Watch Claude for Chrome create Kid Pix art from your requests.
Current Queue
Loading queue...
Recent Artworks
Frequently Asked Questions
▶ How do I submit my idea for Claude Draws to work on?
Just head over to the submission form and describe your artwork idea in as much detail as you'd like!
Claude Draws processes the most-upvoted submissions first. After you submit, you can upvote other submissions in the queue - and others can upvote yours too! You can watch Claude Draws work on the livestream at the top of this page. And if you'd like to be notified when your artwork is ready, just include your email address in the form - it's completely optional.
So don't worry if the queue is backed up - you don't have to watch the livestream to catch your finished artwork. If you provided an email, you'll get a notification when it's done. Otherwise, you can check back in the gallery to see your completed creation! Plus, each artwork page includes a recording of the entire drawing process, so you can watch Claude work through your design anytime.
▶ Who built this?
Hi, I'm Andrew. I'm a software engineer and in the 1990s I was a kid who loved Kid Pix. You can learn more about me and this project on the about page.
▶ Is this project affiliated with Anthropic?
No. These days I'm a solo engineer and I built this project as part of my exploration of AI browser agents.
▶ How can I get access to Claude for Chrome?
I joined the waitlist when Anthropic first announced Claude for Chrome and got access a couple weeks ago as a Claude Max subscriber.
Check out this doc on Anthropic's site for up to date instructions on how to get access yourself.
▶ Wait… this Kid Pix looks a little different than I remember!
Good eye! Vikrum's Kid Pix implementation is pretty faithful to the original - you can use it here. I forked that repo and made a few changes to help Claude (and other browser agents) use Kid Pix more easily.
Some of those changes were necessary because of Claude for Chrome's limitations. For example: it can't emulate a "shift + click", necessary to create enlarged versions of Kid Pix stamps. So I added a toggle it can use to hold down a virtual "shift" key instead.
Other changes, though, just help Claude stay on track. Browser agents work best when they can use a combination of screenshot and HTML data, and when that HTML data has useful information included in each element. My fork adds a lot of little details to the HTML to help Claude understand what will happen when it presses each individual button.
Finally, I added a status bar so Claude and livestream viewers alike can more easily see what tools and subtools are active at any given moment.
If you are a Kid Pix purist, you can find a full emulator loaded with v1.1 on the Wayback Machine.
▶ What are those screensavers I see between artworks?
Those screensavers come from After Dark, classic screen saver software that launched in 1989. If you spent time around computers in the 1990s, you probably remember Flying Toasters, the fish tank, or one of the many other iconic modules that came with After Dark.
I pulled a selection of my favorite After Dark screensavers to use with Claude Draws. You can see more in this playlist on YouTube.
▶ So, is this an art project?
Not really. When I think back to my days using Kid Pix, the finished artwork wasn't really the most important part. In fact, it probably got blown up with the dynamite tool a few times along the way!
To me, this project is more of an interactive educational demo you can use to learn about how AI browser agents work. This type of AI is still in its early years, but one day something like Claude for Chrome might be capable and reliable enough to help you get things done.
That "thing" probably won't be "create a Kid Pix artwork." But I found it interesting to explore AI browser agents via assessing their ability to use one of the first pieces of software I encountered in my youth.
▶ Have you tried it with ChatGPT Atlas or Gemini 2.5 Computer Use?
I try just about every browser agent model and tool! This project, however, was not about benchmarking. After all, there aren't a lot of websites out there that work like Kid Pix.
I did briefly try out my custom prompt and Kid Pix UI with ChatGPT Atlas and Gemini 2.5 Computer Use. I got the best initial results with Claude for Chrome.
It also made financial sense to use Claude for Chrome for this project, since its usage is included in my existing Claude Max plan which I use with Claude Code. As any engineer who has built browser agents knows, there are a lot of tokens involved in processing all those screenshots and HTML!
You can certainly try out my prompt and UI with your browser agent of choice, however. Each entry in the Claude Draws gallery contains the original prompt passed to Claude for Chrome. My custom fork of the Kid Pix UI is available at https://kidpix.claudedraws.xyz.
▶ Is Claude Draws open source?
Yes, Claude Draws is open source under the MIT license. You can find the GitHub repo here.
I don't recommend trying to set up the repo yourself just yet. The docs will get a bit more love this weekend. If you're itching to try anyway, you're welcome to open an issue if you get stuck.
▶ What tools did you use to build this?
Anthropic's Claude for Chrome browser extension does the heavy lifting here. Most of the Claude Draws code is focused on accepting requests via the submission form, programmatically injecting prompts into the browser extension, and then processing the newly created artwork before adding it to the gallery site.
I wrote most of that code in Python using Cursor and Claude Code. The other notable tools in this project are Playwright, Temporal, and BAML. The gallery site runs on Cloudflare and Claude Draws itself is currently deployed on a spare PC in my garage.