vidau-video-generator（vidau 视频生成器）

Vidau Video Generator

Version: 1.0.8

When to use

• User asks to "generate a video", "create a short video with Veo3/Sora", "generate video from this prompt/image", "make a clip from this script", etc.
• User asks "how many credits do I have", "check my Vidau balance", "query Vidau credits", etc.
• User asks to "check my video task status", "has my Vidau task finished", "query task by UUID", etc.

When NOT to use

• User only wants to edit or process existing local video files (e.g. trim, merge, transcode) without calling any cloud API.
• User asks for a different video API (e.g. Runway, Pika) by name; use that provider’s skill if available.
• User has not mentioned Vidau/Veo/Sora/Seedance or "generate video" / "create video" and is only discussing concepts or other tools.

Reply language

When returning results (video link, credits, or error message) to the user, use the same language as the user’s question (e.g. Simplified Chinese if they asked in Chinese, English if they asked in English).

Output format (task success)

When a video task succeeds, return:

• Video URL (verbatim, from [VIDAU_VIDEO_URL] … [/VIDAU_VIDEO_URL])
• Thumbnail URL (verbatim, from [VIDAU_THUMB_PATH] … [/VIDAU_THUMB_PATH])

Always present both URLs in a code block so they are preserved verbatim and less likely to be altered by downstream processing. Example:

Video URL: <paste exact URL from script output>
Thumbnail URL: <paste exact URL from script output>

Do not rewrite, truncate, or wrap the URLs in other markup; copy them exactly from the script output.

Prerequisites

Environment variable VIDAU_API_KEY must be set (injected by OpenClaw from skills.entries.vidau.apiKey or env). If the user has no API key, they must register at https://www.superaiglobal.com/ to get one. Do not trigger this skill if it is not configured.

Environment check

Before running any script, ensure Python 3 is available:

1. Run python3 --version or python --version. If either succeeds, use that command when invoking the scripts below.
2. If both fail, try to install Python for the current platform:

- macOS: brew install python3 (requires Homebrew). - Linux (Debian/Ubuntu): sudo apt-get update && sudo apt-get install -y python3. - Windows: winget install Python.Python.3.12 if available; otherwise tell the user to download and run the installer from python.org.

1. If install fails (e.g. no permission or unsupported OS), reply with a short message that Python 3 is required and link to python.org/downloads, then stop. Do not run the Vidau scripts until Python is available.

The scripts use only the standard library; no pip packages are required.

Available scripts

• scripts/query_credit.py — Query current available credits; no args; outputs JSON with data.availableCredit, data.userId.
• scripts/query_models.py — Query model list and capability config; no args (optional --no-cache to force refresh); outputs JSON with data array. Each item has model, scaleList, durationList, resolutionList, generateAudio. Cached 10 minutes (env VIDAU_MODELS_CACHE or ~/.vidau_models_cache.json). Use before create_task to validate user parameters. After new models or capability changes, use --no-cache to force refresh so the agent does not validate against a stale list.
• scripts/upload_asset.py — Upload a local file (image or video) to Vidau; outputs JSON with data.url and data.assetId. Use data.url as --image-url / --last-image-url / --ref-image-urls in create_task when the user only has a local file. Caches by file content hash (default ~/.vidau_asset_cache.json, override with env VIDAU_ASSET_CACHE) so the same file is not re-uploaded.
• scripts/create_task.py — Create video task; required --prompt, optional --model, etc.; outputs JSON with data.taskUUID.
• scripts/wait_task.py — Poll until task succeeds or fails (or timeout); required --task-uuid; optional --interval (default 10s), --timeout (default 600s = 10 min). On success outputs same JSON as query_task and the same [VIDAU_VIDEO_URL] / [VIDAU_THUMB_PATH] blocks. Exit code 0 = succeeded, 1 = failed, 2 = timeout. Always use wait_task by default; when using wait_task you do not need to handle DELETE_SCHEDULED_TASK.
• scripts/query_task.py — Query task status and result once; required --task-uuid; outputs JSON with taskStatus, and on success also prints [VIDAU_VIDEO_URL] / [/VIDAU_VIDEO_URL] and [VIDAU_THUMB_PATH] / [/VIDAU_THUMB_PATH] blocks. Use query_task only when the platform requires scheduled polling. When the task reaches succeeded or failed, the script prints [OPENCLAW] DELETE_SCHEDULED_TASK task_uuid=<uuid> to stderr so the platform can remove the scheduled polling task.

Task polling (wait_task vs query_task): By default always use wait_task; use query_task only when the platform requires scheduled task polling. When using wait_task, there is no need to handle DELETE_SCHEDULED_TASK.

Workflow (generate video)

1. Optional: Call scripts/query_credit.py to check availableCredit before creating a task to avoid failures due to insufficient credits.
2. If the user only has a local image/video (no URL yet): Call scripts/upload_asset.py <path> to upload the file; read data.url from stdout and use it as --image-url, --last-image-url, or in --ref-image-urls when calling create_task.
3. Validate parameters against model capabilities: Call scripts/query_models.py to get the model list (cached 10 min). Find the entry in data where model equals the chosen --model. Validate user inputs (resolution, ratio, duration, generate_audio) against that entry:

- resolution: must be in the model’s resolutionList. If not, do not create the task; tell the user the resolution is not supported and list the allowed values (e.g. resolutionList). - ratio (if user specified one): must be in the model’s scaleList. If not, do not create the task; tell the user the ratio is not supported and list the allowed values (e.g. scaleList). - duration (if user specified one): for the chosen resolution, allowed durations are in durationList[resolution] (e.g. durationList["1080p"] = [5, 10]). If the user’s duration is not in that list, do not create the task; tell the user the duration is not supported for this resolution and list the allowed durations (e.g. for 1080p: 5, 10 seconds). - generate_audio (if user requested true): the model entry’s generateAudio must be true. If it is false, do not create the task; tell the user this model does not support audio generation. If any check fails, reply with what does not meet requirements and the allowed values; do not change the user’s parameters automatically and do not call create_task. If the chosen model is not in the list, skip validation and proceed to create_task.

1. Call scripts/create_task.py with --prompt (required), --resolution (required), --model (default veo@3:normal), and optional args (see parameter table below).
2. Read data.taskUUID from the script’s stdout JSON.
3. Always use wait_task by default: run scripts/wait_task.py --task-uuid <taskUUID> to poll until done (default 10s interval, 10 min timeout). When using wait_task you do not need to handle DELETE_SCHEDULED_TASK. Only when the platform requires scheduled polling use scripts/query_task.py --task-uuid <taskUUID> and, when the task reaches a terminal state, remove the scheduled task using the [OPENCLAW] DELETE_SCHEDULED_TASK task_uuid=<uuid> line on stderr.
4. Success: Return the video and thumbnail links to the user per Output format (task success) above: copy URLs verbatim from the script output blocks and present them in a code block. Failure: Explain using the API message or error code (e.g. insufficient credits, policy violation).

Workflow (query credits)

When the user asks for balance, call scripts/query_credit.py (no args), read data.availableCredit from stdout JSON and reply; optionally include data.userId.

Parameters (create_task.py)

Defaults when unspecified: 720p / 8s / 16:9 (see references/parameters.md). Note: create_task currently requires passing --resolution explicitly; the agent must always include this parameter for consistent behavior.

Valid values: resolution and ratio are constrained by the model’s capability (see workflow step 3 and references/capabilities.md). See references/parameters.md for more detail. Run python scripts/create_task.py --help for full parameters.

Model list

Common model IDs (aligned with API):

• Veo 3: veo@3:normal
• Veo 3.1: veo@3.1:normal, veo@3.1:fast
• Sora 2: sora@2:normal, sora@2:pro
• Seedance: seedance@1:pro, seedance@1:pro_fast
• Wan 2.5: wan@2.5:preview
• Vidu Q2: vidu@q2:turbo, vidu@q2:pro

See references/models.md for more models and use cases.

Error handling

• If VIDAU_API_KEY is not set, scripts exit with an error; tell the user to register at https://www.superaiglobal.com/ to get an API key, then configure it in OpenClaw.
• When API or script returns an error, relay the API message first, then use the one-sentence user message below. Reply in the user’s language.

See references/errors.md for details.

Example usage

• Generate video: User says e.g. "Generate an 8s video with Veo3 from this script" or "Generate a 10s sunset beach clip with Sora2". Agent: (optional) query_credit.py → query_models.py → validate resolution/ratio/duration/audio against model capabilities (step 3); if invalid, tell user and stop; else create_task.py --prompt "..." --resolution 720p --duration 8 --model veo@3:normal → read taskUUID → wait_task.py --task-uuid <uuid> → return video URL from output blocks or explain failure.
• Local image as first frame: User has only a local image file. Agent: upload_asset.py /path/to/image.png → read data.url from stdout → create_task.py --prompt "..." --resolution 720p --image-url "<data.url>" → then wait_task as above.
• Query credits: User says e.g. "How many Vidau credits do I have?" or "Check my Vidau balance". Agent: query_credit.py → reply with data.availableCredit (and optionally data.userId).
• Check task: User says e.g. "Is my video task done? UUID is xxx". Agent: use wait_task.py --task-uuid xxx by default → return status or video link (no need to handle DELETE_SCHEDULED_TASK). Use query_task.py only when the platform requires scheduled polling, and remove the scheduled task on terminal state via DELETE_SCHEDULED_TASK on stderr.