Skip to main content

Kling 3 Omni integration

Generate AI videos with multi-modal inputs: text prompts and reference images.
Kling 3 Omni is a versatile video generation API that supports multiple input modes: text-to-video and image-to-video with reference images. It offers advanced features like multi-shot mode for scene-by-scene control and reference images for style guidance.
Looking for video-to-video? Use the dedicated Reference-to-Video endpoints to generate videos from a reference video using video_url.

​
Key capabilities

  • Text-to-video: Generate videos from text prompts up to 2500 characters
  • Image-to-video: Use image_url for start frame, end_image_url for end frame control
  • Reference images: Add style guidance with image_urls, reference as @Image1, @Image2 in prompts
  • Multi-shot mode: Create multi-scene videos with multi_prompt for shot-by-shot control
  • Duration control: Generate videos from 3-15 seconds
  • Audio options: Generate native audio or use voice IDs for narration
  • Async processing: Webhook notifications or polling for task completion

​
Generation modes

​
Standard, Pro and 4K

4K is generated natively by the model at 3840x2160. It is not an upscale of a lower-resolution output, so there is no additional processing step and no loss of detail from interpolation.
All three accept the same request body: the resolution is fixed by the endpoint you call.

​
Use cases

  • Character animation: Animate a character image with controlled, consistent motion
  • Product visualization: Animate product images with controlled motion
  • Storyboarding: Create multi-scene videos with shot-by-shot prompts
  • Style transfer: Apply visual style from reference images to generated content
  • Marketing content: Generate promotional videos from brand imagery

​
Generate videos with Kling 3 Omni

Create videos by submitting prompts with optional images. The service returns a task ID for async polling or webhook notification.

POST /v1/ai/video/kling-v3-omni-pro

Generate video with Kling 3 Omni Pro

POST /v1/ai/video/kling-v3-omni-std

Generate video with Kling 3 Omni Standard

POST /v1/ai/video/kling-v3-omni-4k

Generate native 4K video with Kling 3 Omni

GET /v1/ai/video/kling-v3-omni

List all Kling 3 Omni tasks

GET /v1/ai/video/kling-v3-omni/{task-id}

Get task status by ID

​
Video-to-video with reference video

For video-to-video generation using a reference video, use the dedicated Reference-to-Video endpoints. These endpoints accept video_url and let you reference the video in your prompt as @Video1.
Audio works differently here. The Reference-to-Video endpoints do not generate new audio, so generate_audio does not apply to them. The output carries the audio track of the reference video, kept by default and controlled with keep_audio:If you need generated audio, use the Pro or Standard endpoints with generate_audio: true.

POST reference-to-video Pro

Generate video from reference video (Pro)

POST reference-to-video Std

Generate video from reference video (Standard)

GET reference-to-video tasks

List all reference-to-video tasks

GET reference-to-video task

Get reference-to-video task by ID

​
Parameters

​
Frequently Asked Questions

Kling 3 Omni is a multi-modal video generation API that creates videos from text prompts, images, or a combination. It supports reference images for style guidance and multi-shot mode for scene-by-scene control. Video durations range from 3-15 seconds.
Call the 4K endpoint instead of Pro or Standard. The resolution is fixed by the endpoint, so the request body is the same. The model renders 3840x2160 natively, so no upscaling step is involved. Generation takes longer and is priced differently from Pro and Standard.
Because the reference video had none. The Reference-to-Video endpoints never generate new audio: they carry over the audio track of the video you pass in video_url, which is kept by default and can be turned off with keep_audio: false. A silent reference produces a silent output, and generate_audio has no effect on these endpoints — use the Pro or Standard endpoints if you want the model to create audio.
For video-to-video generation (using a reference video), use the dedicated Reference-to-Video endpoints. These endpoints accept a video_url parameter and let you reference the video in your prompt as @Video1.
Multi-shot mode lets you create videos with multiple scenes, each with its own prompt. Provide an array of prompts via multi_prompt (max 6 shots). Each shot must be at least 3 seconds. The total duration is the sum of all shots.
Images must be publicly accessible URLs in JPG, JPEG, or PNG format. Requirements: minimum 300x300 pixels, maximum 10MB file size.
Rate limits vary by subscription tier. See Rate Limits for current limits and quotas.
Pricing varies based on model tier (Pro vs Standard) and video duration. See the Pricing page for current rates.

​
Best practices

  • Reference images: Use clear, well-lit reference images. Reference them as @Image1, @Image2 in your prompt for best results.
  • Duration planning: Start with 5-second videos to test, then increase duration for final output.
  • Multi-shot flow: Plan shot transitions carefully; each shot should have a coherent prompt.
  • Audio options: Use generate_audio: true for ambient sound, or voice_ids for narration.
  • Production integration: Use webhooks instead of polling for scalable applications.
  • Kling 3: Standard Kling 3 without Omni multi-modal features
  • Kling 2.6 Pro: Previous generation with motion control
  • Runway Gen 4.5: Alternative video generation model
  • VFX: Apply visual effects to generated videos