Kling V3 Omni Video Generation

Problem

When prompts, reference images, audio, and other assets are scattered across local files or pipelines, using Kling v3 Omni directly requires handling authentication, parameter mapping, file upload, async polling, and error interpretation. This skill turns those steps into a single agent-facing generation task, making it easier to validate multi-modal video generation workflows and reducing ad hoc parsing of CLI output, async status codes, and media URLs.

How It Works

The skill is a thin client over the dLazy hosted API. The caller supplies a prompt and media parameters; the CLI sends the request to api.dlazy.com for inference, while local image, video, or audio paths are uploaded to files.dlazy.com so the model can read them. Results are usually hosted URLs, and asynchronous tasks can be tracked by task ID and polled until complete. For agents, the key is keeping inputs traceable: prompts, media paths, wait mode, and output URLs should remain correlatable in logs.

Boundaries

It requires a valid dLazy API key, and failures can be caused by missing balance, invalid permissions, or safety-policy rejections. It does not perform local rendering, post-production editing, or self-hosted deployment; sensitive inputs should be checked against the third-party API's data-handling terms. For batch generation, limit concurrency, retain task IDs, and avoid duplicate submissions that could consume credits.