> ## Documentation Index > Fetch the complete documentation index at: https://docs.imini.ai/llms.txt > Use this file to discover all available pages before exploring further. # Kling v3 Omni > Kling 3.0-Omni video generation model. Adds reference-video support (video source edit / video feature reference) on top of Kling v3. ## OpenAPI ````yaml /en/openapi/videos/kling-v3-omni.yaml POST /v1/videos/generate openapi: 3.1.0 info: title: Kling v3 Omni version: '1.0' description: >- Kling 3.0-Omni video generation model. Adds reference-video support (video source edit / video feature reference) on top of Kling v3. servers: - url: https://openapi.imini.ai/imini/router security: - ApiKeyAuth: [] paths: /v1/videos/generate: post: summary: Kling v3 Omni description: > Submit an asynchronous video generation task. You will need to poll the query endpoint to retrieve the result. > Model ID: `kling/kling-v3-omni` · Underlying: Kling 3.0-Omni Compared to `kling/kling-v3`: - Additionally supports `reference_videos` (video source edit / video feature reference), up to **1** clip; - `extra_params` does not support `camera_control`. operationId: generateKlingV3Omni requestBody: required: true content: application/json: schema: type: object required: - model properties: model: type: string description: Fixed value `kling/kling-v3-omni`. prompt: type: string description: > The prompt, which may contain both positive and negative descriptions. Up to **2500** characters. The Omni model supports referencing subjects, images, and videos in the prompt using the `<<<>>>` syntax, e.g. `<<>>`, `<<>>`, `<<>>`. > **Required when**: > - `extra_params.multi_shot=false`, or `extra_params.multi_shot=true` with `extra_params.shot_type=intelligence`. > > When `extra_params.multi_shot=true` and `extra_params.shot_type=customize`, this field is ignored — per-shot prompts are provided via `extra_params.multi_prompt`. example: >- A slow pan across a misty mountain valley at sunrise, cinematic resolution: type: string description: Output resolution. Defaults to `720P`. default: 720P enum: - 720P - 1080P aspect_ratio: type: string description: Output aspect ratio. Defaults to `16:9`. default: '16:9' enum: - '16:9' - '9:16' - '1:1' duration: type: integer description: > Video duration in seconds. Defaults to `5`. Allowed values: `3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15`. > When using video editing (`reference_videos[].refer_type=base`), this parameter is ignored — the output duration matches the input video and is billed based on the input video duration, rounded to the nearest second. default: 5 enum: - 3 - 4 - 5 - 6 - 7 - 8 - 9 - 10 - 11 - 12 - 13 - 14 - 15 generate_audio: type: boolean description: Whether to generate audio. Defaults to `false`. default: false reference_images: type: array description: > Reference image list. Can provide subject / scene / style references, or serve as the first or last frame of the video. **Input modes** (determined by the `reference_type` combination): - **Image-to-video**: 1 image with `first_frame`; - **First/last-frame interpolation**: 1 `first_frame` + 1 `last_frame` (last-frame-only is not supported — when `last_frame` is provided, `first_frame` must also be provided); - **Multi-reference image-to-video**: N images with `asset` (non-first/last-frame assets may omit `reference_type`, which defaults to `asset`). **Combining with `reference_videos`**: - When `reference_videos[].refer_type=base` (video to be edited), `first_frame` / `last_frame` are not allowed — only `asset` is supported; - Otherwise, `asset`-type images can coexist with `reference_videos`. **Image requirements**: - Formats: `jpg` / `jpeg` / `png`; - Size ≤ **10 MB**; - Dimensions ≥ **300px**, aspect ratio between `1:2.5` and `2.5:1`. **Count limits** (total of "reference images + multi-image subjects"): - No reference video, multi-image subjects only: **≤ 7**; - No reference video, with a video subject: **≤ 4**; - With a reference video, multi-image subjects only: **≤ 4**. > Multi-image subjects and video subjects are referenced from `prompt` via placeholders like `<<>>` / `<<>>`. maxItems: 7 items: type: object required: - url properties: url: type: string description: > Image address, in one of two forms: **1)** A publicly accessible URL (`http://` or `https://`); **2)** A Data URI base64 encoding (starting with `data:image/`). reference_type: type: string description: | Reference type: - `asset` — generic asset reference (default) - `first_frame` — first frame - `last_frame` — last frame default: asset enum: - asset - first_frame - last_frame reference_videos: type: array description: > Reference video list. At most **1** clip. **Two modes** (distinguished by `refer_type`): - `base` (default) — **video to be edited**: the source video is repainted/edited directly. In this mode, `first_frame` / `last_frame` cannot be supplied via `reference_images`; - `feature` — **feature reference video**: only motion / style features are extracted to guide generation. Use `keep_original_sound` to decide whether to preserve the original audio track — this also takes effect for `feature`. > When a reference video is provided, `generate_audio` is forced to `false` (no audio will be generated). **Video requirements**: - Formats: `MP4` / `MOV`; - Size ≤ **200 MB**; - Duration ≥ **3s** (upper bound depends on the model version); - Dimensions: **720px – 2160px** inclusive; - Frame rate: `24 – 60 fps` (output is always 24 fps); - Publicly accessible URL only; base64 is not supported. maxItems: 1 items: type: object required: - url properties: url: type: string description: >- Publicly accessible video URL. base64 is not supported. refer_type: type: string description: > Reference type: - `base` — video to be edited (default) - `feature` — feature reference video (extracts motion/style only) default: base enum: - base - feature keep_original_sound: type: boolean description: >- Whether to preserve the original audio track. Leave unset to follow the model default. extra_params: type: object description: | Kling's official extension parameters. All optional. properties: multi_shot: type: boolean default: false description: > Whether to enable multi-shot mode. - `true`: multi-shot enabled. `prompt` is ignored and first/last-frame generation is not supported; - `false`: multi-shot disabled. `shot_type` and `multi_prompt` are ignored. shot_type: type: string description: > Shot type. Only effective when `multi_shot=true`. - `intelligence` — model splits shots automatically from `prompt`; - `customize` — each shot is explicitly defined via `multi_prompt`. enum: - intelligence - customize multi_prompt: type: array description: > Per-shot definitions. **Required when `multi_shot=true` and `shot_type=customize`.** - Minimum **1**, maximum **6** shots; - Each shot prompt is up to **512** characters; - Each shot duration must be ≥ 1s and ≤ the total `duration`; - The sum of all shot durations must equal the total `duration`. minItems: 1 maxItems: 6 items: type: object required: - index - prompt - duration properties: index: type: integer description: Shot index, starting from `1`. prompt: type: string description: Prompt for this shot. Up to 512 characters. maxLength: 512 duration: type: string description: >- Shot duration in seconds, passed as a string (e.g. `"5"`). Must be ≥ 1 and ≤ the total `duration`. example: '5' additionalProperties: true examples: Text to Video: summary: Text to Video value: model: kling/kling-v3-omni prompt: >- A slow pan across a misty mountain valley at sunrise, cinematic resolution: 720P aspect_ratio: '16:9' duration: 5 Image to Video: summary: Image to Video value: model: kling/kling-v3-omni prompt: The mountain comes alive as birds fly past reference_images: - url: https://example.com/first.jpg reference_type: first_frame resolution: 1080P aspect_ratio: '16:9' duration: 5 Video Edit (base): summary: Video Edit (base) value: model: kling/kling-v3-omni prompt: Re-imagine the scene in neon cyberpunk style reference_videos: - url: https://example.com/source.mp4 refer_type: base keep_original_sound: false resolution: 1080P aspect_ratio: '16:9' duration: 5 Video Feature Reference: summary: Video Feature Reference value: model: kling/kling-v3-omni prompt: A tiger walking across a snowy plain reference_videos: - url: https://example.com/reference-motion.mp4 refer_type: feature resolution: 720P aspect_ratio: '16:9' duration: 5 responses: '200': description: Submitted successfully (asynchronous mode) content: application/json: schema: type: object properties: task_id: type: string description: Unique task ID, used for subsequent polling example: task_2042864638838083584 model: type: string description: The model ID used for this task example: kling/kling-v3-omni created_at: type: string description: Task creation time (ISO 8601) example: '2026-04-13T07:17:39.146Z' request_id: type: string description: Unique request ID example: 291a4117-9df8-4349-aee1-bdf8ffcacb9a '400': description: Validation Error (e.g. invalid parameter, invalid prompt) content: application/json: schema: type: object properties: error: type: object properties: code: type: string description: Error code example: INVALID_PARAMETER message: type: string description: Error message details example: reference_videos only supports at most 1 entry status: type: integer description: HTTP status code example: 400 request_id: type: string description: Unique request identifier example: req_abc123 security: - ApiKeyAuth: [] components: securitySchemes: ApiKeyAuth: type: http scheme: bearer bearerFormat: API Key description: > Pass `Authorization: Bearer ` in the request header. Go to the [API Keys management page](https://imini.ai/api-keys) to create and manage API Keys. ````