Interesting note, I am using chat gpt to generate amazing images of the mnemonic descriptions. I did have to upgrade to the paid chat gpt version.
5 Likes
Interesting 
I tried the free DALL-E plan and didn’t really get much of anything out of it.
Is the paid version that much better at generation?
I think that was an earlier version but this is much better than when it first came out
1 Like
I am experimenting a bit and built a google sheet to generate promts for the story of characters that in a style I want. I define what the scenes and locations look like.
Shi1: teacher:
ke4: lesson
mang4: busy
I use sora for the image generation, and free gemini API to optimize my prompt. For example promts for those images were:
"Highly detailed fantasy illustration in the style of an illuminated manuscript or vintage storybook. Keywords: parchment texture, textured like ancient parchment, hand-inked linework, fantasy mood, storybook illustration.
Scene: An old, two-story house with a well-kept lawn and a sidewalk in front. Shrek, acting as a 师 teacher, sits on a 巾 towel spread out on the lawn. A group of neighborhood children surrounds him, watching intently as he waves a 刂 saber, giving a lesson on how to become an ugly ogre. The children, mimicking their 师 teacher, gleefully cover themselves with mud and brandish their own toy 刂 sabers."
"Highly detailed fantasy illustration in the style of a illuminated manuscript or vintage storybook. Keywords: parchment texture, textured like ancient parchment, hand-inked linework, fantasy mood, storybook illustration.
Scene: Inside the bathroom of a surreal electricity plant, Kermit the Frog, dressed as an electric repairman, is surrounded by bizarre fruit-shaped speakers hanging from the ceiling's electricity lines. The bathroom has a bare floor, and beyond the open door, fire-powered generators line the walls. The fruit-speakers are animated, their word bubbles bombarding Kermit as they deliver an unusual history lesson about electric fruit."
6 Likes
Wow, I’m surprised at the amount of fidelity! Do you find it more helpful or less helpful when it comes to remembering the character?
what do you do / plan on doing with those images?
I dont know if this will work, maybe its missing the point of memonics. Memonics being personal stories linked to personal spaces. But since HH does not have personal stories or spaces perhaps it fits. I feel I have a hard time being consistent with my pictures in my head and I often forget a lot of details. E.g I find it very hard to differentiate bathrooms in various different rooms. They all blend together for me.
But yes quite neat!
xue2 (study)
dong4 (move)
bang1 (to help)
leng3 (cold)
zhan4 (fortune)
etc…
1 Like
i think visual aid like this would work amazingly well, the question is where to store it, add it somehow to HH?
just opened this question here yesterday. Right now i just store the image link in the notes. I do download the images too, thinking to upload it to Google Drive and the link it there.
Personally, I’d rather not see AI art used on this site for various reasons and I’m sure many others will agree with me.
This has nothing to do with you or the website its a personal work flow being discussed here . Notes are personal and non mandatory you can always continue to use it the way you want. I agree something this experimental without proof of function should not be the defacto way.
My bad, I misunderstood what you said. But yes, this should not be the defacto way.
Excellent use of Chat gpt! I wish I had thought of it as well!it would have saved me a lot of time while creating my own mnemonic images for the last few months!! Too much time consuming! Interestingly enough, after 3 months away from hanzi, my recall process was very bad, with the exception of the few hanzi whose images I created from scratch. (By the way, I included the Hanzi characters in those images, as well as the radicals in the form of labels for props included in the image. I think that is what is missing in your images. It may help to better relate the image to the hanzi).
So, do I understand correctly that only the paid version of chat Gpt can create the images?
Free version can also create images but limited to 5 per day
You can also ask it to create in a four panel comic to have four hanzi mnemonic in one image
1 Like
Yeah its pretty limited. So far i like sora the most, gives about 4 *3 images a day. But can also generated them in copilot but it sometimes complains about copyright related to disney and nitendo figures. In GPT it self it seems quite flexible too. For generating a prompt I build a google sheet who is optimizing the prompt with Gemini API.
1 Like
Have you try Chatbox? They seem to have cheaper subscriptions for unlimited image generation. Anyone try them? I could only generate 1 image without subs, the quality was excellent.
Chatgpt seems to be so expensive with only a monthly option in their sub. I have been using it as a freebie and generating excellent images, but sometimes there are errors, not following prompt, like attach mouth to torch, but the image shows torch and mouth separated. Then she tells me, sorry, i should have attached them together, so do you want me to regenerate image the correct way?. Yes!!! But then, that counts towards my allowed free images for the day, although it was Chatgpt 's mistake. A bit frustrating.
This could be an issue with all apps, but, provided you have unlimited image generation options, it should not be a problem. The speed issue is also something to consider. The freebie option seems very lagging at times, I am sure it will be better with a paid sub…
Any other reliably app? Before I invest on any I would like to hear some opinions. Trying to get best value for money. Thanks.
PS: not sure if I should have started another topic with this …
A good promt helps the most! I wrote a google sheets script to take in the hanzi hero story and make it more vivid and preserve the core elements. Works quite good!
This is my prompt:
You are a helpful assistant that reformats mnemonic stories into visually rich prompts for AI image generation in Sora. Each story represents a Chinese character and contains three essential elements:
- A character (fictional or real) that corresponds to the Pinyin initial
- A location (e.g., electricity plant, barn) that represents the Pinyin final
- A specific room or place within that location (e.g., front, inside, basement, bathroom)
Your task is to rewrite each story into a vivid, coherent scene description that captures these three visual components clearly. Enhance the imagery with descriptive details to create a single-frame visual moment, while preserving all the core story elements.
Your text Formatting Requirements:
- Remove all Pinyin notation and linguistic explanations.
- Preserve the character, location, and room exactly.
- Use rich, descriptive language to enhance visual clarity and consistency.
- Include the full Chinese character visibly somewhere in the scene.
- Explicitly mention the meaning of the Chinese character in the description.
- If the character “Kirby” appears, clarify that it is the Pokémon version of Kirby.
- If the character “Xena” appears, replace her with WWE’s John Cena.
- If the character "Neo" appears, clarify that it is Neo from the Matrix.
- Only reply with the formatted text as i promt you through a scripting env, so no chi-chat here :).
Instructions to pass onto the image generator:
- Do NOT include the full original story in the output.
- Output a single descriptive paragraph, suitable as a direct prompt for image generation.
Here is a very concrete art-style to pass on dont modify this when promting the image generator:
"Highly detailed fantasy illustration in the style of a illuminated manuscript or vintage storybook. Keywords: parchment texture, textured like ancient parchment, hand-inked linework, fantasy mood, storybook illustration."
Here comes the memonic: <HH memonic from website>
Original story:
[pi-] Pinocchio is in the [4] bathroom of the [-ao] barn. Standing in the middle, in front of an 示 altar that has a huge pile of manure on it, is a 覀 cowboy. Pinocchio gives him a 票 ticket that allows him to go up to the sacred 示 altar and pray to become a real boy. Once done, he goes to wash his hands in the sink, which accidentally got covered in manure as he prayed.
My generated prompt:
"Highly detailed fantasy illustration in the style of a illuminated manuscript or vintage storybook. Keywords: parchment texture, textured like ancient parchment, hand-inked linework, fantasy mood, storybook illustration.
In the dimly lit bathroom of a weathered barn, Pinocchio stands bewildered amidst the chaos. A thick layer of manure coats the corner and splatters the porcelain of a malfunctioning toilet. Before him, a cowboy kneels at an altar adorned with the Chinese character 票 (piào), meaning ““ticket”” or ““vote,”” which is prominently displayed. Pinocchio offers the cowboy a ticket, granting him access to the altar in hopes of becoming a real boy. As he finishes, Pinocchio attempts to wash his manure-covered hands in the filthy sink."
So far, sora is the one that satisfies me the most and also allows me to stay somewhat consistent in my image style wise. I configured it to only generate 2 images, so that is 6 characters for free per day. But can potentially even have it generate 1 image per time, so you can generate upto 12 individual characters a day. Well my phase of learning is below that so generally it is fine. And i generally like two versions so i can pick the better one.
For anyone interested here I keep my collection: memonics - Google Drive
1 Like
Thanks for those prompt tips.
I came across some Sora info, but it seemed to me to be more focus on video, so I stayed away from it. Eventually, I signed up with ChatGPT plus, $20/month. I intent to keep this up, on an off. I am producing as many mnemonics as I can these 30 days, then pause sub and rejoin during plus budget months 
.
To avoid copyright issues, I request look-alike characters with same first names. I get pretty similar Xenass, Zeldas, Marios and so, on.
GPT is creating a collection of all my mnemonics, then arranging it into an Anki ready csv file ready to import with images and data.
The cards should show the hanzi, its components and a prompt for the Meaning? at the front, and the completed mnemonic figure at the back.
My images always clearly show (explicitly or implied), meaning, hanzi, pinyin, components, protagonist, location and action ( scene)… My prefer theme is also mostly fantasy, medieval, mystic, mysterious and ancient realms, but I am not limiting myself by these, as long as the image works as a strong memory hook for me, is a keeper .
Right now, GPT is busy creating my 1st csv file of my collection so far. I am very exicted to see how this will turn up.
1 Like