I was listening to the latest Latent Space podcast with Steve Ruiz of tldraw. (You can watch the full video below). What I was taken back by is the “Make Real” demo that included different approaches to using GPT4-Vision. Of course, my first instinct was to play test it and see what I could make and where the limitations are.
As a part-time designer, I was curious to see a working demo of a wireframe I had in mind.
“Make Real” outputs in HTML, CSS, and JS (with tailwind), and I was able to cobble together this little demo of telling you the timezones based off of the selected input. One thing to note is the component below is actually a Svelte component, which I created using ChatGPT to convert the HTML to Svelte.
And of course, I’ve made slight modifications so there are more timezones than the ones given to me.
Design aesthetic aside, this is a fully functional prototype that I could reuse and prettify. With a little bit of coding knowledge, I can be off racing towards putting this in a production-ready app. Of course, the way to make these modifications are easier if you have some implementation knowledge of what you’re doing.
Thoughts on Make Real
It’s not an career killer for front-end engineers. We are far from that. Instead, these should be seen as tools to augment our work and allow us to spend more energy playing over mulling over small implementation details that might get changed later.
With each iteration, I was able to get “Make Real” to output different variations of the same idea, to really see what the realm of possibilities are. There’s a balance though of giving annotations and adding or subtracting more to the drawings. For example, when I asked it to give me 50 timezone selections from the most populated areas on the earth, it only gave me 10 items. And even then, ChatGPT came up with TZs not in use.
In a different iteration, I wanted an explorable map. When I asked it to give me a map of where this selected timezone was, it gave me a placeholder. Not so helpful.
The following is the video with Steve Ruiz, and I highly recommend watching it to see more.
Where do I think we can go from here
I haven’t tried state machine drawings yet, and I think that can help extend what we can do with drawings. I also would love to explore what other people have done as helpful aids for the drawings to get GPT4-Vision to come up with something better.
Try it out today by giving it your OpenAI API Key.