Turn briefs and sources into controlled, source-grounded video prompts that run on HiAPI Seedance 2.0 or HappyHorse 1.0.
Video Prompt Generator • HiAPI Handoff • HiAPI
This skill itself does not need an API key.
HIAPI_API_KEYis only required for the target render skill (Seedance 2.0 or HappyHorse 1.0).
HiAPI Docs · All HiAPI Skills · Remote MCP · Pricing · Get API Key
HiAPI Matrix: 🎨 Image Prompts · 🎬 Video Prompts · 🛠️ Agent Skills (you are here) · 🤖 Remote MCP · 📖 API Docs
AI Agent? Skip the README and read llms-install.md. It contains installation steps and the agent contract for this skill.
A prompt-engineering skill for OpenClaw / Claude Code / OpenCode / Codex-style agents. After installation, your agent can take a one-line idea, a product link, or a research topic and return a directed video prompt ready to paste into one of HiAPI's video skills.
This skill produces prompts. It does not call a video API by itself. The output is built to plug into:
| Target | When to use |
|---|---|
| hiapi-seedance-2-0-video-skill | Text-to-video and image-to-video, cinematic quality, 4–15 seconds |
| hiapi-happyhorse-1-0-video-skill | Fast text-to-video drafts |
A directed prompt specifies what appears, what moves, what is typed, what is heard, and what the model must not show. Every key fact in the source appears on screen, in narration, or as a labeled visual cue.
Use this skill when the brief is too short, the source has not been read closely, or the user has produced a vague prompt and expects a generic video back. If the user already has a final, scene-by-scene prompt, skip this skill and call the target video skill directly.
For a library of proven Seedance 2.0 prompts to anchor on, see awesome-seedance-2-0-prompts.
npx -y github:HiAPIAI/hiapi-video-prompt-generator-skill -yThe installer auto-detects Codex (~/.codex/skills) and Claude Code (~/.claude/skills). To target a specific agent or directory:
npx -y github:HiAPIAI/hiapi-video-prompt-generator-skill --codex # ~/.codex/skills only
npx -y github:HiAPIAI/hiapi-video-prompt-generator-skill --claude # ~/.claude/skills only
npx -y github:HiAPIAI/hiapi-video-prompt-generator-skill --target=/path # custom directory
AGENT_SKILLS_DIR=/path npx -y github:HiAPIAI/hiapi-video-prompt-generator-skill -yopenclaw skills add https://github.com/HiAPIAI/hiapi-video-prompt-generator-skillgit clone https://github.com/HiAPIAI/hiapi-video-prompt-generator-skill.git
export AGENT_SKILLS_DIR="/path/to/your/agent/skills"
mkdir -p "$AGENT_SKILLS_DIR"
cp -R hiapi-video-prompt-generator "$AGENT_SKILLS_DIR/hiapi-video-prompt-generator"Install the HiAPI Video Prompt Generator skill:
1. Run: npx -y github:HiAPIAI/hiapi-video-prompt-generator-skill -y
(auto-detects Codex / Claude Code skill directories)
2. Read SKILL.md for the output contract.
3. After this skill returns a prompt, run it through one of:
- https://github.com/HiAPIAI/hiapi-seedance-2-0-video-skill
- https://github.com/HiAPIAI/hiapi-happyhorse-1-0-video-skill
Talk directly to your AI agent:
Use
$hiapi-video-prompt-generatorto turn https://github.com/HiAPIAI/hiapi-skills into a 5-second product intro.
Use the HiAPI Video Prompt Generator to plan a 9:16 social short about why agent skills beat one-shot prompts.
Use
$hiapi-video-prompt-generatorto plan an 8-second image-to-video fromoutputs/product.png, soft camera, studio lighting.
This skill returns a structured prompt with scene-by-scene direction and a ready-to-paste command for the target HiAPI skill.
Every skill run returns these sections, in order:
- Video Type — product demo, teaching short, social short, explainer, pitch, historical, or visual concept.
- Target Model —
hiapi-seedance-2-0-videoorhiapi-happyhorse-1-0-video, with one reason. - Duration And Aspect — picked from the chosen target's allowed list. Seedance: any integer from
4to15seconds +--ratio∈16:9|9:16|1:1|4:3|3:4|21:9|adaptive. HappyHorse: any integer from3to15s +--size∈16:9|9:16|1:1|4:3|3:4. The output names the right flag. - Core Objective — single sentence the viewer should remember.
- Source Extraction Summary — 5–10 bullets. Real facts are tagged
[source]; only staging choices (camera, layout, lighting, transition copy) may be tagged[creative assumption]. Guesses are never tagged as sources. - Narrative Through-Line — one-sentence arc.
- Scene Prompts — time, visual, on-screen text, action/camera, narration, transition.
- Required Screen Text — every exact quoted string.
- Motion And Camera Control — dominant and ambient motions.
- Style Requirements — color, lighting, typography, mood.
- Negative Constraints — what the model must not show.
- Final Copy-Ready Prompt — a compact block to paste into
--prompt; must keep scene order with time cues (e.g.[0–1.2s]), all Required Screen Text strings verbatim, dominant motions, and at least one Negative Constraint. - Handoff Command — one
cdline into the installed target skill directory, then the exactnode scripts/...line with all parameters filled (Seedance uses--ratio, HappyHorse uses--size).
See SKILL.md for the full agent contract and references/ for the patterns and source-extraction guidance.
- Duration: defaults to
5seconds. This skill never produces a30-second plan, because HiAPI's video models do not run that length. - Aspect:
16:9for demos and explainers,9:16for social shorts. - Image-to-video: only on Seedance 2.0. The starting image is passed as
--first-frame-url. HappyHorse 1.0 is text-to-video only. - Model-specific parameters — this skill constrains the handoff to the chosen target's actual API:
| Model | Durations (--seconds) |
Resolutions | Aspect flag |
|---|---|---|---|
| Seedance 2.0 | any integer from 4 to 15 |
480p, 720p, 1080p, 4k |
--ratio ∈ 16:9, 9:16, 1:1, 4:3, 3:4, 21:9, adaptive |
| HappyHorse 1.0 | any integer from 3 to 15 |
720p, 1080p |
--size ∈ 16:9, 9:16, 1:1, 4:3, 3:4 |
If the user asks for an unsupported duration or ratio, this skill offers the closest supported value and notes the change in the output.
This skill is inspired by the upstream video-prompt-director, which defaults to 30-second prompts for HyperFrames-style generators. HiAPI's video models output single clips up to 15 seconds. Defaulting to 5 keeps the prompt directly runnable against Seedance 2.0 or HappyHorse 1.0 without the user having to remember to compress the plan. Longer pieces are still supported up to 15 seconds where the target model allows it.
.
├── README.md
├── README.zh-CN.md
├── SKILL.md
├── LICENSE
├── package.json
├── agents/
│ └── openai.yaml
├── references/
│ ├── prompt-patterns.md
│ ├── source-extraction.md
│ └── hiapi-handoff.md
├── scripts/
│ └── install.mjs
└── llms-install.md
| Question | Answer |
|---|---|
| Does this skill call the HiAPI API directly? | No. It produces prompts. Pair it with a HiAPI video skill to render. |
Do I need HIAPI_API_KEY to use this skill? |
No. This skill runs offline. The key is only needed for the target render skill. |
| Why does it default to 5 seconds and not 30? | HiAPI's video models do not support 30-second outputs. Seedance supports integer durations from 4 to 15 seconds; HappyHorse supports 3/5/8/10/15. |
| Can it pick the target model for me? | Yes. If you do not name one, it defaults to Seedance 2.0 for image-to-video and cinematic work, HappyHorse 1.0 for fast text-only drafts. |
| What if my brief lacks facts? | This skill only labels staging choices (camera, layout, lighting) as [creative assumption]. Missing facts become a question for you, not an invented claim. |
| What if I provide a URL? | This skill extracts facts from the source first, then writes the prompt. See references/source-extraction.md. |
| Agent | Install Method |
|---|---|
| Codex | npx -y github:HiAPIAI/hiapi-video-prompt-generator-skill --codex |
| Claude Code | npx -y github:HiAPIAI/hiapi-video-prompt-generator-skill --claude |
| OpenClaw | openclaw skills add https://github.com/HiAPIAI/hiapi-video-prompt-generator-skill |
| OpenCode | AGENT_SKILLS_DIR=~/.opencode/skills npx -y github:HiAPIAI/hiapi-video-prompt-generator-skill -y |
| Cursor / other | npx -y github:HiAPIAI/hiapi-video-prompt-generator-skill --target=/your/skills/dir |
MIT
HiAPI — One API, all AI models