Brøndby Zoo (Lovable)
Feb 23, 2026 · live
A serious website for a fake zoo in a social housing apartment.
This was a test of Lovable as a “spin up a landing page fast” tool, using one prompt and no follow-up adjustment prompts. The result was better than expected, which is annoying, because it means the tools are getting competent.
What this experiment was testing
I wanted to test three things:
- First-prompt quality (how good is the output with no back-and-forth?)
- Code quality for manual edits (is the generated code usable later?)
- Prompting vs hand-tuning (when is it faster to just open CSS and fix things yourself?)
Result (short version)
Lovable is very good at producing a fast first pass for a landing-page style site.
- The initial quality is surprisingly solid.
- The generated site looked coherent and “real” without me nudging it.
- For this kind of task, it can absolutely produce something that is already stronger than what a lot of low-end agencies shipped a few years ago.
- Once the rough design is there, small refinements are still usually faster in manual CSS than prompting your way through micro-adjustments.
That is not a criticism. It is just a practical workflow reality:
- use prompt -> get strong draft
- use code/CSS -> do precision cleanup
The generated site
Live link:
The site is for a fictional “Brøndby Zoo” and was intentionally absurd, while the requested visual style was serious and modern. That tension is the whole point.
The prompt (included exactly as used)
I want a website for Brøndby Zoo - a zoo residing in a social housing apartment in the suburb Brøndby outside Copenhagen. It features a petting zoo in the bathroom, two goats and lots of parrots in the livingroom, the kitchen has a couple of snakes and spiders as well as a lynx cat. In the bedroom lives an alpaca (stolen from a nearby farm). Tickets are available online - 80 dkk for kids and 120 dkk for adults. I want a serious looking website for the zoo - with an modern and urban look
Yes, the “alpaca (stolen from a nearby farm)” part was in the prompt on purpose. If the model can keep a straight face through that, it deserves partial credit.
Notes on output quality
1) Visual quality (first prompt)
The first render looked surprisingly credible for a modern landing page:
- clear sections
- decent hierarchy
- coherent style choices
- “serious brand” tone despite ridiculous content
This is exactly where tools like Lovable shine: getting from nothing to something usable very quickly.
2) Code quality (Vite / TypeScript)
The generated project was built with Vite + TypeScript, and the code looked decently clean and structured for manual work afterwards.
That matters more than people think.
A lot of AI tools can make pretty screenshots. Fewer produce code that you can open later without immediately regretting your life choices. This one was workable.
3) Prompting vs manual CSS edits
For small visual adjustments, prompting can be slower than just editing CSS directly.
In practice:
- prompting is great for directional changes
- manual CSS is better for fine control (spacing, exact sizing, small layout corrections, polish)
The useful workflow is not “AI or manual.” It’s:
- AI for momentum
- manual edits for taste
Takeaway
Lovable feels strongest as a rapid landing page generator and prototype launcher.
If you need:
- a first visual direction
- a serious-looking mockup
- a quick campaign/site concept
- a decent codebase to refine manually
…it’s a strong tool.
If you need pixel-perfect control immediately, you will still end up in the code. Which is fine. That’s what the code is for.
What I’d test next
- How well it handles second-pass revisions from design feedback
- Whether it keeps code quality after multiple prompt edits
- How fast it is to adapt the same concept into a different visual identity
- Where the “prompting becomes slower than editing” threshold actually is (my guess: very low for CSS polish)
Sponsorship note
This experiment is a good example of an in-kind tool sponsor format on UNUSABLE.Ai: a tool provides access/credits, I build something real with it, and the result stays public.
No paywall. No fake “community.” Just a weird project and some honest notes.