Why we built per-scene regeneration
Early on, the regenerate button worked the way you would expect: you did not like the video, you pressed it, you got a different video. It tested well in a demo and it was the wrong design. Here is why we tore it out.
The unit of control was wrong
A ten-minute documentary is maybe forty scenes. When scene 31 has the wrong footage, the other thirty-nine are fine. Regenerating the whole thing to fix one of them throws away thirty-nine good scenes to re-roll one bad one, and hands you a completely different video where the one problem might now be in scene 12 instead. You were not editing. You were pulling a slot machine and hoping.
That is the actual complaint behind "the AI video tool gave me something I could not use". It is not that the output was bad. It is that there was no way to fix the one thing wrong with it without risking everything that was right.
Scenes are the natural unit
So the unit of control became the scene. Scene 31's footage is wrong: you regenerate scene 31. The narration is good but the clip is generic: you swap the clip and keep the narration. The line is clunky: you rewrite the line and re-record just that. Forty scenes, forty independent dials, and the thirty-nine you were happy with never move.
It changes what "AI-assisted" means
When the only control is regenerate-everything, you are not really involved in the video. You are a judge giving it a thumbs up or down. When the control is per-scene, the video is yours: the structure, the footage choices, the phrasing, the pacing, all editable in place. The pipeline did the assembly so you could spend your attention on the decisions that actually change the result.
It also changes the economics. Re-rolling a whole render to fix one scene is expensive and slow. Regenerating one scene is neither. The control got finer and cheaper at the same time, which is usually a sign the design is right.
What this looks like in practice
You get a finished video. You watch it once. Three scenes are not pulling their weight. You regenerate those three, swap one clip you would have chosen differently, tighten two lines, and re-render. Ten minutes, not a re-roll, not a rebuild, and the thirty-seven scenes you liked are exactly as they were.
That is the whole argument. The editor is built around it end to end, and what it does walks through the rest of it. If you want to see what it costs to work this way, that is on the pricing page.
Cheers, Carl