What’s new
- Synchronized audio-video generation: clips come with matching sound built in, no separate audio pass needed — Kuaishou lists this as a core 3.0-series capability.
- Multi-shot storyboarding: up to 5 shots in one generation, each with its own prompt, so a whole short story can play out in one take.
- Element references: tag up to 3 reference elements with
@namesyntax — people, products, or sounds — and the model keeps them consistent throughout the clip. - Native 4K output: Kling AI said in April 2026 it was the first AI video model with native 4K output, and the 3.0 series can render directly at that tier.
What it’s good for
- Multi-shot shorts that need a real beginning, middle and end: ads, product films, short-drama scenes.
- Projects where a character or product has to look the same in every shot — pin it down with element references.
- Clips that need built-in sound instead of a separate audio pass afterward.
Prompting tips
- Pick a quality tier first: std, pro, or 4K — higher tiers hold more detail but render slower.
- State duration in seconds: 3–15 seconds per task; in multi-shot mode, split that total across the shots explicitly.
- Turn sound on when you want it, and describe the actual sound rather than just saying "with audio".
- Reference elements with
@name: up to 3 elements per task, each holding 2–4 images, one 3–8 second video clip, or one 5–30 second audio clip. - Skip the aspect ratio when you supply an image: it's picked up automatically from the reference image.
- Mark shot boundaries clearly in multi-shot mode: one sentence per shot, not several scenes crammed into one line.
16:9, pro quality, 15 seconds, three connected shots, sound on.
Reference @girl: same face, same red coat, consistent throughout.
Shot 1 (0-5s): early morning, she steps out of a café holding a coffee cup.
Shot 2 (5-10s): medium shot, she sits on a bench as the wind lifts her hair.
Shot 3 (10-15s): wide shot, she looks down the street and smiles.
Sound: street ambience, wind, distant traffic. No extra people in frame.
Known limits
- Multi-shot mode only supports a first-frame image — it can't be combined with first-and-last-frame mode.
- Each task can reference at most 3
@elements, and each element has its own caps (2–4 images, one video clip, one audio clip). - Higher quality tiers, especially 4K, take longer to render — the docs don't give an exact multiplier.
Using it in Nomi
On the generation canvas, add a video card, pick “Kling 3.0” in the model picker and choose a mode (Text to video, First frame to video, First and last frame). Add your references, write the prompt and generate. In the storyboard you can pick it per shot. The first time, connect any one of the providers above under Settings → Models.
