What actually shipped this week
Kuaishou officially launched Kling 4.0 globally on October 16, 2026, after a staged rollout that started with a limited early-access build called Kling 4.0 Flash on September 28, 2026, for top-tier annual subscribers. Other trackers confirm the full flagship model went live a day earlier, on October 15. Either way, this is a brand-new release, not a minor patch, and it's built around a problem most short-form creators know well: keeping faces, outfits, and voices consistent across cuts.
The headline upgrades are native 30-second clip generation, an expanded Omni Reference system that accepts up to 15 multimodal inputs, and voice references that can be bound to a saved character. Kuaishou is positioning all three around narrative coherence, not just visual quality, which is a notable shift from earlier model releases that mostly competed on resolution and motion smoothness.
Why the consistency features matter for episodic content
Omni Reference now supports up to 10 keyframes for scene planning plus up to 15 reference assets total, including images, video, and audio. In practice, that means a creator can lock in a character's appearance, a location, and even a vocal tone, then generate multiple shots that all pull from the same reference set instead of drifting further from the original look with each new clip.
The voice-binding feature is the more unusual addition. Instead of generating a new synthetic voice per clip, a creator saves a voice to a subject, and that subject speaks in the same voice across different scenes. For anyone building a recurring cast rather than one-off clips, that closes a gap that's been a persistent weak point in AI video: dialogue-heavy scenes where the voice resets or drifts between shots.
Independent testing backs up the consistency claims to a degree. One reviewer tested a scene where a character's back was turned to the camera in the source image, forcing the model to invent a face and expression once the character turned around. Kling 4.0 handled it, maintaining stylistic consistency through a full dialogue exchange without breaking down. A separate side-by-side comparison found that Kling 3.0 tends to drift from reference images after the first few frames, especially when the reference conflicts with the model's default style, while 4.0 holds the reference across the entire clip.
The 30-second limit change, and what it actually saves you
Kling 4.0 doubles the previous native generation limit from 15 seconds to 30 seconds. That's a practical change more than a flashy one: it means a complete narrative beat, like a full line of dialogue plus a reaction shot, can happen in one continuous take instead of being stitched together from two or three separate generations.
For short-form creators, fewer stitched clips per episode means less time spent matching lighting, color, and motion between segments in post. It also reduces the number of times a character's look has to be re-established from scratch, since each new generation is a new opportunity for drift.
Pricing, access, and the catch worth knowing about
Kling 4.0 is priced per second of output: 15 credits per second at 1080p, which works out to 450 credits for a full 30-second clip, or 10 credits per second at 720p, or 300 credits for the same length. There's a free Basic plan, and non-subscribers can claim free credits monthly, valid for 31 days, but free-tier videos are capped below 1080p, carry a watermark, and can't be used commercially.
Access right now is through Kling's own web app only. The API for enterprise customers and developers was not live as of Kling's October 15 post, so any tool that wants to call Kling 4.0 programmatically, including automated production pipelines, will likely have to wait. Early rollout also wasn't smooth for everyone: comments under Kling's official Instagram announcement showed users in some regions, including Brazil, asking why they couldn't access it, alongside complaints about premium-only availability persisting well into October. The full October 15 release appears to resolve most of that, but it's worth checking regional and tier access before planning a workflow around it.
The pricing above doesn't account for rerolls, and that's the detail most worth remembering. One reviewer put it plainly: the real cost of AI video is the cost per usable clip, and if a 30-second shot needs three attempts to get right, the effective price triples. Budget for that reality rather than the sticker price.
What this means if you're building a recurring cast
If your short-form content relies on the same character or characters appearing across multiple episodes, the honest upside here is fewer regenerations per episode, not a magic fix. A 15-asset reference system and voice binding should meaningfully cut down on the trial-and-error of re-establishing a character's look and voice scene to scene, which is exactly where most AI video budgets quietly leak away in credits and editing time.
The tradeoffs are real too: no live API yet means no direct integration into automated workflows for now, and the reroll math means the actual cost of a consistent 30-second episode could be two to three times the sticker price if the first generation doesn't land. For creators evaluating whether to add Kling 4.0 into a production pipeline, including tools like Episodo that are built around solving character consistency, the web app is the only entry point today, and that's a reasonable place to start testing before committing budget to a full workflow.