Kling 4.0 Explained: Keyframes, Omni Reference, and What Changed in AI Video
Kling 4.0, announced September 28, 2026 by Kuaishou's Kling AI team, is the most control-oriented text-to-video release so far. It doubles the maximum clip length to 30 seconds in a single pass, adds up to 10 keyframes per generation, and introduces "Omni Reference" — up to 15 multimodal references that lock characters, products, and lighting across shots. If you build with video models, the…
Kling 4.0, announced on September 28, 2026, is a major upgrade in control-oriented text-to-video technology. It expands the maximum clip length to 30 seconds and introduces up to 10 keyframes per generation, allowing developers to choreograph the video's progression using individual storyline panels. Additionally, Kling 4.0 includes Omni Reference, enabling up to 15 multimodal references that maintain consistency across shots for characters, products, and lighting.
These features shift the focus from "prompt and pray" to a more directed approach, resembling the process of creating a shot list. This change brings AI video generation closer to traditional video production techniques. The release of Kling 4.0 comes in two tiers: 4.0 Flash, a limited early access version with up to 20 seconds per pass at 720p, and the full version with 30-second clips, 10 keyframes, and the complete reference system, scheduled for public release in October 2026.
While the full model's release date is not set, 4K output at 10-bit HDR is expected to follow. For builders considering video models for their products, this upgrade emphasizes the importance of iteration, reference asset preparation, and planning for multiple attempts per finished clip.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
