Post

It does good Mermaid diagrams though.

It’s been a while But there are markup languages for diagrams like mermaid. Not sure if there are any styling mechanisms for them similar to css. Perhaps a mermaid > structured HTML then style that with css?

One way I got to some interessting results was think in a more multi modal way. A prompt is translated into a blueprint diagram, then translate into node based logic gates structure (they are good at that), and then render the outcome. This one was prompt to new game.

We’ve reached identical conclusions. If you figure this out I’d be very keen to hear how you did it!

Vector graphics in general is a bit of a blind spot. It's not easy generating e.g. icons using LLMs.

Using Claude because I need these to be in working code – SVG, CSS, HTML, JS. It allows me to have have fine-grained control over details. And I usually add in animations and interactions, plus I need them to be responsive for mobile. Nano banan still has a ways to go 😄 bsky.app/profile/magg...

What until it can work in layers, or generate native photoshop, illustrator files. Then we're all c̶o̶o̶k̶e̶d̶ unlocked for utopian "understand through diagrams" future (working on my ai optimism).

I've kind of enjoyed what it Opus 4.5 produces out to excalidraw diagrams. However, it does end up blocky pages in structure instead of flows similar to wireframes. It also struggles to leave enough room for text when wrapped in boxes

what a pretty theme!

try mermaid for standard diagrams, and d2 for more general diagrams. d2 can output svg. I intermixed AI to get the right layouts and clustering.

It's OpenCode! It's delightful. With the theme catppuccin-macchiato. Also using Ghostty for my terminal and have same theme set there.

The dream is to use AI to get started with a set theme .. something you can point the AI at (url, screenshot, etc) .. prompt it and get a first pass. But the iteration step is tricky. More prompting? At that point I want to take over in a tool (PS, Illustrator) and get what I want. Or code.

Generating for draw.io used to work well (GPT-4 Turbo, Claude 3.5 Sonnet, …). Wrote a walk-through (€, 🇩🇪) here: www.heise.de/ratgeber/Pro...

For my use cases, Claude worked best at generating Mermaid diagrams ( which you can then render however you want). Downside was that they’re pretty ugly by default.

Oh I didn't know it was customisable at all! I'll take a look. Maybe models will be better at generating initial ideas with a more structured language like that, and I can work off it

Have a look at at Plot and d3 from observablehq.com - they have higher level ways of expressing graphs and diagrams, so you're not relying on the LLM to understand collisions

Yeah I thought this too – that starting with mediocre, predictable stuff narrows the range of exploration substantially. We can ward this off a bit by having high standards and/or a specific vision for the final result pre-prompting. There's a tradeoff calc here: easier/faster but slightly worse

Ask it to create a mermaid diagram. For that it does great.