How to Use Kling 4.0: Modes, Keyframes, References and Prompts

Kling 4.0 (Kling AI) is Kuaishou's video model that makes a 3 to 30 second clip in one generation, with sound and lip-synced speech included. Using it comes down to three choices: pick a mode — Keyframes (up to 10 ordered frames), References (a character photo, a video with the camera move you want, a voice recording) or Edit (change an existing clip) — set duration, aspect ratio and 720p or 1080p, then describe what happens in the prompt. In the IIshki studio, Kling 4.0 runs in the browser with no foreign card needed. Below: which model fits which job, a step-by-step walkthrough of each mode, working prompts, and the mistakes that most often waste a generation.

What Kling 4.0 is and what's new

Kling 4.0 is the fourth generation of Kling video models. The headline change is length and control: up to 30 seconds in one pass (Kling 3.0 tops out at 15), plus ways to steer how the clip develops over time.

What the model does on IIshki:

• Duration from 3 to 30 seconds, in one-second steps.
• Resolution 720p or 1080p.
• Aspect ratios 16:9, 9:16, 1:1, wide 21:9, or "match files", which takes the ratio from your uploaded frames.
• Sound and speech are generated with the video: footsteps, rain, character lines with lip sync.
• Up to 15 files per generation in References mode: up to 10 photos, up to 5 videos (30 seconds in total) and up to 3 voice tracks (30 seconds in total).
• Prompt up to 3,000 characters.

Kling 4.0 or Kling 3.0

More on the previous version in the Kling 3.0 guide and the prompt collection.

Which model to pick for the job

The full list of video models and their strengths is in the best AI video generators roundup.

How to use Kling 4.0, step by step

Step 1. Choose a mode

Open Kling 4.0 in the studio and pick the input mode. Modes don't mix: if you upload both keyframes and photo references, the studio asks you to keep one before any tokens are charged.

• Keyframes You have images of the key moments and want the clip to pass through them in order.
• References You want to keep a character, product, style, camera move or voice, and let the model build the scene.
• Edit The clip already exists and something in it needs replacing.

Step 2. Keyframes mode: plan the key moments

Don't upload 10 frames just because you can. A keyframe belongs where the picture genuinely changes: the character walks in, the product appears, night falls, the location switches. If nothing important happens between two frames, the extra one only gets in the model's way.

A plan for a 20-second ad:

Mind the limit: shorter clips allow fewer frames — up to 4 at 3 seconds, 6 at 5 seconds, 9 at 8 seconds, all 10 from 9 seconds. Generate the frames in advance with Seedream 5.0 Pro from one shared reference so the character and lighting match.

In this mode, the prompt describes what happens between the frames, not the frames themselves:

Morning in the kitchen, the camera glides slowly along the table. A hand sets down a mug, steam rises from the coffee. A woman walks to the window and takes a sip, soft sunlight. The camera lowers to the coffee pack on the table. Quiet clink of dishes, birds outside.

Step 3. References mode: name every file

Upload photos of the character, product or style, and if needed a video with the camera move and a voice recording. In the prompt, refer to files in upload order: @Image1, @Image2, @Video1, @Audio1. The model doesn't guess why you gave it a file — state the role in words.

@Image1 walks down a night street in the rain, the camera follows smoothly, neon reflections in the puddles, sound of footsteps and rain.

Repeat the camera move from @Video1, but the frame shows the woman from @Image1 in a red coat on a rooftop at sunset.

The man from @Image1 sits in an armchair by the fireplace and speaks in the voice from @Audio1, calmly, looking into the camera. The camera slowly pushes in.

For a specific line without your own recording, put it in quotes and say who speaks and how: "The woman from @Image1 smiles and says: 'This is my favourite city.'"

Step 4. Edit mode: one clip, one change

Upload one clip and a photo of what should go into the frame, then describe the swap: "Replace the mug in her hand with the bottle from @Image1, keep everything else as is." To keep the original audio, turn on "Sound from video". One change per generation works far more reliably than a list of five.

Step 5. Settings and run

Pick duration (3 to 30 seconds), aspect ratio and resolution. A tip from reviews that holds on IIshki too: run a short 5–8 second draft in 720p first to check the model got the character and motion right, then go for full length and 1080p. The generation cost is shown on the model card before you start. The finished clip lands in your generations, and if you signed in with Telegram it is also sent to your chat.

How to write prompts for Kling 4.0

• Write along the timeline. For anything over 10 seconds, split the action into 3–4 sentences in the order events happen; the model holds the story better when it sees a sequence.
• One camera move per segment. "The camera follows slowly", "static camera", "the camera rises over the rooftops". Three camera moves in one sentence give you a jittery shot.
• Describe the sound. List what should be heard: footsteps, rain, café noise, a spoken line. If you don't, the model picks its own.
• Describe motion, not a still. "A beautiful woman in a coat" is a photo. "The woman turns to the camera and straightens her collar" is a video.
• Repeat the character's traits. In a long clip, briefly remind the model who is on screen ("the same woman in the red coat"), especially after a location change.

More video prompt templates in the Seedance 2.5 prompt guide: the "who → action → camera → sound" structure works for Kling 4.0 as well.

Common Kling 4.0 mistakes

When to use Kling 4.0 and when to pick something else

Kling 4.0 is our best pick when the clip is long and tells a story: an ad with several beats, a character's dialogue, a scene that moves from day to night. For a short 5-second shot, Kling 3.0 is quicker and simpler. If several characters interact, compare the result with Seedance 2.5 — the differences are covered in Seedance vs Kling. To chain several steps — frames, animation, voice-over — use the Canvas.

What's next

1. Open Kling 4.0 and make a 5-second draft in References mode with one character photo.
2. Once the character and motion look right, extend the duration and add keyframes or a voice.
3. To recreate a viral dance with your own character, try Swap — no prompt needed.