And it holds steady for the whole video.
This week, xAI updated Grok Imagine Video. Give it one photo and one voice sample, and it holds the same face and the same voice consistent across every scene of a generated video, not just the first few seconds. Here's why that specific detail matters,
The Hard Part Was Consistency
Plenty of tools could fake a voice for a short clip; almost none could hold it steady scene after scene without drifting. This update claims to have solved that, natively, at 1080p.
The Price
$0.25 a second. Up to seven reference images for extra control. Currently gated to paid subscribers, with a developer API path for the video side, voice-reference specifically still requires contacting sales, not yet fully self-serve.
The Department This Actually Lands On
Not editors, not VFX. Voice. ADR, dubbing, narration, voice acting. A solved consistency problem at a quarter a second is a number you can do real math against this week, not a someday-threat.
An Honest "Not Yet"
I'm not saying it sounds as good as a real voice actor who's spent years in a booth, I haven't put it against a real project yet. But I've watched enough of these tools go from "good enough for now" to "good enough, period" quietly.
My Honest Take
If you do any kind of voice work, this is worth testing yourself this week, not filing away as someday-news. The tools that get dangerous are never the ones that announce themselves as finished; they're the ones that quietly closed the one gap everyone was relying on.
Has AI Voice Cloning Come Up at Your Studio Yet?
xAI's Grok Imagine update this week solves AI video's actual hard problem: holding a cloned voice consistent across an entire video, natively, at $0.25/second.