Show IH: Kotoba Dojo — describe an image, AI redraws it, you get scored
Launched today.
Kotoba Dojo is a training game for putting what you see into words. You get a reference image, describe it (type or speak), an AI redraws it from that description only, and a Shiba Inu sensei scores the match and points at which words changed the picture.
White belt → black belt. 22 languages.
I can't write code — no degree, no bootcamp, started from zero in March. I ship by describing the product to Claude + an agentic IDE. This game trains that exact skill. #6 since March. About 40 hours, solo, Tokyo.
Try it: https://kotobadojo.letiziainc.com
Also on Product Hunt today: https://www.producthunt.com/products/kotoba-dojo
One thing I want roasted: does the scoring feel fair after a couple of rounds?
Curious how the scoring works under the hood — does the sensei compare the redrawn image to the reference with an embedding similarity, or does it diff the two images at the pixel/region level to point at which words changed the picture?