A new web project called UI Palooza asks whether AI models have a design taste you can spot. It gives several models the same design brief and then shows the interfaces they build next to each other. Hacker News reports that the project launched as a “Show HN” post and picked up a score of 164, which suggests developers and designers care about the question.
The pitch is short: “A test of taste.” Each model gets the same brief, and UI Palooza shows “what do they choose for the type, the layout, the way an interface works.” Instead of scoring code or reasoning, it looks at the design choices models make when nobody tells them exactly what to do.
What UI Palooza Lets You Do
According to Hacker News, the site is built around three ways of comparing results:
- Compare models head to head. You see how different models answer one brief, side by side. That makes it easy to notice when one model always picks a certain font or layout.
- Add a design skill. You can attach a design skill (a set of instructions that steers how a model handles design work) and see how much it changes the result. This tests whether a model’s defaults are fixed or easy to steer.
- Look across attempts. You can compare several runs from the same model and see “what changes and what keeps showing up.” This one matters most. A choice that shows up once could be random. A choice that shows up every time looks like a habit, and the habits are the real taste.
The project also includes a write-up of the story behind it, though the launch post doesn’t go into detail.
Why This Matters
Anyone who has shipped an AI-generated frontend knows the look: soft gradients, rounded cards, a centered hero, and the same few fonts over and over. The joke that every AI-built landing page looks alike has been around for a while. UI Palooza turns that joke into something you can actually check.
What stands out here is the focus on repeated runs. Most model comparisons treat one output as the answer. But design taste is about patterns, and a pattern only shows up when you run the same prompt several times. Showing which choices survive across attempts lets you see a model’s real defaults.
There’s a practical side too. If you use AI coding tools to build interfaces, knowing each model’s habits tells you what to push back on. If a model always reaches for the same layout, you know to spell out a different one in your prompt. If a design skill clearly moves the output, that’s evidence that writing one is worth the time.
Who It’s For
Here’s who’s likely to find it useful:
- Frontend developers picking a model for UI work and wanting to see real output, not benchmark numbers.
- Designers curious whether AI tools have a recognizable style and how far that style can be pushed.
- Teams building design skills or system prompts who want to see whether their instructions actually change anything.
Limits Worth Noting
The launch post is light on details. It doesn’t say which models are included, how many attempts each one gets, or whether visitors can submit their own briefs. It doesn’t mention pricing or access terms either, though as a Show HN project it looks like a site anyone can browse.
Keep in mind that “taste” is subjective. UI Palooza can show you what models choose, but it can’t tell you which choices are good. That part is still up to the person looking.
What Comes Next
As AI coding agents build more of the web’s interfaces, their default habits end up shaping what the internet looks like. Tools like UI Palooza make those habits visible, and that’s the first step toward breaking them when you need to. Model makers may also start paying closer attention to how their models design, not just whether the code runs. You can find more details, including the story behind the project, in the original Hacker News post.