What I learned drawing dozens of Arabic books with AI

When I started Sutur, I thought drawing would be the hard part. It turned out to be the easiest.

An image comes out in half a second. What takes time is everything around it.

First: the model draws what you mention, even when you tell it not to. I wrote “no speech balloons” in the page description, and every page filled with empty balloons. I wrote “head without a hat,” and it drew me a bald man. I learned to describe what I want, not what I fear.

Second: Arabic lettering is the weakest link. Ask it to draw a word floating in the air and it draws shapes that look like letters and say nothing. So every line in a book is short, no more than sixty characters, and I check each page myself.

Third: consistency is harder than beauty. A face that works on page one can change by page five. Whatever the model can’t remember, I have to write down every time, in exactly the same detail.

Fourth: reviewing is the job. I don’t publish a book until I have looked through every page. AI makes the drawing faster, but it doesn’t carry the responsibility for it.

I’m writing all this in public because I want to know, honestly, what these tools can and can’t do for Arabic culture.

The machine draws in half a second, but it doesn’t know when to stop. That decision is still mine, and yours when you read.

Try reading one of the books, then write me one thought: what stayed with you, and what I broke.