing (もう) your neighbor’s lawn. Your neighbors love you so much, because you’re so humble, and you mow their lawn for them.Hints
Imagine mowing your neighbor’s lawn while you say humbly how bad you are at mowing lawns. > What a good neighbor you are.
Result:
I also agree it would burn some planet - but probably a lot less that lots of other people doing it individually; and sadly you just aren’t going to stop that happening. I think the best hope here is for tech to improve to reduce the energy cost of AI - pandora’s box is opening and you are never going to close it.
I also definitely agree images are quite personal, and I generally don’t like leaning into other people’s ideas of images for remembering things. I would probably turn this off if WK had it in favour of my own images. Would be nice to be able to add them, but should WK pay for storing thousands of images? Just being able to add a link to a personal image store would be nice, and a plugin can do that nicely.
I think this is a great idea. I’d definitely find some of my own images to upload, ones that resonate with me the most. Right now we can only add text comments.
I would hate AI generated images, that is a LOT of infringed copyrights and a lot of energy wasted. In the first place, I don’t understand how adding an image would help much. If say, the word is “yellow”, would you need a whole image to remember the color yellow? It seems redundant and unnecessary.
Another thing to consider that, not all words are tangible and can be explained by a single image. On top of that, people’s interpretation of a single picture can differ significantly. Putting images for can actually be misleading or confusing as the image would oversimplify the word definition(s).
