Agent Skills
› KunAgent/Kun
› media-video
media-video
GitHub根据文本或授权首帧图像生成短视频。支持定义主体、动作、风格等参数，处理图文转视频流程，验证元数据并确保运动连贯性，最终交付视频文件。
Trigger Scenarios
用户请求生成短视频
基于现有图像制作动态视频
Install
npx skills add KunAgent/Kun --skill media-video -g -y
SKILL.md
Frontmatter
{
"name": "media-video",
"description": "Generate a short Kun video from text or an authorized first-frame image."
}
Video Generation
Copyright (c) 2026 KunAgent. Licensed under the MIT License.
Purpose
Generate a short Kun video from text or an authorized first-frame image.
Tool routing
| Tool or skill | Use |
|---|---|
generate_video |
Generate a 6- or 10-second video. |
generate_image |
Create a first frame when the workflow needs one and the user permits it. |
send_im_attachment |
Send the generated video through IM when requested. |
Workflow
- Define subject, action, camera movement, scene continuity, style, aspect ratio, duration, and resolution.
- Use an authorized workspace image as first_frame_image_path when doing image-to-video.
- Write a motion-focused prompt that preserves important identity and geometry.
- Generate the video.
- Inspect the returned file metadata and report the path.
Completion gates
- Use only supported duration, resolution, and aspect-ratio values.
- Keep motion physically coherent for the short duration.
- Check that the first frame is a PNG, JPEG, or WebP workspace file.
Boundaries
- Do not use temporary attachment files as paths without authorization.
- Do not impersonate or fabricate real events involving identifiable people.
Delivery
Lead with the outcome, name the evidence used for verification, and disclose any real limitation that remains.
Version History
- a984912 Current 2026-09-09 01:50
