How to prompt Kling O3: start from a first frame and write the motion and the camera, not the picture. A reusable template, three example prompts and mistakes.
Kling O3 is Kuaishou's image-to-video model in Cinemagiq. You give it a still, and the prompt tells it what happens next: what moves, how the camera behaves and when something changes. This guide covers how Kling O3 reads its inputs, a motion-first template, three example prompts with the technique behind each, and the mistakes worth avoiding.
Kling O3 needs a first frame. That still sets the subject, composition, light and style, so the prompt should not describe them again. A last frame is optional; add one when the shot must end on a specific image. Kling O3 References takes up to 4 reference images, with first and last frames optional, for shots where a character or prop has to match other material.
Clips run from 3 to 15 seconds, in 16:9, 9:16 or 1:1. Sound is optional, so you decide per clip whether the model adds audio.
Two related lanes cover other jobs. Kling O3 Video Edit edits an existing clip by prompt, with up to 4 reference images. Kling V3 Motion Control makes a character image perform the movements of a reference video.
[Main subject motion], [secondary motion in the scene], [camera instruction + speed + purpose], [one timed event, if any]. [Sound: ambience and effects, when sound is on.]
Write it in the present tense, in the order things happen. Name objects that are already in the frame ("the camels", "the window") so the model knows what to move. Leave out style words and descriptions of the picture, because the first frame already has them.
Made in Cinemagiq. The exact prompt is under each one.
PromptWaves crash against the rocks and spray over the cottage, rain lashes the window, the camera holds steady, lightning flashes once in the distance.
First frame made with Nano Banana Pro in Cinemagiq.
PromptRack focus from the raindrops on the glass to the people crossing the street, umbrellas moving, neon reflections rippling in the puddles.
First frame made with FLUX.2 Pro in Cinemagiq.
PromptThe mechanical camels march forward across the dunes, gears turning, sand blowing off the ridge, slow crane up revealing the endless desert.
First frame made with Nano Banana 2 in Cinemagiq.
Example 1 animates a storm still made with Nano Banana Pro in Cinemagiq. It stacks environmental motion ("Waves crash against the rocks", "rain lashes the window"), then locks the camera with "the camera holds steady" so all the energy stays in the weather. The single event, "lightning flashes once in the distance", is limited with "once", which asks for one beat rather than a repeating effect.
Example 2 starts from a rainy street still made with FLUX.2 Pro in Cinemagiq. Its camera move is a focus change, not a physical move: "Rack focus from the raindrops on the glass to the people crossing the street". That suits this frame, which was composed with raindrops in the foreground and the street behind them. The rest of the prompt adds small motion ("umbrellas moving, neon reflections rippling in the puddles").
Example 3 animates concept art made with Nano Banana 2 in Cinemagiq. Subject motion comes first ("The mechanical camels march forward across the dunes"), with detail motion behind it ("gears turning, sand blowing off the ridge"). The camera move ends the prompt and has a purpose: "slow crane up revealing the endless desert".
For the full vocabulary of moves and lenses, see the prompt engineering cheat sheet.
Not in its main lane. Kling O3 needs a first frame, and a last frame is optional. Kling O3 References can work from up to 4 reference images, with first and last frames optional. For framed shots, make the still first, then use the prompt for motion, camera and sound.
Kling O3 runs from 3 to 15 seconds per generation, in 16:9, 9:16 or 1:1. Short lengths suit inserts and reaction shots, and longer ones give a camera move room to finish. Sound is optional, so you can switch it on or off per clip.
Name the move the way an operator would: hold steady, slow push-in, pan, crane up or rack focus. Give it a speed and a purpose, such as 'slow crane up revealing the endless desert', so the move has somewhere to go.
Yes, through Kling O3 Video Edit, which changes an existing clip by prompt and can use up to 4 reference images. For performance, Kling V3 Motion Control makes a character image perform the movements of a reference video.
Run Kling O3 and the other leading models in one AI-native workspace.