Kling 3.0 Omni Video Generator

This skill enables the generation and manipulation of videos using the Kling 3.0 Omni model. It provides a structured workflow for constructing API requests based on user intent, ensuring compliance with the model's complex parameter constraints.

Reference Files

This skill includes the following reference files:

• - references/api_reference.md — Complete official API parameter reference, including all fields, types, constraints, mutual exclusion rules (R1–R10), capability matrix, and invocation examples. Read this file before constructing any API call.
• INLINECODE1 — Kling 3.0 Omni prompt writing principles, official formula, template syntax, and few-shot examples for all major scenarios.
• INLINECODE2 — Python utility class for JWT authentication, task creation, and polling.

---

Core Capabilities

• - Text-to-Video: Generate a video from a textual description.
• Image-to-Video: Animate a static image with a descriptive prompt.
• Video-to-Video (Editing): Modify an existing video based on a prompt (e.g., change subject, style).
• Video-to-Video (Reference): Use an existing video as a reference for camera movement and style.
• Multi-shot Generation: Create a video with multiple distinct scenes or shots.
• Audio Generation: Generate video with synchronized audio, including speech and sound effects.

---

Workflow: From User Intent to API Call

To correctly use the Kling API, you MUST follow this decision-making workflow to construct the API payload. The process is divided into two main stages: Prompt Design and Parameter Construction.

Stage 1: Prompt Design

Before constructing the API call, you must first design the prompt(s) based on the user's request. The quality of the prompt is the single most important factor for a good result.

1. 1. Consult the Prompting Guide: Read /home/ubuntu/skills/kling-video-generator/references/prompt_guide.md to understand the core principles, official formula, and few-shot examples for writing effective prompts.

1. 2. Identify the Scenario: Determine which of the following scenarios the user is requesting:

- Single-shot video (from text, image, or video) - Multi-shot video (storyboard with multiple scenes)

1. 3. Write the Prompt(s):

- For single-shot, write a single, detailed prompt following the guide's formula. - For multi-shot, write a separate prompt for each shot/scene. - Use Template Syntax: If the user provides reference images, elements, or videos, you MUST use the <<<image_1>>>, <<<element_1>>>, <<<video_1>>> template syntax in the prompt to explicitly reference them. This is a core feature of the Omni model.

Stage 2: Parameter Construction

Once the prompt(s) are ready, construct the final API request payload by following this decision tree. This ensures all parameter constraints and interdependencies, discovered through extensive testing, are respected.

CODEBLOCK0

Key Parameter Rules (from testing)

This is not an exhaustive list, but a summary of the most critical, non-obvious rules that you MUST follow. For a complete guide, refer to the prompt_guide.md.

Parameter	Rule
INLINECODE8	MUST be explicit. Do not omit. Defaults to base but this is unreliable. Use base for editing, feature for reference.
INLINECODE12

Ignored in base mode. In customize mode, it MUST equal the sum of multi_prompt durations. |
| sound | Incompatible with video_list. Cannot be on if a reference video is provided. |
| shot_type | MUST be customize for multi_shot: true with the Omni model. intelligence is not supported. |
| multi_prompt | index MUST start from 1. Total duration MUST match top-level duration. Max 6 shots. |
| aspect_ratio | Required for feature mode. |
| image_list | Max 7 images without video input, max 4 images with video input. |

---

Execution

To execute a video generation task, use the provided Python script which handles authentication and polling.

1. 1. Set Environment Variables: Ensure KLING_ACCESS_KEY and KLING_SECRET_KEY are set.

1. 2. Construct the Payload: Follow the workflow above to create the JSON payload for the API call.

1. 3. Run the Script:

CODEBLOCK1

This structured approach ensures that all the nuances and constraints of the Kling 3.0 Omni API are handled correctly, leading to fewer errors and more predictable results.