← 返回 Skills 市场
全局配置文件:
132
总下载
0
收藏
1
当前安装
3
版本数
在 OpenClaw 中安装
/install doubao-video-creator
功能描述
基于火山引擎豆包视频生成模型,辅助用户配置项目并分场景确认提示词,逐步生成并合成专业短视频。
使用说明 (SKILL.md)
doubao-video-creator - 豆包视频创作助手 v3.0
🆕 v3.0 更新 (2026-04-28)
- ✅ 默认时长:从 5 秒改为 10 秒
- ✅ 用户可选时长:10 秒或 15 秒
- ✅ 用户可选模型版本:1.0 / 1.5 / 2.0
- ✅ 每个场景独立选择时长和模型版本
- ✅ 文生图模型升级:
wan2.6-t2i→wan2.7-image-pro
📋 技能描述
使用火山引擎豆包(Doubao Seedance)视频生成模型,帮助用户创作专业视频。支持项目化配置管理、分场景生成方式选择、完整状态追踪。
🎯 触发条件
用户提到以下关键词时触发:
- "豆包生成视频"
- "帮我做视频"
- "制作短视频"
- "视频创作"
- "生成宣传视频"
- "doubao 视频"
- "火山引擎视频"
🔄 工作流程(v3.0 增强版)
阶段 0:首次配置
您好!我是豆包视频创作助手 🎬
检测到您是首次使用,需要先配置豆包 API 信息:
1️⃣ 请提供您的火山引擎豆包 API Key
- 获取地址:https://console.volcengine.com/ark
配置将保存,后续使用无需重复输入。
阶段 1:需求收集
为了生成更符合您需求的视频,请告诉我:
**1. 视频主题**:是什么产品/服务?
**2. 目标受众**:视频给谁看的?
**3. 视频风格**:科技感?温馨?专业?电影感?
**4. 参考资料**(可选):
📄 文档资料 | 🔗 网页链接 | 🖼️ 参考图片 | 📝 文字描述
阶段 2:脚本规划
- 分析参考资料
- 生成视频场景规划脚本
- 发送脚本给用户确认
阶段 3:分场景生成确认
对每个场景,单独确认生成方式、时长和模型版本:
🎬 场景 1/3:开场展示
请选择生成方式:
A) 文生视频 - 直接用文字提示词
B) 图生视频 - 使用参考图片生成
请选择时长:
1️⃣ 10 秒(默认)
2️⃣ 15 秒
请选择模型版本:
① 2.0 - 最新版本,效果最佳(推荐)
② 1.5 - 稳定版本,广泛使用
③ 1.0 - 基础版本,兼容性好
阶段 4:提示词确认(关键!)
⚠️ 重要:必须先确认提示词,才能生成视频!
阶段 5:分场景生成(逐个进行!)
⚠️ 重要:逐个场景生成,不要批量!
阶段 6:视频合成
- 将所有确认的场景视频按顺序合成
- 添加转场效果
- 发送最终视频
🛠️ 工具调用
1. 火山引擎豆包 API(视频生成)
API Key: 65ae8f92-134c-4194-a3af-6e6cb74284e0
API 端点: https://ark.cn-beijing.volces.com/api/v3/contents/generations/tasks
文生视频 (T2V) ✅
import requests
payload = {
"model": "doubao-seedance-2-0-260128",
"content": [{"type": "text", "text": "一只可爱的小猫在草地上玩耍"}],
"parameters": {
"ratio": "16:9",
"duration": 10, # 10 秒(默认)或 15 秒
"watermark": False
}
}
response = requests.post(url, headers=headers, json=payload, timeout=30)
task_id = response.json()["id"]
图生视频 (I2V) ✅
payload = {
"model": "doubao-seedance-2-0-260128",
"content": [
{"type": "image_url", "image_url": {"url": "http://example.com/coffee.jpg"}},
{"type": "text", "text": "镜头缓缓推进,展示咖啡店温馨环境"}
],
"parameters": {
"ratio": "16:9",
"duration": 10,
"watermark": False
}
}
查询任务状态
status_url = f"https://ark.cn-beijing.volces.com/api/v3/contents/generations/tasks/{task_id}"
response = requests.get(status_url, headers=headers, timeout=10)
result = response.json()
# status: "pending" | "running" | "succeeded" | "failed"
video_url = result["content"][0]["url"]
可用模型列表
| 模型 ID | 版本 | 推荐 | 说明 |
|---|---|---|---|
doubao-seedance-2-0-260128 |
2.0 | ✅ 推荐 | 最新版本,效果最佳 |
doubao-seedance-2-0-fast-260128 |
2.0-fast | ✅ | 2.0 快速版 |
doubao-seedance-1-5-pro-251215 |
1.5 | ✅ | 稳定版本,广泛使用 |
doubao-seedance-1-0-pro-250528 |
1.0 | 基础版本,兼容性好 | |
doubao-seedance-1-0-pro-fast-251015 |
1.0-fast | 1.0 快速版 |
2. 通义万相 API(图片生成)
python3 /root/.openclaw/workspace/wanxiang_generate.py "提示词" output.png
⚙️ 配置参数
全局配置文件: ~/.openclaw/workspace/doubao-config.json
{
"default_api_key": "65ae8f92-134c-4194-a3af-6e6cb74284e0",
"default_text_to_video_model": "doubao-seedance-2-0-260128",
"default_duration": 10,
"last_updated": "2026-04-28T18:00:00"
}
视频生成参数
- 时长: 10 秒或 15 秒(默认 10 秒,用户可选)
- 分辨率: 480p / 720p / 1080p
- 比例: 16:9 / 9:16 / 1:1
- 帧率: 24 fps
⚠️ 注意事项
1. API 格式
- ✅ Seedance 使用
content: [{type: "text"}, {type: "image_url"}] - ✅ 已验证通过(2026-04-28)
2. 模型 ID 必须带版本号
- ❌ 错误:
doubao-seedance-1-5-pro - ✅ 正确:
doubao-seedance-1-5-pro-251215
3. 图片 URL 要求
- ✅ 必须使用 HTTP/HTTPS 公网可访问的 URL
4. 成本控制
- 先确认参考图再生成视频(避免浪费)
- 逐个场景生成
5. 时长说明
- 10 秒:约 1-2 分钟生成
- 15 秒:约 2-3 分钟生成
📁 文件管理
项目目录结构
/root/.openclaw/workspace/doubao-video-projects/
└── video_20260331_120000/
├── project.json
├── config.json
├── references/
├── characters/
├── scenes/
├── videos/
└── final_video.mp4
🔄 更新日志
v3.0 (2026-04-28)
新增功能:
- ✅ 默认时长从 5 秒改为 10 秒
- ✅ 支持用户选择时长:10 秒或 15 秒
- ✅ 支持用户选择模型版本:1.0 / 1.5 / 2.0
- ✅ 每个场景独立选择时长和模型
v2.2 (2026-04-28)
- ✅ 更新模型列表:添加 2.0 版本
- ✅ 文生图模型升级:wan2.6-t2i → wan2.7-image-pro
v2.1 (2026-04-03)
- ✅ 文生图模型升级:wan2.6-t2i → wan2.7-image-pro
v2.0 (2026-03-31)
- ✅ 首次使用配置流程
- ✅ 分场景生成方式选择
- ✅ 完整项目记录文件
Skill 版本: v3.0
创建时间: 2026-03-31
更新时间: 2026-04-28
作者: 卡妹 🌸
安全使用建议
This skill appears to be a legitimate Doubao/Seedance video-creator, but there are several red flags you should address before installing or running it:
1) Do not run publish scripts or any included shell scripts (publish.sh, quick_push_github.sh) until you inspect them line-by-line. They may invoke network operations or push repositories and were distributed with credential examples.
2) Treat any embedded API keys/tokens in the repository as leaked/compromised. Search the codebase for strings like 'ghp_', 'clh_', '65ae8f92...' and rotate those credentials if they belong to you or your org.
3) The skill's metadata declares no required env vars, yet SKILL.md expects you to provide and persist a Doubao API key; this mismatch is sloppy and could cause accidental credential persistence. Prefer prompting for API keys at runtime and storing them in a secure store rather than shipping defaults.
4) Review config write locations (~/.openclaw/workspace/doubao-config.json and project directories). Running as root or following examples with /root paths can cause files to be written with elevated privileges—run in a sandbox or non-root user.
5) Audit the Python modules (doubao_video_creator.py, publish scripts and any network code) for unexpected network endpoints, telemetry, or code that reads unrelated system files. Pay special attention to any code that sends data to endpoints not documented (SKILL.md lists ark.cn-beijing.volces.com for Doubao and a wanxiang_generate.py invocation).
6) If you decide to use the skill: remove hard-coded secrets from the repository, update SKILL.md to not include example secrets, add requires.env or explicit prompts for credentials, and re-check that no publish or CI helper contains credentials. Run the package in an isolated environment (container/VM) first.
If you want, I can: (a) search the repository for all likely secret patterns, (b) summarize any network calls in the Python files, or (c) list exact lines where credentials appear to help you remediate.
功能分析
Type: OpenClaw Skill
Name: doubao-video-creator
Version: 3.0.0
The skill bundle contains multiple hardcoded API keys for video and image generation services (Doubao and Tongyi Wanxiang) and a functional GitHub Personal Access Token (PAT) within the script `quick_push_github.sh`. This script is designed to force-push the skill's directory to a specific external repository (https://github.com/465367/doubao-video-creator.git). While these credentials appear to belong to the developer and there is no explicit instruction for the agent to exfiltrate user-specific secrets, the presence of automated push scripts and hardcoded authentication tokens in a distributable bundle constitutes a high-risk security flaw.
能力标签
能力评估
Purpose & Capability
The code files and SKILL.md align with a video-generation assistant (prompt generation, scene confirmation, calling Doubao Seedance). However the repository contains many publishing helpers, GitHub/ClawHub publishing instructions and example credentials which are peripheral to the core feature and increase the surface area.
Instruction Scope
SKILL.md instructs the agent to persist an API key to ~/.openclaw/workspace/doubao-config.json and shows concrete code that will call the Doubao API endpoint. It also references running local scripts at absolute paths (/root/.openclaw/...). The document contains explicit example API keys and other credential-like strings; storing/using them is outside of a minimal 'prompt-helper' scope and risks leaking secrets. The instructions also include publishing/automation steps (publish.sh, quick_push_github.sh) which may perform repository operations and prompt for credentials.
Install Mechanism
There is no install spec (instruction-only in registry metadata) but the package actually contains multiple Python modules and shell scripts. Not having an install step is plausible for a pure-skill distribution, but the included publish and push scripts (and absolute root paths) increase risk if a user executes them. No external download URLs or archive extracts were used.
Credentials
Registry metadata declared no required env vars or primary credential, yet SKILL.md expects a Doubao API key and even embeds a default API key string in multiple files/config examples. Several documentation files include what look like real tokens/credentials (GitHub token, ClawHub token, example API key). This is inconsistent and disproportionate: the skill should declare required credentials formally (and not ship hard-coded secrets).
Persistence & Privilege
The skill persists configuration and project files under ~/.openclaw and project directories; this is consistent with its purpose. It does not request always:true or other elevated persistent platform privileges. Still, the use of absolute root paths (/root/...) in examples can be problematic if run with elevated privileges.
如何使用
- 确保已安装 OpenClaw(本地或 Docker 部署)
- 在对话框中输入安装命令:
/install doubao-video-creator - 安装完成后,直接呼叫该 Skill 的名称或使用
/doubao-video-creator触发 - 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v3.0.0
默认时长改为 10 秒,支持用户选择 10s/15s,支持模型版本选择 1.0/1.5/2.0
v2.0.0
## doubao-video-creator v2.0.0 Changelog
- Added a publish log file (CLAWHUB_PUBLISH_LOG.md).
- Added a publish success confirmation file (CLAWHUB_SUCCESS.md).
v2.1.0
doubao-video-creator v2.1.0
- 新增首次使用配置引导,支持 API Key、模型版本等参数一次性项目化保存
- 分阶段优化:需求收集、脚本规划、关键元素生成、分场景方式选择、提示词确认、场景生成、合成管理,流程更清晰
- 支持文本生视频和图片生视频两种生成方式,并允许每个场景独立选择
- 引入提示词和原型图多轮确认机制,大幅减少废片,提高生成效率
- 文件与项目结构规范化,实现全项目全流程状态追踪
- 丰富交互示例,全面提升用户体验
元数据
常见问题
豆包视频创作助手 是什么?
基于火山引擎豆包视频生成模型,辅助用户配置项目并分场景确认提示词,逐步生成并合成专业短视频。 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 132 次。
如何安装 豆包视频创作助手?
在 OpenClaw 或 Claude Code 对话框中运行命令「/install doubao-video-creator」即可一键安装,无需额外配置。
豆包视频创作助手 是免费的吗?
是的,豆包视频创作助手 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。
豆包视频创作助手 支持哪些平台?
豆包视频创作助手 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。
谁开发了 豆包视频创作助手?
由 Cindypapa(@cindypapa)开发并维护,当前版本 v3.0.0。
推荐 Skills