"My, have you heard? Who’d you hear it from? The Rumor of the Anime A.I Image Generator.! With some typing and a click, you can create anime images instantly. It’s a big deal that it’s a big rage. Talk about standalone software”
Yes, it’s true, a version of Mnemonics is now in anime form. This is like the other project using DALL-E, Midjourney, and Stable Diffusion. These images are generated from NovelAi and Waifu Diffusion. I use NovelAI primarily as it’s easy to use, but a paid subscription. The highest plan allows for unlimited generations for normal images.
You can see some examples at the project’s github. You can contribute to it by forking it, comitting it and creating a push request. Note the guidelines in the readme.
If you don’t know how these images are made, this video will explain everything:
I’m curious what people adapt WaniKani Mnemoics in anime image form. The main focus is on the Kanji, for now.
I decided to subscribe just for one month to try it out without actually setting up Waifu Diffusion. My GPU on my virtualized gaming machine is out of date (Geforce GTX 980 TI) and it can’t probably handle it. I probably need to get a RTX 4080 or 4090, which is a future upgrade. NovelAI is the fastest way to get one’s hands dirty.
Commissioning art for every WaniKani mnenomic would be infeasible since there are 2024 Kanji on WaniKani and can get really expensive to commission art for each mnemonic. It’s not worth it if the image is going to be used for education purposes. However, speaking of commissioning, I do plan to commission some anime art for my Anime blog on the character design I came up with in text form, although I generate some concept images, which will become the primary identity of the blog and I have money reserved for that.
AI Images are a controversial topic since it goes in that grey area as the algorithms used non-free materials. I think AI images are okay for brainstorming ideas or for educational use, but not to carry out as a finished product or art. Even the pictures AI generates while getting better is still far from what a human can do, especially in terms of innovation and creativity. AI picture generation is akin to tracing an image or emulating an art style. Artists need to start somewhere before building their own style.
Also, the models themselves don’t actually have the actual images, but what the machine learned from looking at the images.
True, feels more fulfilling to commission artists and get personalized art than ai subscriptions imo
I think thats fair, i just dont trust ai art usage tbh. Had too many artists buddies run into issues with people who use ai and the whole discourse its causing
Yeah, I feel like using AI images for mnemonics is one of the few actual use cases for AI art, since you need such a high quantity of images, it’s unfeasible to commission them individually, and the quality of the images only needs to be good enough for them to successfully serve as a memory aid. Plus, they’re being generated for educational use and not for profit.
It feels to me like it’s the kind of project that isn’t taking any money from any actual artists because illustrating thousands of kanji mnemonics is not something that anyone would be likely to pay for, and no one is trying to sell the AI generated images, and the AI’s biggest weakness (cannot remotely compete with human artists in terms of quality) doesn’t matter as long as the resulting images are memorable.
It’s been a week since I started the project and we have finally hit 205 generated pictures in about a week. Coming up with prompts that fit with the WaniKani is the tough part and some pictures require me to do some Photoshop editing.
My goal is to probably get close to 1000-1500 by early November before I take a break, as in cancel NovelAI subscription until there is something interesting or I need to generate a bunch of pics for reference materials for commissioning art. But coming up with the prompts and fitting the mnemonics are the fun part, but can be a bit tricky due to limitations. I interpreted a wolverine as a bear girl with claws since it’s pretty close. Either way, it’s fun generating these pictures on my free time.
I was curious abou the kind of art an AI would come up using the mnemonics, and just as a test I looked up the one you’ve got commited for horse and it’s:
I must admit, I’m not really seeing the horse there. =P
It has horse ears as mentioned, but if you look up Uma Musume or World Witches, there are girls with horse ears and tails. I decided to use girls with animal ears as those generate properly with the AI image generation opposed to trying to get the actual image as it doesn’t really generate animals that well. That is a reason you will see shellfish shown as a crab and a turkey replaced with a chicken.
I am going to take a break from this project since well, for the past two weeks, I have been generating images, but not making any progress with the things that matter, such as immersion and finishing Quartet 2, although I am still progressing in WaniKani. There are plenty of Manga I want to get to. So I am putting this project on hiatus for now as I am going to cancel my NovelAI subscription. May come back to it at some point, maybe during the holidays.
So far, there are 557 unique kanji images, which covers 27.9% of the Kanji available on WaniKani, little over 1/4th. I think that is not too bad, but could be better. Hopefully there will be some improvements down the line when I come back to this project.
