Story Video Generator

从图片或文字描述自动生成完整视频故事。支持灵活输入（1-N张图片/纯文字/混合），可选时长和风格。当用户要求生成视频故事、视频短片、图片转视频、文字转视频时使用。

Security Scan

The skill's instructions are generally consistent with a video-generation helper, but it omits and glosses over important dependencies and external-tool behavior (FFmpeg installation and opaque gen_* tools), which is disproportionate and unclear.

Purpose & Capability

Name and description match the runtime instructions: script → frames → segments → BGM → merge. However, the skill assumes the presence (or automatic installation) of FFmpeg and relies heavily on opaque tools (images_understand, gen_images, gen_videos, gen_music) without declaring them as required binaries, services, or credentials — this omission is unexpected for an end-to-end generator.

Instruction Scope

SKILL.md stays within the stated purpose: it reads user images/text, generates scripts, writes outputs under output/, and runs FFmpeg to merge. It does not instruct access to unrelated system files or secrets. The only scope concern is the broad autonomy given to call external generators (gen_* and LLM) whose network/data handling is unspecified.

Install Mechanism

There is no install spec, yet the instructions say 'check FFmpeg and install if not available'. Without an install mechanism, it is unclear how FFmpeg would be installed (which package source, whether network downloads occur, or whether elevated privileges are required). This ambiguity increases risk.

Credentials

The skill requests no environment variables, credentials, or config paths. That is proportionate to its stated purpose. However, because it invokes opaque gen_* services and an LLM, the absence of any declared API keys or endpoints is a gap: if those tools use remote APIs, credentials or network access will be needed but are not documented.

Persistence & Privilege

The skill is not always-enabled and does not request persistent system-wide privileges. It writes files only under an output/ directory as part of normal operation. No modifications to other skills or global agent settings are described.

What to consider before installing

This skill appears to implement the claimed video-generation workflow, but before installing you should verify a few things: (1) Confirm what the gen_* tools and the 'images_understand' and LLM calls actually are — are they local binaries/plugins, or cloud APIs that will send user images/text off-host? (2) Clarify how FFmpeg will be installed if missing (which installer, will it download from the network, require sudo?). (3) Ask whether any external services require API keys or billing (none are declared). (4) If privacy is a concern, require explicit confirmation that user media will not be uploaded to third-party endpoints, or run the skill in a sandboxed environment. If the developer can provide an install spec (or declare that gen_* are built-in safe tools and how FFmpeg is installed), the remaining issues would be resolved.

故事视频生成助手

从用户提供的图片或文字描述自动生成完整的视频故事。完整流程：脚本生成 → 主体参考图 → 首帧图片 → 视频片段 → 背景音乐 → 最终合成。

快速开始

输入要求

• 图片模式：1-N 张图片（不强制分类，AI 自动识别）
• 文字模式：纯文字描述故事
• 混合模式：图片 + 文字补充

可选参数

• 时长：24秒（4段）/ 48秒（默认，8段）/ 72秒（12段）
• 风格（纯文字时）：吉卜力 / 赛博朋克 / 写实 / 水彩 / 像素 / 动漫 / 油画 / 极简 / AI推荐

输出格式

生成视频后用以下格式输出：

<deliver_assets>
<item>
<path>视频路径</path>
</item>
</deliver_assets>

执行流程

Step 0: 环境检查

1. 检查 FFmpeg 是否可用，不可用则安装
2. 接收用户输入（图片/文字）
3. 确定时长（未指定默认 48 秒）
4. 纯文字时确定风格（未指定则 AI 推荐）

Step 1: 生成故事脚本

• 有图片 → 用 images_understand 分析图片，生成脚本
• 纯文字 → 用 LLM 生成脚本
• 输出：output/story_script.json

Step 1.5: 主体参考图生成

• 基于 analysis.subject 生成主体参考图
• 作为整个视频视觉一致性的锚点
• 输出：output/subject_reference.png

Step 2: 首帧图片生成（串联）

• 必须逐帧生成，每帧用主体参考图 + 上一帧作为双重参考
• 禁止并行生成
• 输出：output/frames/frame_01.png - frame_N.png

Step 3: 视频片段生成

• 从首帧图片生成视频片段
• 参数：duration=6秒，resolution=768P
• 输出：output/videos/segment_01.mp4 - segment_N.mp4

Step 4: 背景音乐生成（可与 Step 3 并行）

• 生成无歌词 BGM，时长等于视频总时长
• 输出：output/bgm.mp3

Step 5: 视频拼接与音乐合成

• 使用 FFmpeg 拼接视频片段并叠加 BGM
• 输出：output/final_video.mp4

关键约束

参数	值
每段时长	6秒（固定）
视频分辨率	768P
背景音乐	无歌词纯音乐

常用工具

工具	用途
images_understand	分析图片生成脚本
gen_images	生成首帧图片
gen_videos	生成视频片段
gen_music	生成背景音乐
exec + FFmpeg	拼接视频与合成音频

常见错误

1. 首帧并行生成 → 角色一致性无法保证
2. 每段未重复主体特征 → 角色长相漂移
3. BGM 带歌词 → 必须强调 instrumental, no vocals
4. 跳过主体参考图 → 视觉一致性无法保证