Seedance Video Generation

Problem

Seedance Video Generation targets engineering workflows that need short AI videos from text prompts or reference images. It wraps the asynchronous Volcengine Ark API flow: model selection, task submission, polling, and artifact download, while handling local image base64 conversion, defaults, timeouts, and error feedback.

Core Capabilities

• Models: defaults to doubao-seedance-1-5-pro-251215; also supports Seedance 1.0 Pro / Lite variants. 1.5 Pro adds audio, draft mode, and a wider duration range.
• Modes: text-to-video, first-frame image-to-video, first-and-last-frame image-to-video, and Seedance 1.0 Lite I2V reference-image mode with 1–4 images referenced by [image 1] / [image 2] in the prompt.
• Workflow: create a task, poll status every roughly 15 seconds, then download immediately because URLs expire after 24 hours; keep the task ID for status checks, cancellation, or cleanup.
• Controls: configure ratio, duration, resolution, seed, camera_fixed, watermark, generate_audio, draft, return_last_frame, and service_tier; use return_last_frame to chain consecutive videos by feeding the last frame into the next first frame.
• Input constraints: images must meet format, aspect-ratio, dimension, and file-size limits; local files should be converted to base64 data URLs.

Boundaries

This skill requires ARK_API_KEY and Volcengine Ark access. It suits prototypes, previews, or consecutive-shot experiments in controlled environments. Generation is asynchronous; draft is a cheap preview forced to 480p, and final output still needs resolution, duration, audio, and compliance checks. History is retained for 7 days, and failed or queued tasks should be debugged by reviewing errors and adjusting prompts, images, or parameters.