Sometimes AI is great, other times it fails miserably. But everything in this journey is all learnings at this point. The more familiar you are with the technology you're prompting toward as an outcome, the better results you'll receive. Example of a failed prompt result based on a challenging image and description. (I would have been shocked if it had worked.)
In contrast, when sharing a simple styled "eyebrow" component I wanted to animate, I provided Claude a specific and descriptive prompt to build requesting Tailwind CSS, Next.js, and Motion.js. It took about 7 re-prompts to polish out the idea starting from a simple image. But I described the component — font + weight + sizing, hex colors, motion style and direction, end state blinking, etc.
TLDR: Break down large ideas to smaller consumable snippets. Generate working micro-prototypes. Cobble them together into a larger idea.