WeryAI Video Generator

Generate WeryAI videos with the official base skill for text-to-video, image-to-video, video from image, storyboard-to-video, and first-frame/last-frame transition workflows. In agent environments, default to result-first execution: submit and poll until playable videos are ready or timeout is reached. Use bounded wait (short tasks: 10 minutes, long tasks: 30 minutes) and avoid unbounded polling loops.

Example Prompts

• - INLINECODE0
• INLINECODE1
• INLINECODE2
• INLINECODE3
• INLINECODE4

Quick Summary

• - Main jobs: text-to-video, image-to-video, video from image, first-frame to last-frame video, storyboard-to-video, INLINECODE10
• Default model: Seedance 2.0 (SEEDANCE_2_0)
• Default parameters: duration=5, resolution=720p, aspect_ratio=9:16, generate_audio=true (for audio-capable models)
• Main trust signals: dry-run support, model capability lookup, paid-run warning, media-source validation with auto upload for local references

Authentication and first-time setup

Before the first real generation run:

1. 1. Create a WeryAI account.
2. Open the API key page at https://www.weryai.com/api/keys.
3. Create a new API key and copy the secret value.
4. Add it to the required environment variable WERYAI_API_KEY.
5. Make sure the WeryAI account has available balance or credits before paid generation.

OpenClaw-friendly setup

• - This skill already declares WERYAI_API_KEY in metadata.openclaw.requires.env and primaryEnv.
• After installation, if the installer or runtime asks for required environment variables, paste the key into WERYAI_API_KEY.
• If you are configuring the runtime manually, export it before running commands:

CODEBLOCK0

Quick verification

Use one safe check before the first paid run:

CODEBLOCK1

• - models-video.js confirms that the key is configured and the models endpoint is reachable.
• INLINECODE23 confirms the request shape locally without spending credits.
• Real wait or submit-* commands still require available WeryAI balance.

Prerequisites

• - WERYAI_API_KEY must be set before paid runs.
• Node.js >=18 is required because the runtime uses built-in fetch.
• Reference media (image, images, videos, audios) can be http/https URLs or local/file sources. Local/non-http(s) sources are uploaded first via /v1/generation/upload-file.
• Real submit and wait commands consume WeryAI credits.

Security And API Hosts

• - Keep WERYAI_API_KEY secret and never write it into the repository.
• Optional overrides WERYAI_BASE_URL and WERYAI_MODELS_BASE_URL default to https://api.weryai.com and https://api-growth-agent.weryai.com. Only override them with trusted hosts.
• Review scripts/ before production use if you need higher assurance.

Supported Intents

• - Text brief -> generate a video from scratch.
• One reference image -> animate the source image into a video.
• Start frame + end frame -> generate a controlled transition video.
• Multiple ordered images -> turn a storyboard or shot sequence into one video.
• Existing task -> check status instead of creating a new paid job.
• Parameter or model questions -> inspect models first, then submit.

Default Configuration

Unless the user explicitly changes them, prefer:

• - model: SEEDANCE_2_0 (SEEDANCE_2_0)
• INLINECODE47: INLINECODE48
• INLINECODE49: INLINECODE50
• INLINECODE51: INLINECODE52
• INLINECODE53: true (for audio-capable models)

Always allow the user to override model, duration, resolution, aspect_ratio, and generate_audio. When the user asks for unsupported settings, check models-video.js and keep only values supported by the chosen model.

Model Switching And Parameter Guidance

Guide the user progressively instead of explaining every parameter up front.

• - If the user only wants a video, proceed with the default Seedance 2.0 (SEEDANCE_2_0) configuration.
• If the user asks for a different model, better quality, stronger motion, longer duration, landscape output, vertical output, or generated audio, switch into parameter-confirmation mode.
• If the user already knows the exact parameter they want, apply it directly and only validate model support when needed.
• If the user sounds unsure, translate their creative request into the closest supported parameters rather than asking them to choose raw API fields.

Recommended guidance pattern

Use short operator-style guidance like this:

• - General help: When the user asks "how to use this skill", DO NOT paste raw shell commands. Instead, explain the capabilities in natural language and give 2-3 prompt examples (e.g., "You can ask me to animate a portrait image...").
• Default run:

• - Model switching:

• - Parameter changes:

• - Safety before paid runs:

When to ask follow-up questions

Ask only for the smallest missing detail needed to submit safely.

• - Ask about aspect_ratio when the user implies platform intent such as TikTok, Reels, YouTube Shorts, or landscape trailer.
• Ask about duration when the user asks for a longer clip or a slower beat.
• Keep generate_audio enabled by default for audio-capable models unless the user asks to mute.
• Ask about model choice only when the user explicitly wants a different model or when capability support is uncertain.
• Do not ask every parameter question if the default configuration already fits the request.

Map user language to parameters

Use these common mappings:

• - vertical, portrait, short video, TikTok, Reels, Shorts -> INLINECODE75
• INLINECODE76 -> INLINECODE77
• INLINECODE78, widescreen, YouTube, cinematic frame -> INLINECODE82
• INLINECODE83, slower pacing -> increase duration to a supported value such as 10 or INLINECODE87
• INLINECODE88, higher quality -> use the highest supported resolution for the chosen model
• INLINECODE91, with audio -> INLINECODE93
• INLINECODE94, not Seedance, check supported models -> run models-video.js before submission

Model lookup workflow

When the user asks to change the model or requests parameters that may be unsupported:

1. 1. Identify the intended mode: text_to_video, image_to_video, or multi_image_to_video.
2. Run the matching models-video.js command first.
3. Keep only supported values for duration, aspect_ratio, resolution, and audio support.
4. If multiple models fit, recommend one concise default instead of dumping raw metadata unless the user asked for a comparison.
5. If support is unclear, say so explicitly and use --dry-run or a safe model query before the paid call.

Confirmation block before submission

Before a paid run, show a concise confirmation block with the final payload choices.

CODEBLOCK2

Wait for confirmation or requested edits before running a paid submission.

Intent Routing

Use result-first execution as the default path so users receive final playable output whenever possible.

• - If the user provides only prompt, route to text-to-video.
• If the user provides image, route to image-to-video.
• If the user provides first_frame + last_frame, or image + last_image, normalize them into ordered images and route to the guided multi-image flow.
• If the user provides images, route to multi-image-to-video.
• If multi-image support is unavailable for the chosen model, the runtime may downgrade to image-to-video with the first image only.
• Default run path:

• - Use submit-* only when the user explicitly asks to create a task without waiting.

Preferred Commands

CODEBLOCK3

Workflow

1. 1. Identify the user's intent: text-only, single-image, first/last-frame, multi-image, status lookup, or model lookup.
2. Collect prompt and, if needed, ordered public https image URLs.
3. Apply defaults: Seedance 2.0 (SEEDANCE_2_0), 5s, 720p, 9:16, and generate_audio=true for audio-capable models, unless the user asks otherwise.
4. If the user wants a custom model or non-default parameters, run models-video.js first when support is uncertain.
5. Use --dry-run when you need to preview the final payload before a paid submission.
6. Default to bounded wait execution:

1. 7. Use submit-* only when the user explicitly asks for task creation without waiting.
2. Use status-video.js to re-check an existing task or batch safely.

Poll Timeout Classes

• - short task class: text_to_video, default timeout 10 minutes (600000ms).
• INLINECODE135 task class: image_to_video, multi_image_to_video, almighty_reference_to_video, default timeout 30 minutes (1800000ms).
• INLINECODE141 task class (default) maps from the effective submission mode.
• INLINECODE142 remains the highest-priority explicit override for compatibility.
• Optional environment overrides:

Input Rules

• - prompt and duration are required for every generation mode.
• INLINECODE147, images, first_frame, last_frame, and last_image must be public https URLs.
• If both images and image are provided, images wins during mode detection.
• INLINECODE156 + last_frame and image + last_image are accepted aliases for start/end-frame intent.
• Only send aspect_ratio, resolution, negative_prompt, or generate_audio when supported by the selected model.
• Do not invent undocumented fields.

Output

All commands print JSON to stdout. Successful results can include:

• - taskId, taskIds, INLINECODE166
• INLINECODE167
• INLINECODE168
• INLINECODE169
• INLINECODE170
• INLINECODE171, INLINECODE172

User-facing delivery requirement:

• - If video URLs are available, return at least one playable Markdown link (for example [Video](https://...)). If multiple videos are generated, render all of them using markdown links consecutively.
• Alongside video output, include key generation parameters when available: model, duration, aspect_ratio, resolution, generate_audio.
• If timeout is reached, return the taskId to the user and ask if they want you to check the status again. Do NOT show the raw node status command to the user; use it internally.

See references/error-codes.md for common failure classes and recovery hints.

Definition Of Done

The task is done when:

• - local validation passes without CLI-side errors,
• INLINECODE180 reaches terminal state with at least one playable video URL,
• or bounded wait reaches timeout and returns the taskId plus an offer to check status again later,
• or status-video.js returns a clear in-progress or terminal state,
• and the output makes it explicit whether video URLs are present.

Constraints

• - Do not assume every model supports every parameter.
• Do not use local file paths for reference images.
• Do not re-run submit or wait casually because each run can create a new paid task.
• Do not default to wait-video.js in agent environments for long-running generations.
• Do not broaden this skill into general video editing outside the documented WeryAI API surface.

Re-run Behavior

• - submit-text-video.js, submit-image-video.js, and submit-multi-image-video.js are not idempotent.
• INLINECODE189 is also not idempotent because it submits first and then polls.
• INLINECODE190, models-video.js, and balance-video.js are safe to re-run.

References

• - Model metadata and parameter rules: references/api-models.md
• Error handling guidance: references/error-codes.md