expectedwrong hindsight

There Are No Components

Ideogram just shipped readable text in generated images, and the logical endpoint is that UI components don't exist anymore.

2 min read 224 words #generative-ai #ui #design-systems #ideogram #interfaces
hindsight — half right

text rendering in images improved dramatically — flux, dall-e 3, ideogram 2. but the 'no components' thesis was too aggressive. we still have react, still have design systems. generative UI exists as a pattern but didn't replace the component paradigm.

Ideogram dropped something today that renders legible, accurate text inside generated images — not the usual hallucinated glyph-soup, but actual words, correctly spelled, correctly placed.

That's the boring description.

The less boring description: for the first time, generating an image of a UI is nearly indistinguishable from building one.

There's a threshold that matters here, and it's not quality — it's latency. The moment generation is fast enough to update on interaction (a click, a keypress, a scroll position), the entire premise of interface development collapses. You don't need a button component because you don't need a button. You need a model that renders pixels showing a button, and then renders the next state when the user touches it.

Design systems, component libraries, accessibility trees, z-index wars — all of that is load-bearing scaffolding for a building that hasn't been constructed yet. Swap the foundation and the scaffolding just... doesn't go up.

The usual objection is performance. Real-time generation is too slow, too expensive, too power-hungry. Fine, today. This argument ages poorly and everyone knows it.

The weirder objection is that humans want deterministic interfaces — the same button in the same place every time. Maybe. Or maybe that preference was trained by thirty years of no alternative.

There are no components. We're just waiting on the hardware to catch up to the claim.