Future Vision Custom Video Generation

Problem

Calling Future Vision video generation directly can hide operational details: missing auth, mismatched display names and stable enums, backend size validation, and tasks that are created before the video is finished. The skill focuses on direct custom video generation and keeps the workflow aligned with a shared CLI and a stable request contract.

How It Works

The skill first routes away from template remake, batch creative exploration, product TVC, and image-only output. It then calls the backend through a stable contract:

• Required parameters: prompt; model uses standard, pro, or stable, with display names such as “Alibaba model” mapped to pro; seconds commonly uses 5, 10, or 15; and size is selected by model and validated by the backend.
• Optional inputs: resolution, referenceImage, referenceVideoUrls, referenceAudioUrls, and requestKey.
• Task state: it preserves reference media order, uses x-idempotency-key to avoid duplicate submissions from retries, and emits [SORA_TRACK] for queued, pending, or running states so downstream agents can keep polling.

Boundaries

A successful task creation does not mean the video is complete; work progress must be queried. When a required field is missing, ask for only one field at a time rather than asking for model, duration, and size together. If only a taskId is available, resolve it to a workId before polling. Authentication failures should be handled through the referenced docs, and the skill should not duplicate the full size/duration matrix.