The demos are seductive and, in their way, honest: a prompt becomes a few seconds of plausible motion, the lighting holds, the camera drifts with something like intention. What the demos cannot show is the part that actually matters to anyone making real work — what happens when you need the next shot to match the last one.
We spent two weeks trying to cut a genuine ninety-second short using generative video as the primary source. The footage was never the problem. Continuity was.
The good
Per-shot quality has crossed a threshold. Individual clips now hold up at viewing distance, motion is coherent, and the failure modes have shifted from “obviously fake” to “subtly wrong” — which is real progress. For establishing shots, abstract interstitials, and anything where a human face is not the subject, the output is genuinely usable.
The editing-adjacent features matter more than the generation. The ability to extend a clip, to retime it, to nudge a camera move, is where the tool starts to feel like production rather than a slot machine.
The wall you hit
Character consistency remains the wall. Across shots, faces drift, wardrobe mutates, and the small continuity contracts that an audience reads unconsciously get broken. For narrative work with recurring people, the tool fights you at every cut.
The second wall is direction. You can prompt, but you cannot truly direct — there is no reliable vocabulary for “the same room, from the reverse angle, forty minutes later.” Until that exists, generative video is a source of material, not a camera.
Who it is for, today
If you cut explainer content, motion-driven social pieces, or anything where the shot is a texture rather than a performance, Runway has quietly become part of a working pipeline. If you are making a scripted narrative with characters an audience must believe in, treat it as a tool for backgrounds and transitions — and keep your camera nearby.
