How to use Kling 4.0: prompts, 21:9, keyframes and Omni Reference

What Makes Kling 4.0 a Major AI Video Upgrade

0:00 to 1:06 in the video, about 1 minute

Kling 4.0 introduces the ability to generate up to 30-second video clips in a single pass, moving past the short, disjointed clips common to earlier generation models. It natively incorporates camera movements, synchronized sound effects, and character dialogue with built-in lip sync, all following instructions provided in the text prompt.

Because Kling only refunds credits if a generation crashes or completely fails technically, any generated clip that turns out incorrectly still charges the user full price. Creating precise prompts and utilizing the new reference tools is essential to getting desired results on the first attempt without wasting expensive subscription credits.

• Generates up to 30 seconds of video natively in one pass.
• Integrates video, sound effects, and lip-synced spoken lines.
• Failed generations refund credits automatically, but flawed generations do not.

The Four-Part Prompt Formula to Prevent AI Guesswork

1:07 to 2:14 in the video, about 1 minute

Whenever information is left out of an AI prompt, the model fills in the gaps on its own, which often leads to unwanted surprises and burnt credits. Kling 4.0 performs best when given a structured, four-tier text prompt that dictates what happens from second to second.

The formula starts with a main concept line defining the style, subject, and hook. The body consists of timestamped segments specifying one clear action and one specific camera movement per interval. Finally, dedicated lines define atmospheric audio and a strict list of elements to avoid, such as warped limbs or distorted faces.

• Opening line: Specifies the subject, visual aesthetic, and central action.
• Timestamp blocks: Outlines second-by-second changes with dedicated camera angles.
• Dialogue insertion: Enclosing spoken words in quotes within timestamps triggers automatic lip-syncing.
• Sound and Avoid tags: Explicitly calls out required audio cues and artifacts to ban from the frame.

Testing Five Video Styles Across 30-Second Shots

2:15 to 8:03 in the video, about 6 minutes

To test how Kling 4.0 handles diverse aesthetics, five distinct genres were rendered across full 30-second lengths using the structured prompting method.

The model excels noticeably with stylized visuals. A 2D cel-shaded anime sequence of a bike messenger showed consistent frame-to-frame lines despite rapid motion, and a 3D animated lighthouse sequence maintained feature-film lighting without introducing unwanted weather. In live-action scenarios, such as a handheld UGC selfie vlog, small natural imperfections like camera shake and breath mist enhanced realism. However, intense, complex fantasy action scenes with dragons can sometimes take on a plastic sheen, requiring extra prompt keywords for texture and lighting to resolve.

• Cel-shaded 2D Anime: Maintains stable lines and character design during fast movement.
• 3D Feature Animation: Accurately renders volumetric lighting and complex weather rules.
• UGC Phone Video: Simulates believable camera shakes, whispered audio, and lighting changes.
• Cinematic Fantasy: Handles dense choreography and lip-synced yells, though heavy action can require texture fine-tuning.

Framing for the 21:9 Ultrawide Aspect Ratio

8:04 to 9:38 in the video, about 2 minutes

Kling 4.0 supports a 21:9 ultrawide cinematic aspect ratio alongside standard 16:9 and vertical formats. Standard prompting tends to concentrate the action in the center of the screen, which leaves the wide edges of a 21:9 canvas empty and underutilized.

To properly use the widescreen format, write prompts that send objects across the entire frame. For example, describing a starship entering from the far-left edge, sliding across the screen, and kicking up debris across the entire width allows the model to leverage the broad framing effectively.

• Widescreen framing requires prompting horizontal motion rather than centered push-ins.
• Use duration controls to match the narrative length; a 20-second shot avoids the credit cost of generating an unused 30-second clip.

Using Keyframes for Precise Scene Transitions

9:39 to 11:01 in the video, about 1 minute

While text prompting defines the overall trajectory, keyframing allows creators to lock down the exact visuals at crucial turning points. In Kling 4.0, users can define up to 10 keyframes across a sequence by uploading images directly into the generation bar.

By setting first, middle, and last frame images, the user dictates the composition, and the prompt merely outlines how the camera transitions between those fixed points. Keyframing works most reliably when the movement between images is continuous and logically simple, significantly cutting down on character and object morphing errors.

• Keyframe mode allows between 3 and 10 image anchors within a single shot.
• Guides the AI on the exact starting, midpoint, and ending frames of an action.
• Prevents mid-sequence hallucination and saves credits on rerolls.

Preserving Character and Scene Continuity with Omni Reference

11:02 to 12:58 in the video, about 2 minutes

Maintaining a consistent character appearance over 30 seconds can be challenging in AI generation. Kling 4.0 solves this with the Omni Reference feature, allowing creators to upload a reference image sheet and tag it directly in the prompt using an '@' mention.

A practical method for character sheets is providing three angles: front, back, and a face closeup. Leaving the head off the full-body mannequins forces the system to pull facial features solely from the closeup portrait, locking in identity throughout dynamic motion. The prompt should reinforce the sheet by repeating key clothing and physical descriptions.

• Upload turnaround character sheets and reference them via @Image tags.
• Use headless front/back outfit photos paired with a face closeup to avoid facial morphing.
• Omni Reference ensures character faces remain identical even during dramatic actions.

Advanced Multi-Reference Workflows

12:59 to 15:05 in the video, about 2 minutes

Kling 4.0 can handle up to 15 different visual references in a single 30-second clip. This capability makes it possible to construct multi-scene story sequences featuring hard cuts and multiple environments within a single generation pass.

In this setup, each uploaded reference serves a specific role: one for the character, one for a hero prop, and several for distinct background locations. By assigning each image an sequential reference tag (@Image1, @Image2, etc.) and scripting hard cuts between timestamps, the AI teleports the persistent character across diverse worlds without losing visual fidelity.

• Supports up to 15 image references in one generation run.
• The upload order directly dictates the numbered @Image tags used in the prompt.
• Enables multi-shot narratives with hard cuts between totally different environments in 30 seconds.

Final Thoughts

Kling 4.0 represents a significant leap forward for AI video production by uniting extended 30-second lengths, integrated audio, native lip syncing, and advanced asset referencing into a single platform. The key to maximizing its value lies in eliminating ambiguity from every generation.

By adopting a structured timestamp prompt formula, drafting concepts at 720p, and leveraging Omni Reference sheets for characters and props, creators can produce cinematic, narrative video without wasting credits on endless rerolls.