PROMPTS
5 Sites For Better AI Design ⚡
If your AI designs keep coming out generic, you are missing the most important step. The best AI designers are not better at prompting, they are better at.
The people getting the best results out of AI design are not better at prompting. They are better at collecting inspiration.
Here is what I figured out. AI can build almost anything now, but it cannot read your taste. Ask it for a landing page with no direction, and it hands you the average of every landing page it has ever seen. That is exactly why so much AI design looks the same. The fix is not a cleverer prompt. The fix is giving AI real creative direction, and direction starts with references.
So before I ever open a chat window, I pull from a short list of sites. Think of collecting references like a mood you are handing off, not a homework assignment. The more specific your inspiration, the less generic the result. These are the five I actually use, what each one is good for, and the exact way I turn them into direction that AI can follow. Save this so you have the whole list in one place.
Mobbin
Mobbin is a huge library of real, shipped app screens, hundreds of thousands of them, plus the actual user flows behind them. Every screen came from an app that truly launched, so you are looking at design that already works in production, not concepts that never shipped. When I want AI to build something clean and usable, I show it Mobbin screens as proof of what good actually looks like. It even connects to Claude Code through its MCP, so AI can pull real screens directly instead of me copying and pasting screenshots.
visit →
Awwwards
Awwwards collects award-winning websites from some of the best studios in the world. This is where I go when I want high-end web layouts, bold animations, and creative direction that feels expensive. I find one or two sites with the exact energy I am after, then I point AI at that feeling instead of trying to describe a look I do not have words for yet.
visit →
Cosmos
Cosmos is like Pinterest built for designers. It is where I build my moodboards before a project starts, gathering colors, type, and layouts that feel like they belong together. The quality of what gets saved here tends to be a step above, so my boards come out cleaner. Once the board feels right, that board becomes the brief. I hand AI the vibe I already collected instead of guessing at it in the moment.
visit →
Pinterest is still one of the fastest ways to explore colors, branding, and overall visual direction. I know it is not the fancy answer, but nothing beats it for casting a wide net in a hurry. I search a feeling, save fifteen things in two minutes, then keep the three that keep showing up. Those three become my direction.
visit →
NameThatUI
NameThatUI is basically a dictionary for UI components. If you spot a design element and have no idea what it is called, you look it up here and get the real name. This one matters more than it sounds, because AI cannot build what you cannot describe. The right word is often the whole difference between a generic result and the exact thing you pictured in your head.
visit →
How To Actually Use Them
Collecting references is only half of it. What you do next is where most people go wrong, so here is the exact flow I use once I have gathered a few things I love. It takes about five minutes and it changes everything about what comes back.
- Gather 3 to 5 references from the sites above. Not fifty. Just the few that keep pulling you back.
- Describe what you like about each one. Write a quick sentence per reference. The spacing, the color, the boldness, the calm. Name it out loud so AI knows what caught your eye.
- Give AI the combine instruction. Do not ask it to copy any single reference. Ask it to blend the feeling of all of them into something new for your product.
The real unlock of AI agents is not full autonomy, but human-in-the-loop systems design. Structure your agent inputs with explicit constraints, negative prompts, and automated test checkpoints. When building tools, keep token consumption lean by caching system prompts and isolating tasks into specialized sub-agents.