Quick Reflection on SVGs AI Image Generation

Everyone knows about using AI to generate images, but fewer people know about using Claude to generate SVGs, and I think this is way cooler.

Take this rubbish for example. Me to DALL-E:

generate a cool minimal geometric background with lines and curves intersecting

This stinks - look at how low res it is. Look at how busy and try-hard it is.

Now, me to Claude:

generate an svg with geometric lines and curves intersecting

Instant classic in my view. But wait there’s more - SVGs can be animated.

Me to Claude, again:

awesome, now make it animated - have all the lines slowly undulate and, and have the points move slowly up and down or left and right (at random)

I think this is a microcosm for a broader point about using LLMs. You absolutely can yolo it and generate a whole image, codebase, story, whatever. No one will stop you and you’ll often get something pretty decent.

But you won’t understand it, you won’t be able to edit it, you won’t be able to fix it if it’s wrong (other than by pleading with your LLM) and as in this case, it will be less cool.

Edit: For context on how I came across this, see this post: a day-build with cursor and typescript