TLDR: the author tasked a bunch of AI models to generate commands for a custom brush script language to simulate Japanese calligraphy, for 5 each of word/phrase and emojis, 10 total.
From what I see:
- in script division: Gemini 4 Argon is about as good as it gets. 3.x not so much. GPT-6.0 models are not bad, 6.1 models are a bit Chinese coded, Claude models are model dependent and a bit Chinese coded as well, and most others are somewhere between barely recognizable to just trash. This includes many Chinese models and national origin of models don't seem to affect the result much.
- in emoji division: styles vary, some look creative, most look fine. A lot better than the script division results.
My affiliation with the author: none, I think I've seen this domain name in the past. I just thought this is interesting.
TLDR: the author tasked a bunch of AI models to generate commands for a custom brush script language to simulate Japanese calligraphy, for 5 each of word/phrase and emojis, 10 total.
From what I see:
- in script division: Gemini 4 Argon is about as good as it gets. 3.x not so much. GPT-6.0 models are not bad, 6.1 models are a bit Chinese coded, Claude models are model dependent and a bit Chinese coded as well, and most others are somewhere between barely recognizable to just trash. This includes many Chinese models and national origin of models don't seem to affect the result much.
- in emoji division: styles vary, some look creative, most look fine. A lot better than the script division results.
My affiliation with the author: none, I think I've seen this domain name in the past. I just thought this is interesting.