video-generation-minimax

视频生成技能，使用 MiniMax 视频生成 API 创建视频

Security Scan

The skill mostly does what it claims (call a MiniMax video API) but has mismatches between its manifest/instructions and the included script—most importantly an undeclared required API key and a misleading claim about where outputs are saved/sent.

Purpose & Capability

Name/description match the code: the script calls a MiniMax video-generation API. However the registry metadata lists no required environment variables while both SKILL.md and scripts/video_gen.py require MINIMAX_API_KEY. Also SKILL.md claims generated videos are 'automatically downloaded to {WorkspaceDir}/video-generation/ and then sent to the user' but the script only writes a file to the current working directory (or --output path) and contains no logic to move files to a workspace directory or to transmit them to the user.

Instruction Scope

Runtime instructions are limited to calling the MiniMax API and downloading the produced file. SKILL.md instructs installing requests and setting MINIMAX_API_KEY (consistent with code). The instructions' claim that the agent will place outputs in {WorkspaceDir}/video-generation/ and send them to users is not implemented in the script, giving the agent broader implied behavior than the code actually performs. The 'subject' mode accepts face images — a privacy consideration not emphasized in the SKILL.md.

Install Mechanism

No install spec; this is instruction+script only. The only runtime dependency is python3 and Python package 'requests' (requested), which is proportionate for an HTTP client script. The script makes network calls only to the declared API host and to returned download URLs.

Credentials

The script requires a single API credential MINIMAX_API_KEY (read from environment) but the skill manifest/registry metadata declared no required env vars. This mismatch means the catalog entry understates the credential needed. Requesting one API key is reasonable for this purpose, but it must be declared so users can make an informed decision.

Persistence & Privilege

The skill does not request always:true and does not modify other skills or system settings. It runs as an on-demand script and saves output locally; no persistent elevated privileges are requested.

What to consider before installing

Before installing: (1) Be aware the skill requires an API key MINIMAX_API_KEY — the registry metadata does not declare this, so you must provide it manually. Only use a key with limited scope and from a trusted MiniMax account. (2) The SKILL.md claims outputs will be placed in {WorkspaceDir}/video-generation/ and sent to users; the included script instead saves to the current directory or the --output path and does not implement sending — treat the 'automatic send' claim as incorrect. (3) The 'subject' mode accepts face images; consider privacy and consent implications before uploading or pointing to pictures of real people. (4) Verify the MiniMax endpoint/domain (api.minimaxi.com / platform.minimaxi.com) is legitimate for the provider you expect. (5) Because the script downloads a file from a URL returned by the API, run it in a sandbox or isolated environment if you are concerned about untrusted content. If you want to proceed, update the registry metadata to declare MINIMAX_API_KEY and optionally correct the SKILL.md to match the actual file save/send behavior.

使用 MiniMax 视频生成 API 创建视频。支持4种生成模式：

1. 文生视频：根据文本描述生成视频
2. 图生视频：基于图片+文本描述生成视频
3. 首尾帧：首图+尾图+文本生成视频
4. 主体参考：人脸照片+文本，保持人物特征一致

前置要求

• 安装 Python 依赖：pip3 install requests
• 设置环境变量 MINIMAX_API_KEY

使用方法

方式一：Python 脚本（推荐）

python3 {baseDir}/scripts/video_gen.py --mode text --prompt "描述文字"
python3 {baseDir}/scripts/video_gen.py --mode image --prompt "描述文字" --image "图片URL"
python3 {baseDir}/scripts/video_gen.py --mode start_end --prompt "描述文字" --first "首图URL" --last "尾图URL"
python3 {baseDir}/scripts/video_gen.py --mode subject --prompt "描述文字" --subject "人脸图片URL"

参数说明

参数	说明	必填
--mode	模式：text/image/start_end/subject	是
--prompt	视频描述文本	是
--image	图生视频的首帧图片URL	image模式必填
--first	首尾帧模式的首帧图片URL	start_end模式必填
--last	首尾帧模式的尾帧图片URL	start_end模式必填
--subject	主体参考模式的人脸图片URL	subject模式必填
--duration	视频时长：6/10	否，默认6
--resolution	分辨率：720P/1080P	否，默认1080P
--output	输出文件名	否，默认 output.mp4

示例

# 文生视频
python3 {baseDir}/scripts/video_gen.py --mode text --prompt "镜头拍摄一个女性坐在咖啡馆里，女人抬头看着窗外"

# 图生视频
python3 {baseDir}/scripts/video_gen.py --mode image --prompt "Contemporary dance" --image "https://example.com/image.png"

# 首尾帧生成
python3 {baseDir}/scripts/video_gen.py --mode start_end --prompt "A little girl grow up" --first "https://example.com/start.jpg" --last "https://example.com/end.jpg"

# 主体参考
python3 {baseDir}/scripts/video_gen.py --mode subject --prompt "在街头走路的男士" --subject "https://example.com/face.jpg"

数据来源

• API Key：从环境变量 MINIMAX_API_KEY 读取
• API 文档：https://platform.minimaxi.com/docs/llms.txt

注意事项

• 视频生成是异步过程，需要轮询任务状态
• 推荐轮询间隔为10秒
• 生成完成后自动下载视频到你的工作空间 {WorkspaceDir}/video-generation/目录下随后发送给用户
• 模型版本：文生视频用 MiniMax-Hailuo-2.3，主体参考用 S2V-01