video-generator
GitHub用于将文本需求转化为专业的AI视频生成提示词,指导模型生成包含主体、动作、运镜和光影的短视频片段。
Trigger Scenarios
Install
npx skills add holaboss-ai/holaOS --skill video-generator -g -y
SKILL.md
Frontmatter
{
"name": "video-generator",
"description": "Create short videos and motion clips from a text description using AI video generation."
}
Video Generator
Act as the director who briefs the video model. A video prompt has to carry everything a still image does plus time — what moves, how the camera behaves, and what changes between the first frame and the last. Your job is to turn a one-line request into a shot that reads as deliberate rather than as a wobbling still.
When to use this skill
Use Video Generator for short social clips, product motion, b-roll, animated backgrounds, and any request that names a video, clip, animation, or motion. Reach for it whenever the deliverable moves.
Anatomy of a strong video prompt
Build the shot deliberately rather than describing a picture and hoping it animates. Cover:
- Subject — who or what is in frame, concretely, including how it is framed at the start.
- Motion — what the subject actually does, as one continuous action. One clear action beats three vague ones.
- Camera — static, pan, tilt, dolly in/out, tracking, orbit. Name it; "cinematic" is not a camera move.
- Setting & lighting — where this happens and how it is lit, including time of day.
- Style & mood — live-action / 3D / anime / stop-motion, plus the feeling and pacing.
- Duration fit — what can plausibly complete in the clip's length. A four-second shot holds one beat, not a sequence.
Be specific where it matters and silent where it doesn't. Contradictory motion (a static camera that also orbits) is the most common cause of a smeared result.
Working within the clip's length
- Short clips (4–6s): one subject, one action, one camera move.
- Longer clips (8–12s): a single action that develops — an approach, a reveal, a turn — not a cut between scenes.
- Never ask for cuts, scene changes, or on-screen text; generate separate clips and assemble them instead.
How to work
- Clarify the subject, the motion, the camera, and the target platform's aspect ratio and length.
- Compose a prompt covering the anatomy above, in one paragraph.
- Generate, then judge the result against the brief before offering it — a clip that ignored the camera move is a prompt problem, not a taste problem.
- Offer variations that change a meaningful axis (camera, pacing, or lighting), not trivial reworks.
Output format
Return the generated clip when generation is available; otherwise return the finished, ready-to-use prompt. For each, note the aspect ratio and duration it targets and one line on the directorial choice. When delivering variations, label each by the axis it explores.
Version History
- f6013b5 Current 2026-08-19 23:17


