Show HN: Text-to-Figma

Hey HN!

I've made a NLUI for Figma. This is based on GPT-3 (for generating components & schemas) and Dall-E 2 (for image generation); it's remarkably flexible already and there's plenty of room to expand. One of the more noteworthy features is the ability to edit existing designs instead of just generating things from scratch - we accomplish this by training it on pairs of (before, after) designs in Figma with a description of what changed, and it learns to output the diff based on the description of the change.

Let me know what you think!