Kākāpō Party
Simon Willison’s Weblog Simon Willison
Simon Willison had Claude make a kākāpō pixel-art party for a keynote slide. Then Claude Code used Playwright to turn the page into a video.
Based on reporting by Simon Willison’s Weblog, Simon Willison — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Simon Willison needed one last flourish for a closing keynote at WeAreDevelopers World Congress North America, so he went looking for a way to put kākāpō on screen. The hook was a record-breaking breeding season in 2026, but the real story is how quickly the tooling turned a whim into something usable on stage.
He grabbed three kākāpō photos from Google image search and fed them to Claude Opus 5.5 with a very specific prompt: make animated pixel art in HTML5 canvas, make it obviously pixel art, make the birds jump around at a party, and include at least 20 of them. The result was not just passable. By his account, it was pretty great.
But a keynote slide needs to live inside Keynote, not sit in a browser tab, so he handed the HTML off to a local Claude Code session and asked for a 15-second video. The instructions were equally particular: load the file in a browser, click around a few times to trigger the confetti effect, and don’t start clicking until 3 seconds in.
Claude Code used Playwright to do the job, and the script it produced was short and tidy. It opened the page at a local file path, recorded video, and spread clicks across the center, corners, edges, and middle of the canvas at timed intervals. That was enough to generate the final clip he wanted for the slide.
There’s a nice little chain here: image search, a prompt, generated HTML, then browser automation to turn the result into a stage-ready video. It’s the sort of workflow that makes a keynote prop look easy, which is exactly why people keep underestimating the amount of machinery now hiding behind the pretty stuff.
My take — AI-written commentary, not fact-checked reporting
This is the good kind of AI demo: narrow, useful, and a bit ridiculous in the best way. Closed models still get the glory, but the real win is the plumbing around them — a browser, a script, a video file, done. That’s what actually ships, not the glossy breathless nonsense people usually post between coffee and doomscrolling.
Read more about this at: Simon Willison’s Weblog