Kling AI 3.0: How to Use It — Settings, Sound and Multi-Shot

Kling 3.0 (Kling AI) is a video model that turns an uploaded frame into a clip of 3 to 15 seconds, following the motion described in the prompt, and generates sound together with the picture. In the IIshki studio it renders at 720 or 1080, in 16:9, 9:16 and 1:1, with a “Multi-shot” mode of up to 5 scenes and a transition between a first and a last frame. Below: how to use it, how to write a prompt the model actually follows, and where people most often trip up.

What Kling 3.0 can do in the IIshki studio

Kling 3.0’s main strength is controllable motion: the model follows the description of what happens in the frame and how the camera behaves. That is why it is the usual pick when you need to animate a specific photo — a portrait, a product, a location — and get a predictable result rather than a random scene.

Note the first row: in the studio Kling 3.0 works from an image only. That is not a limitation but how it works: the frame sets the character, light and composition, the prompt sets the motion.

How to use Kling 3.0: 6 steps

1. Prepare the start frame

The model animates what it sees. Upload a clean, uncompressed image with one main subject and a clear composition. If the source is small or old — upscale it first. If you have no frame, generate one in Nano Banana Pro or GPT Image 2 and pass it to Kling.

2. Describe the motion, not the picture

Anything already in the frame doesn’t need describing — the model can see it. The prompt answers “what changes over these seconds”:

The woman in the frame slowly turns her head toward the camera and smiles, a light breeze moves her hair, static camera, soft daylight.

3. Set the camera move — one per clip

Kling understands cinematography terms well. One move per shot, named explicitly: “slow push-in”, “camera pulls back”, “pan left to right”, “camera follows the character”. Two moves at once is a reliable way to get a shaky picture.

4. Pick duration and quality

5 seconds is the working standard for a single action; take 8–10 when something develops in the frame (the character stands up, walks over, turns around). 720 is for drafts and social media, 1080 for the final clip. The price grows with both duration and quality — it is shown on the card before you start.

5. Decide whether you need sound

The “With sound” switch adds atmosphere and ambient noise matched to the scene. If you plan to add music or a voice (ElevenLabs, MiniMax), turn it off — a clean track is easier to edit.

6. Run and review

Generation takes a few minutes; the result is saved under “My generations” and, if Telegram is linked, arrives as a file in the bot. Not happy — change the prompt, not the frame: nine times out of ten it’s enough to soften the motion or remove an extra detail.

How to write a prompt for Kling 3.0

This section covers only the structure. Twenty ready-made examples for portraits, products, landscapes and multi-shot are in a separate article: Kling 3.0 prompts.

A prompt structure that works — four blocks, in this order:

1. Who does what — subject and action in verbs: “a barista pours milk into a cup”.
2. Camera — one move: “slow push-in on the cup”.
3. Light and mood — “warm morning light from the window, light steam”.
4. Sound (if enabled) — “coffee shop noise, clinking cups”.

A barista pours milk into a latte cup, drawing a pattern; slow push-in on the cup; warm morning light from the window, steam over the cup; quiet coffee shop noise.

A sports car parked on a wet night street; the camera slowly arcs around it; neon reflections on the body, light rain; the sound of rain and a distant city.

A cat on a windowsill turns to the window and watches a bird; static camera; soft daylight; silence, a faint rustle.

Camera moves Kling understands

What to avoid in the prompt: negations (“no blur” reads as “blur”), several camera moves at once, descriptions of things not in the frame, and abstract words like “beautiful” or “epic” — they say nothing about motion.

Multi-shot: a clip from several scenes

“Multi-shot” mode assembles up to 5 scenes into one clip of up to 15 seconds. Two ways:

• Your own shots — you write the prompt and duration for each scene (up to 500 characters and up to 12 seconds per scene). This is how ad clips and storytelling are made: “wide shot — detail — reaction”.
• Auto mode — you write one overall prompt and the model builds the storyboard itself. Faster, with less control; good for first tries.

Tip: keep one character and one location for the whole clip — that way the scenes cut together without the character “jumping”.

First and last frame

If you also upload a last frame, Kling builds a transition between the two images: morning → evening, sketch → finished product, a face → the face after make-up. Both frames should match in composition and angle — otherwise the model “plays out” the transition as an abrupt rebuild of the scene. To reproduce someone else’s choreography or gesture exactly there is a separate mode — Kling Motion Control, which transfers motion from a reference video.

Common mistakes and how to fix them

Kling 3.0 or another model

• You need sound with speech and a cinematic picture from text alone — Veo 3.1; see the detailed Veo 3.1 vs Kling 3.0 comparison.
• You need a long clip of up to 30 seconds with several references — Seedance 2.5.
• You need to animate a frame with sound quickly and cheaply — Grok Imagine 1.5; comparison with Kling.
• You need precise control over the motion of your own image — Kling 3.0, and that is its main use case.

What’s next

Start with one portrait, 5 seconds, 720 and a two-sentence prompt — and see how the model reads your description. Then add camera motion, sound and multi-shot. If the task is specifically to animate a photograph, the step-by-step breakdown with model choice is in how to animate a photo with AI, and the roundup of video models for different jobs is on best AI models for video.