kling-v3-omni model. Combine a required prompt with first and last frames, reference images, or one reference video, then track the asynchronous result by img_uuid.Kling 3.0 Omni API for Multimodal Video Generation
Access Kling 3.0 Omni through GoEnhance with the
Get API Key
One endpoint for prompt, image, and video references
Address reference materials inside the prompt
Images are numbered @image_1 onward in first-frame, last-frame, then ref_imgs order. A reference video is @video_1, so stable array ordering keeps prompt references deterministic.
Use feature or base reference-video behavior
With video_type: feature, the output follows reference motion and style; audio generation is rejected and multi-shot is forced on. With base, the source video is edited in place, first/last frames and multi-shot are unavailable, and existing audio is retained.
Validate role-specific combinations before enqueueing a generation task.
Kling 3.0 Omni API capabilities and validation rules
Model and prompt
POST to <code>https://api.goenhance.ai/api/v1/videos/generations</code> with required <code>model: kling-v3-omni</code> and a required <code>prompt</code>.
First and last frames
Use <code>image_url</code> for a first frame. <code>image_end_url</code> requires that first frame; last-frame-only requests are not supported.
Reference images
Pass ordered <code>ref_imgs</code> values and refer to them with <code>@image_N</code>. The compatible alias <code>image_urls</code> is also accepted.
Reference video
Pass at most one 3–15.5 second MP4 or MOV through <code>ref_videos</code>. <code>video_url</code> is a compatible alias.
Output controls
Choose an integer duration from 3 to 15 seconds, 720p/1080p/4K resolution, and 16:9, 9:16, or 1:1 when no first frame or reference video determines the ratio.
Material limits
First frame, last frame, and reference images can total up to seven materials, or up to four when a reference video is included.
Pricing is per generated second and depends on resolution, reference-video use, and native audio.
Kling 3.0 Omni API pricing
720p · no video, no audio
4.2 tokens per generated second ($0.084/s).
720p · native audio
5.6 tokens per generated second ($0.112/s) when there is no reference video.
720p · with reference video
6.3 tokens per generated second ($0.126/s).
1080p tiers
5.6 tokens/s ($0.112/s) without video or audio, 7 tokens/s ($0.14/s) with native audio and no video, or 8.4 tokens/s ($0.168/s) with a reference video.
4K
21 tokens per generated second ($0.42/s) across the documented no-audio, native-audio, and reference-video scenarios.
Estimate each request
Multiply requested duration by the exact rate for the selected resolution, reference-video use, and audio mode. USD equivalents use the documented $0.02 per token.
Integrate Kling 3.0 Omni in six steps
1
1. Create an API key
Create a GoEnhance API key and send it in the
Authorization header. 2
2. Select the model
Set
model to kling-v3-omni and write a prompt that explicitly addresses any @image_N or @video_1 material. 3
3. Validate materials
Check frame dependencies, material counts, reference-video duration, and the restrictions for
feature or base mode. 4
4. Submit the task
POST the request to
/api/v1/videos/generations and persist the returned img_uuid. 5
5. Track completion
Poll
GET /api/v1/jobs/detail or provide a publicly accessible HTTPS custom_callback_url. 6
6. Handle terminal states
Process success and failure idempotently. Callback receivers that do not return HTTP 200 may be retried up to three times with a three-second timeout per attempt.
Build multimodal video workflows around explicit input roles and deterministic material references.
Developer use cases for Kling 3.0 Omni
Reference-led campaign clips
Combine approved keyframes and visual references with controlled prompt templates for repeatable campaign variants.
Product motion studies
Use a first frame or ordered reference images to explore product movement while your application manages asset validation and approvals.
Video restyling pipelines
Use a feature reference video when motion and style should guide a new output, subject to its audio and multi-shot rules.
Base-video edits
Use base mode when an existing video should be edited in place while retaining its original audio.
Storyboard transitions
Use first-and-last-frame inputs to plan a transition between approved visual beats without changing your job orchestration.
Localized creative automation
Parameterize prompts and material references for regional variants while reusing the same asynchronous task handler.
Practical answers for request validation, pricing, references, and asynchronous completion.