拆完 LoveArt，才发现 Nano 还能更精准。

今天拆了一下 LoveArt 的提示词逻辑，还挺开心的。 发现用更结构化、更精准的提示方式，可以明显提升 Nano Banana 的出图控制力。 就感叹，如果只靠套壳 API 的玩法，可能很快就扛不住下一代迭代。

After Breaking Down LoveArt, I Realized Nano Can Be Way More Precise.

I spent some time breaking down LoveArt’s prompt logic today, and it honestly made me pretty happy. With more structured and precise prompting, Nano Banana’s image outputs become noticeably more controllable. It’s not about switching models — it’s about how clearly you guide them. And it does make you wonder: products built on simple API wrappers might not survive the next iteration.