With Runway's Solaris real-time frame-by-frame UI generation, what I can't wrap my head around is state

资讯 / AI 快讯
1

Scroll a list down to the third screen and back up — is it still the same content as before?

Runway’s Solaris is pitched as real-time frame-by-frame generation of dynamic interactive interfaces. The word ‘frame-by-frame’ immediately made me think of video rendering: the UI isn’t drawn once and left sitting there, it’s generated fresh frame by frame. That problem I mentioned above doesn’t exist in traditional frontend — the data on screen three is fixed and traceable, scroll back and it’s exactly the same. Switch to live generation, and where does the state even live, what guarantees consistency before and after — I can’t figure that out.

Whether it looks good isn’t what I care about, I’m stuck on whether this step can actually be used in a product. The name is pretty grand, before an actual screen recording shows up I’m withholding judgment.

2

If the interface is generated frame by frame, wouldn’t filling out half a form get wiped by a redraw?

3

The concept is pretty fresh, but there’s still a long way to a shippable product

4

Front row

5

Whether the feedback from clicking a button is accurate — that’s the real core, and demo videos usually don’t show that

1 Like
6

You still gotta keep the state saved somewhere outside, and just let the model focus on drawing. Otherwise, if you fill in half and then redo the redraw, all that filling was for nothing.

7

Right, keep state outside, model just draws, once that line’s clear the rest is just the same old frontend work.