I’ve been experimenting with the new #ChatGPT image update to test how far it can go in semi-structured UI generation and cross-asset visual consistency. The strongest signal so far is its ability to carry design tokens across a series (hierarchy, colour logic, UI structure, and object styling) at very high speed; the main gaps are still layout determinism, exact camera/crop anchoring, text/data fidelity, and true component-level repeatability from one output to the next.