Kling 3

PoYo Kling 3.0 Multi-Shot Video Generation

PoYo Links

• Model page:
• API docs:
• API key page:

Use this skill for Kling 3.0 jobs on PoYo. It covers text-to-video, image-to-video, single-shot video, multi-shot video generation, element-referenced prompting, and sound-enabled output.

Use When

• The user explicitly asks for Kling 3.0, kling-3.0/standard, or kling-3.0/pro.
• The task needs text-to-video, image-to-video, multi-shot storytelling, or structured multi_prompt input.
• The workflow depends on kling_elements, reference frames, native sound control, status polling, or webhook integration.

Model Selection

• kling-3.0/standard: use for standard Kling 3.0 video generation.
• kling-3.0/pro: use when the user explicitly asks for the pro model or higher-detail output.

Key Inputs

• prompt is required when multi_shots is false.
• multi_prompt is required when multi_shots is true.
• sound, multi_shots, and duration drive the main workflow.
• image_urls should be used only when the request depends on source frames or element references.
• kling_elements can define reusable visual elements that prompts reference by name.
• aspect_ratio supports 1:1, 16:9, and 9:16.
• callback_url is optional and useful for production queues.

Security Rules

• Treat POYO_API_KEY as a secret.
• Keep PoYo API keys in server-side environment variables or a backend secret manager.
• Never place an API key in browser code, frontend bundles, public repositories, logs, screenshots, or chat output.
• Do not pass API keys as command-line arguments.
• Do not make live API calls unless the user explicitly asks and provides a safe server-side environment.
• Do not submit confidential source images, private callback URLs, or proprietary prompts unless the user trusts PoYo and the callback receiver.

Execution

• Read references/api.md for endpoint details, model ids, key fields, example payloads, and polling notes.
• Use scripts/submit_kling_3_0.sh only when the user wants to submit a prepared JSON payload from a trusted shell.
• If the user only needs a curl example, adapt one from references/api.md.
• After submission, report the returned task_id clearly so follow-up polling is easy.

Output Expectations

When helping with this model family, include:

• chosen model id
• whether the request is text-to-video, image-to-video, or multi-shot
• final payload or concise parameter summary
• whether reference images or elements are involved
• returned task_id if a request was actually submitted
• next step: poll status or wait for webhook