Synthetic art from the depths of my twisted mind.

I tested a sketch to see how the AI (Nano Banana 2) would handle it, and these are the results. The images turned out well, but it struggled with some details, such as the bullets in the belt, the cord from the grappling hook, and the corner of the left ammo box. Even though it's a very simple sketch with reasonably well-defined lines, it just couldn’t get it right after about 70 attempts.

The thing is, when you ask Nano Banana to create, for example, an image of a monkey wearing a hat, sunglasses, red sneakers, and holding a banana, it will either miss one of these items or get something wrong, like changing the color or style. When it doesn’t, it adds some unsolicited extra detail. You squint, scanning the picture for flaws—everything seems fine—but now the monkey has a second banana discreetly protruding from its ass. This goes on in an infinite loop. That’s why I’m convinced the free version is programmed not to deliver exactly what you want, no matter how precisely you craft your prompt.

Cyber Sadhu, the BTL prompt cook.