Kling v3: multi-shot scenes and performed dialogue
Kling v3 will cut between shots inside a single clip, hold fast and complex motion together, and perform quoted dialogue with the emotion you describe.
Real results
Made in solorender with these models. Nothing retouched.
Best for
- ✓Multiple shots in one generation — a standoff, a chase, a scene with a cut in it
- ✓Complex physical motion other models fall apart on: sport, riding, fighting, anything where two subjects must stay connected
- ✓Performed dialogue with an accent and an emotional arc, including the listener's reaction
- ✓Explicit camera grammar — dolly zoom, crane up, tracking shot — carried out as described
Where it struggles
- —Long prompts genuinely help here, which is the opposite of the advice for Nano Banana 2. Describe each beat in order
- —Kling v3 Pro carries native audio; the Turbo tiers are the cheaper way to find the shot first
Prompts that work
Each one opens the studio with the prompt already in the box. They are also one click away inside the app, under the prompt field.
Multi-shot scene
Try it →An action-packed movie scene with multiple shots depicting the final standoff between two characters.
One sentence, several shots, cuts included. Very few models do this in a single pass.
Dialogue with a cut
Try it →The elf says softly, "This crystal's power can't be contained by me alone." The camera cuts to an armoured dwarven man sitting opposite her, who says angrily, "Last time you asked me for help, I lost her. I almost lost everyone."
Keeps the styling of your start image, performs both lines, and cuts between them.
Accent and reaction
Try it →The man looks out to sea, pauses for a moment, then turns to a sailor. He tells him, in a northern British accent and a worried tone, that a huge storm is coming. The sailor's face fills with subtle panic and he asks what they should do.
Describing the listener's reaction is what makes it read as a scene rather than a shot.
Hard motion
Try it →First-person POV of the man riding the dragon. The dragon makes a sharp turn and the man almost falls off.
The test most video models fail — they let the two subjects drift apart. This one holds them together.
Camera move
Try it →Dolly zoom in with a lighting shift to blue as the man's expression turns from worried to horrified.
Cut on an entrance
Try it →The camera follows the unicorn as it runs along the path. The scene cuts to two mages standing on the path as the unicorn passes them. They clear the way for the creature, then look at each other in awe.
Reveal
Try it →She turns around and gasps as the camera cranes upward slightly, revealing a zombie approaching her on the road.
Stillness
Try it →A serene mountain landscape at sunrise, mist rolling through the valleys, cinematic camera movement.
Write beats, not a description
Kling v3 rewards a prompt written as an ordered sequence of events. Who does what, then what the camera does, then who reacts and how. Each clause is a beat the model schedules inside the clip.
This is why a wordy prompt outperforms a terse one here, and why it is worth spelling out the reaction shot. Left to itself the model will give you the action; told about the reaction, it gives you the scene.
Find the shot on Turbo
Kling v3 Turbo and Kling v3 Turbo Pro answer the same prompts as the full Pro model. Iterate there, where a miss costs less, and move to Pro for the take you intend to keep — that is also where you get native sound.
Run these for real
Every model in this guide is live in solorender. Credits never expire.
Open Kling v3 Turbo Pro →