Kling 3.0 is Kuaishou's AI video generation model that produces multi-shot cinematic sequences with native audio from a single text prompt. It is the first video model to offer storyboard-level control, where you can define individual shots, camera angles, and character dialogue all within one generation. This guide covers how to prompt Kling 3.0 for the best results, what changed from Kling 2.6, the full list of capabilities, technical specifications, and where it fits into different creative and commercial workflows. For a quick overview and steps to start generating, see the Kling 3.0 model page.
What is Kling 3.0?
Kling 3.0 is a video generation model released by Kuaishou in February 2026. It was built by merging two earlier models, Kling Video 2.6 and Kling O1, into a single unified architecture. Video 2.6 handled text-to-video and image-to-video generation with motion control. Kling O1 focused on visual quality and consistency. Kling 3.0 combines both into one model that generates video, audio, and element consistency in a single pass.
The result is a model that works less like a clip generator and more like a scene director. You describe a narrative in your prompt, and Kling 3.0 plans the shots, assigns camera angles, generates synchronized dialogue with lip-sync, and keeps characters visually consistent across every cut. Output supports durations from 3 to 15 seconds at resolutions up to native 4K.
On Morphic, Kling 3.0 is available as part of the video generation suite. You can use it in the same workspace as Morphic's image, music, and audio tools, which is useful when a project needs assets across multiple formats.
How to prompt Kling 3.0
The way you write your prompt changes everything about the output. Kling 3.0 is a video model, which means it responds to motion, timing, and camera direction, not just visual appearance. The prompts that produce the best results read like a scene description for a short film, not a caption for a photograph.
Here is a prompting framework for getting strong results across different types of video content.
1. Lead with camera language
The first words of your prompt set the visual tone for the entire generation. Kling 3.0 understands cinematic terminology and responds to it directly. Naming a specific camera behavior before describing anything else locks the model into a consistent visual approach.
| Bad prompt | Good prompt |
|---|---|
| "A woman walking through a city at night, cinematic look" | "Handheld tracking shot following a woman in a dark coat walking through rain-slicked city streets at night, neon reflections on the pavement, shallow depth of field" |
The first prompt leaves the camera behavior entirely up to the model. The second tells it exactly how to move: handheld, tracking, following the subject. It also grounds the scene with specific environmental details that inform the lighting and mood.
Camera terms that Kling 3.0 responds well to: tracking shot, orbital pan, macro close-up, POV, whip-pan, slow push-in, static wide shot, and handheld with subtle drift.
2. Structure multi-shot prompts with labeled shots
When you want multiple camera angles in one generation, label each shot explicitly. Kling 3.0 supports custom multi-shot mode where you define the number of shots, the duration of each, and what happens in the frame. The clearer your shot labels, the more precisely the model follows them.
| Bad prompt | Good prompt |
|---|---|
| "A man orders food at a restaurant, then the waiter brings the meal, then he eats" | "Shot 1: medium shot of a man in a navy shirt sitting at a restaurant table, scanning the menu, warm interior lighting. Shot 2: over-the-shoulder close-up of the menu in his hands, his finger pointing at an item. Shot 3: wide shot of the waiter approaching the table carrying a plate, the man looking up. Shot 4: close-up of the plate being set down on the table, steam rising from the food." |
The first prompt describes a sequence of events but gives the model no visual direction. The second breaks the narrative into distinct shots, each with a specific framing, subject position, and visual detail. This is what custom multi-shot mode is designed for.
3. Tag speakers directly with their dialogue
In scenes with dialogue, Kling 3.0 needs to know which character is speaking which line. Without explicit tagging, the model may assign voices to the wrong faces or create speaker confusion, especially with three or more characters.
| Bad prompt | Good prompt |
|---|---|
| "Two people sit at a cafe table and talk about their weekend plans and whether they should go hiking or stay in the city" | "A young woman in a white blouse and a man in a grey jacket sit at an outdoor cafe table. The woman lifts her coffee cup and says 'I was thinking we could do the coastal trail on Saturday.' The man leans back and replies 'That works, but we should leave early before it gets too hot.'" |
The first prompt summarizes the conversation topic without giving the model any actual dialogue or speaker identification. The second pairs each character with a physical description and their specific line, so the model can match lip movements and voice to the right face.
4. Use reference images to anchor characters
When you upload a reference image, Kling 3.0 uses it as a visual anchor throughout the generation. This is more reliable than describing a character's appearance in text alone, especially for maintaining consistency across multiple shots or separate generations.
To get the most out of references:
- Upload 2-4 reference images that show the character from different angles if possible. This gives the model more visual data to lock onto.
- If you upload a video reference, the model can extract both the character's appearance and their natural voice tone, keeping both consistent throughout the generation.
- For product videos, upload the product image as a reference to keep branding, text, and colors consistent during camera motion.
5. Describe motion and action over time, not static scenes
The most common mistake when prompting a video model is writing a prompt that describes a photograph. Kling 3.0 generates motion, so your prompt needs to describe how things change over the duration of the clip: how the subject moves, how the camera responds, and how the scene develops.
| Bad prompt | Good prompt |
|---|---|
| "A perfume bottle on a velvet surface with soft lighting and rose petals" | "Camera slowly orbits around a glass perfume bottle on a dark velvet surface, soft golden light catching the facets of the bottle as it rotates into view, scattered rose petals shift gently from the air movement, the camera gradually tightens from a wide framing to a close-up of the label" |
The first prompt describes a still image. The second describes how the camera moves, how the light interacts with the object over time, and how the framing changes. This gives the model a clear motion path to follow.
What's new in Kling 3.0
Kling 3.0 is a significant upgrade from Kling Video 2.6. The table below shows what changed, based on the official Kling 3.0 model documentation.
| Capability | Kling Video 2.6 | Kling Video 3.0 |
|---|---|---|
| Text-to-video | Yes | Yes |
| Image-to-video | Yes | Yes |
| Start and end frames-to-video | Yes | Yes |
| Native audio | Yes | Yes |
| Multi-shot generation | No | Yes |
| Start frame + element reference | No | Yes |
| Multi-character coreference (3+) | No | Yes |
| Multilingual support (Chinese, English, Japanese, Korean, Spanish) | No | Yes |
| Dialects and accents | No | Yes |
| 15-second output duration | No | Yes |
| Flexible duration (3-15 seconds) | No | Yes |
| Native 4K resolution | No | Yes |
The most notable additions are multi-shot generation and the element reference system. Multi-shot allows up to six camera cuts in a single generation, which eliminates the need to generate individual clips and stitch them together manually. The element reference system lets you bind a character's visual appearance and voice tone to a reusable element, so consistency carries across shots and even across separate video generations.
Multilingual support with dialect and accent rendering is also new. Kling 2.6 supported native audio, but 3.0 extends this to five languages with the ability to replicate specific accents (American, British, Indian for English; Cantonese, Northeastern, Beijing, Sichuanese, Taiwanese for Chinese) and handle code-switching within a single scene.
Kling 3.0 capabilities
Multi-shot storyboard generation
Kling 3.0 offers two modes for multi-shot video. In auto mode, you enable the multi-shot toggle and the model reads your scene description to plan camera transitions, shot framing, and pacing on its own. In custom mode, you define each shot individually, specifying duration, camera angle, and narrative content. The model follows your storyboard exactly.
Custom mode is especially useful for structured content like product ads or dialogue sequences where the timing of each cut matters. Auto mode works well when you want the model to interpret a narrative prompt and decide the visual coverage.
Native audio with character-specific voice binding
Video and audio generate in a single pass. The model produces lip-synced dialogue, and you can control which character speaks which line by pairing characters with their dialogue in the prompt. Beyond basic lip-sync, Kling 3.0 supports creating character elements with bound voice tones. Once you bind a voice to a character element, that voice stays consistent every time the character appears, without needing to re-specify it.
The model supports dialogue in English, Chinese, Japanese, Korean, and Spanish, with dialect and accent support and multilingual code-switching within a single scene.
Element reference system
You can create reusable character elements by uploading 2-4 reference images or a short reference video. For character elements, you can also assign a voice tone by uploading audio or selecting from available voices. When you use an element in a prompt, the model locks the character's appearance and voice throughout the video, maintaining consistency even through camera movements, scene changes, and multi-shot sequences.
This system supports three or more distinct characters in the same frame without blending features, which is critical for dialogue scenes and any video with multiple people.
Text and logo preservation
The model can identify text content in uploaded images, such as signs, product labels, or logos, and maintain text consistency throughout the video. It can also generate new text content within the video itself. Text stays legible even during continuous camera movement, which is useful for commercial content where brand elements need to remain sharp and readable.
Flexible duration and resolution
Kling 3.0 generates video from 3 to 15 seconds in a single pass, with support for resolutions up to native 4K. The extended duration gives the model room for more complex narrative development, scene transitions, and action sequences that would not fit in shorter clips. Resolution options also include 1080p and 720p.
Kling 3.0 technical specifications
| Specification | Details |
|---|---|
| Generation modes | Text-to-video, image-to-video, start and end frames-to-video |
| Maximum duration | 15 seconds |
| Minimum duration | 3 seconds |
| Maximum resolution | Native 4K |
| Other resolutions | 1080p, 720p |
| Aspect ratios | 16:9, 9:16, 1:1 |
| Multi-shot | Up to 6 camera cuts per generation |
| Multi-shot modes | Auto (model plans shots) and Custom (user defines each shot) |
| Native audio | Lip-synced dialogue, voice tone control |
| Supported languages | English, Chinese, Japanese, Korean, Spanish |
| Dialect and accent support | Yes (Chinese and English dialects, regional accents) |
| Code-switching | Yes (multiple languages in one scene) |
| Character elements | Created from 2-4 images or video reference |
| Voice binding | Voice tone bound to character elements |
| Multi-character coreference | 3+ distinct characters in one frame |
| Text preservation | Reads and maintains text from uploaded images |
| Model lineage | Unified from Kling Video 2.6 + Kling O1 |
| Release date | February 2026 |
Kling 3.0 use cases
Short-form filmmakers and narrative creators
Multi-shot generation is what makes Kling 3.0 particularly useful for short narrative content. You can generate a complete scene with shot-reverse-shot dialogue, establishing shots, and close-ups in a single pass. For creators working on short dramas, micro-series, or story-driven social content, this removes the manual work of generating individual clips and editing them together. The 15-second duration with up to six cuts gives enough room for a beginning, middle, and payoff within a single generation.
Product and e-commerce video
Product ads need the camera to move around an object while brand text and logos stay sharp. Kling 3.0's text preservation handles this natively, keeping labels legible during orbital shots and tracking movements. Combined with the element reference system, you can lock a product's visual identity and generate multiple ad variations with different camera angles, lighting setups, or background environments while the product itself stays consistent. On Morphic, you can generate the product video and then create matching thumbnails or social assets in the same workspace.
Social media content teams
The combination of flexible aspect ratios (16:9, 9:16, 1:1) and quick iteration means you can generate platform-specific video content without separate production workflows for each format. Multi-shot mode with auto storyboarding is useful here, where you describe the content concept and the model handles the shot planning. For teams that need volume across Instagram, TikTok, YouTube Shorts, and feed posts, this speeds up the creation cycle significantly.
Multilingual and localized content
The dialect and code-switching support opens up use cases that most AI video models cannot handle. A training video where a presenter speaks Korean, a tourism ad where characters switch between English and Spanish mid-conversation, or a social clip featuring authentic regional accents all generate with natural lip movements and coherent facial expressions. For brands targeting multiple markets, this means producing localized video content from the same prompt framework without re-recording audio.
On Morphic, you can pair Kling 3.0 with the platform's image and audio tools to build a complete content package, from video to thumbnail to background music, without switching between separate applications.
Frequently asked questions
Kling 3.0 is available on Morphic. Sign up for a Morphic plan, select Video mode from the prompt bar, and choose Kling 3.0 from the model dropdown. It sits alongside image, music, and audio generation tools, so you can work across multiple content types in one workspace.
Both models handle text-to-video and image-to-video, but they serve different use cases. Kling 3.0 is the core generation model with multi-shot storyboarding and native audio. Kling 3.0 Omni extends that with deeper element consistency controls, video-based character references, and voice tone binding. If you need a single polished video from a prompt, Kling 3.0 is the right choice. If you are building a series where the same characters appear across multiple generations, Omni gives you the consistency tools to maintain that.
The model generates lip-synced dialogue in five languages: English, Chinese, Japanese, Korean, and Spanish. It goes beyond basic language support with specific dialect and accent rendering, including American, British, and Indian accents for English, and Cantonese, Northeastern, Beijing, Sichuanese, and Taiwanese dialects for Chinese. Characters can also switch between languages mid-conversation within the same clip.
Multi-shot generates up to six distinct camera cuts within a single video. You have two options: auto mode, where the model plans the shot transitions based on your prompt, and custom mode, where you define each shot's framing, duration, and camera angle yourself. In custom mode, the model follows your storyboard exactly. In auto mode, it interprets your narrative and decides the best shot coverage. Both modes maintain character consistency across all cuts.
The maximum resolution is native 4K, meaning the video is generated at that resolution rather than upscaled. 1080p and 720p are also available for faster generation or smaller file sizes. Duration ranges from 3 to 15 seconds per generation. Supported aspect ratios are 16:9, 9:16, and 1:1, covering widescreen, vertical, and square formats.
