Wasu Video Gen

Problem

AI video generation APIs often require callers to manage model selection, asset upload, task polling, credential refresh, and short-lived video download. This skill wraps Wasu's proxied Seedance interface into a conversational workflow for engineers generating short videos from a local terminal.

How It Works

• Credential management: reads local cache first; when the cache is missing or expired, it uses wasu-cli to log in and write the config, avoiding repeated key retrieval.
• Generation modes: supports one-prompt generation and an HTML config page. The prompt mode extracts prompt, model, resolution, duration, and ratio; the config mode enables local upload of images and videos.
• Model coverage: includes doubao-seedance-1.5-pro, doubao-seedance-2.0, and doubao-seedance-2.0-fast, with differences across 480p/720p/1080p/4K, 4-15 second durations, and first/last-frame, reference image, reference video, or reference audio capabilities.
• Result handling: downloads and opens the video after generation; the video URL expires after one hour, so saving it promptly matters.

Boundaries

Best for text-to-short-video, first/last-frame control, and style transfer with a small number of reference assets. It does not support audio-only input or text-plus-audio combinations; audio must be paired with images or video. Calls should run in the foreground while waiting, and parameter enums, content moderation, plan authorization, and asset size limits can all affect the result.