I was wondering if anyone would be interested in using such a thing if $ was charged per usage (just cover the costs of the AI calls to Gemini and GPT - otherwise i wouldn’t mind offering the entire thing for free)?
I hope this isn’t against the rules to discuss this here, if so please let me know…
I’m not trying to monetize anything, just thought it would be cool to let other people use what I created.
Basically the idea is for you to plug-in your WK api-key, and then it would allow you to pull from WK any subject (kanji, vocab, radical) which you already unlocked, and then run an LLM-based algorithm that involves a “builder” which tries to craft prompts with several judges that debate whether the prompt answers all necessary criteria, and if so - moves on to generate the image and then pass the image to a visual-judge that assesses whether the image is meeting adequacy standards.
When the judges disagree, they try to convince each-other using arguments. When they both agree (of if the more strict one is not convinced), the builder acts accordingly (proceeds, or retries - taking the criticism in account).
I can share some screenshots from the current prototype (in this case i ran it for “梅”, i.e. ume, japanese plum, as a kanji):
The system can take WaniKani’s suggested mnemonic as inspiration, or ignore it - as it pleases.
would love to hear if there is any interest in this.
because it costs a lot of $ to make these AI calls (i didn’t measure yet but probably something between $0.5 and $1 per full generation, due to the ping pong between all the AI judges, and the call to nano banana pro)