Description:
Playing with Astra, I noticed a significant leap in visual, spatial, and graphical reasoning. So I started experimenting with something LLMs have historically struggled with: creating illustrations and drawings. Turns out, Astra is pretty decent at it.
That got me thinking about combining the exploratory potential of AI generation with the precision and control of hands-on editing.
The core idea: Make something, understand its parts, and keep discovering what else it could become.
Generate editable vector drawings from language. Manipulate individual elements or describe what you want to change. Preserve what you like while exploring different styles and interpretations. Combine drawings into new scenes, like asking a bird to ride a bicycle. Get suggestions informed by the artwork itself, not just your prompt.
It's still very much a work in progress, with plenty that doesn't quite work yet. But it's been a fun experiment in what creative tools could become when generation and direct manipulation complement each other.