Seedance 2.0 Open-Source Video Generator

What It Solves

When a natural-language prompt needs to become a publishable video, developers and creators often have to handle API authentication, task polling, and result downloads manually. This skill wraps the Seedance 2.0 workflow around a simple entry point: after providing a prompt, it submits a generation task, tracks status, and saves the finished MP4 to a local directory, reducing the cost of wiring up the underlying integration.

It fits quick creative validation, product demos, short social-media clips, and turning abstract concepts into visual material.

Core Capabilities and Flow

• Text-to-video: prompt accepts Chinese or English descriptions. The underlying model is doubao-seedance-2-0-260128, and the default output is an MP4 with synchronized audio.
• Parameter control: you can set --resolution to 480p/720p/1080p, --ratio to 16:9/9:16/1:1, and --duration to 4-15 seconds or an intelligent duration.
• Task management: --no-download submits the task only, while --task-id queries an existing task. The script polls queued/running/succeeded/failed states and waits up to about 20 minutes.
• Reproducibility and continuation: --seed fixes the random seed, and --return-last-frame can return the final frame for continuous generation.
• People and assets: it supports asset:// references to preset virtual personas; the source notes that directly uploading reference images or videos containing real human faces is not supported.

Boundaries and Notes

You need a redfox.hk account and API Key, and the platform balance and network access should be available. Generation usually takes 3-15 minutes, with longer times for complex scenes; if a task times out, you can query it later by task ID. Output is saved locally by default, and expired download links or an unwritable output directory require re-querying the task or adjusting the destination.