Volcano Engine Seedance Video Generation

Problem

When a user provides only a text prompt or a local image, but the video service expects a public image URL, an async task workflow, and result download, manual integration is error-prone: missing token replacement, passing local file paths directly, over-polling, or creating duplicate tasks can all cause failures and wasted billing. Common failures also include auth errors, inaccessible images, and task timeouts.

How It Works

This skill organizes Volcano Engine Seedance calls through an AI aggregation platform into practical steps:
- Text-to-video: call POST /v1/images/generations with prompt, size, and duration.
- Image-to-video: first upload JPG/PNG/WebP/GIF/BMP via POST /api/upload/image, then call POST /v1/chat/completions using a multimodal content array.
- Task polling: query GET /v1/aigc/task/{task_id} for PROCESSING, SUCCESS, or FAILED.
- Output: after success, provide the playback URL and local file path; failures rely on service-side automatic refunds.

Limits

It fits agent workflows for 4-15 second clips, not real-time editing, long-form post-production, or batch creative rewriting. Keep the user's original prompt unchanged; poll about every 8 seconds, stop after 10 minutes, and use the task ID for later lookup instead of recreating tasks.