Design-GAN
Describe a website. Watch AI build, test, and refine it.
Design-GAN is an experiment in autonomous design: can AI improve a working
website by trying it, evaluating the result, and iterating on feedback?
It was built to explore moving beyond a one-shot mockup toward designs
that help people complete real tasks.
Start with a brief below. An AI generates the page, automated browser tests
exercise its key tasks, and an AI critic suggests improvements for the next
iteration. Task completion determines which eligible design is kept;
accessibility and correctness checks must pass too.
The loop runs automatically. Follow its progress,
compare iterations, and try the selected design. No manual labeling is
needed to run it.
Start a new run
Starting a run requires a shared token on this deployment — ask the owner. Browsing existing runs is open.
Recent runs
#7
exhausted
A simple, accessible landing page for a weekend walking tour. Use concise text, high-contrast dark text on white, semantic HTML, and one cle
100
#6
errored
A simple, accessible landing page for a weekend walking tour. Use concise text, high-contrast dark text on white, semantic HTML, and one cle
—
#5
errored
A simple, accessible landing page for a weekend walking tour. Use concise text, high-contrast dark text on white, semantic HTML, and one cle
—
#4
errored
Recreate the most useful New Tab Page for the Edge browser you can. Consider user browsing history, trending topics and sites, and anything
—
#3
errored
SWEEP TEST: should be errored by boot sweep
—
#2
errored
A website for hikers to choose, rate, and talk about local hikes in Washington state.
—
#1
converged
DEMO: A landing page for a weekend cycling tour in rural Vermont.
95