I created Hosaka Studio using Claude Code. I wouldn't call it vibecoded however! I'm already a professional software dev and it was three months very hard work getting it to work properly across half a dozen different Linux distros, X11 and Wayland compositors (there's still a known issue on Nvidia+Cosmos).
So yeah, it's coherent, complex and non-trivial but also absolutely not accessible to anyone who can prompt.
Depending on the TTS model being used latency can be reduced further yet with an LRU cache, fetching common phrases from cache instead of generating fresh with TTS.
However the naturalness of how it sounds will depend on how the TTS model works and whether two identical chunks of text will sound alike every generation.
As someone who loved the Moomintroll illustrations I find this both familiar and hilarious. I suppose I might have a different opinion if I'd actually read any of Tolkien's works.
> "She even made some of the characters especially tiny to elevate the landscapes." wish there were more examples of this in the images shown in the article.
So yeah, it's coherent, complex and non-trivial but also absolutely not accessible to anyone who can prompt.