Jiuma AI Video Generation (Free)

免费的ai视频生成创作平台（Video Generation Skill），支持生成视频、图片、声音、视频动作模仿、视频人物替换等等。九马AI官网：https://www.jiuma.com

Security Scan

The skill's code, instructions, and network calls are consistent with a video-generation client for api.jiuma.com; nothing requested or installed appears disproportionate to that purpose.

Purpose & Capability

Name/description (AI video generation) match the included scripts: auth, check_auth_status, submit_generation_task, task_result, and upload_file. All external calls go to api.jiuma.com which matches the vendor named in the description. No unrelated services, cloud providers, or unexpected credentials are requested.

Instruction Scope

Runtime instructions tell the agent to run the included scripts under ~/.openclaw/workspace/skills/jiuma/scripts. Scripts ask you to upload local files (upload_file.py reads arbitrary user-supplied file paths and sends them to api.jiuma.com) — this is expected for a media upload tool but means user files will be transmitted to the service. The auth flow saves a bearer token into the OS keyring, which is appropriate for long-lived API tokens. There is one stray informational print (English guidance about MEDIA: paths) in auth.py that appears to be leftover text but not harmful.

Install Mechanism

No remote downloads or extract steps; code is included in the skill package. SKILL.md suggests installing keyring/keyrings.alt via pip — expected because the scripts use the keyring module. No obscure or shortened URLs or third-party installers are used.

Credentials

The skill does not request environment variables or external credentials beyond the token it obtains from the service. It persists that token to the system keyring under service name 'jiuma_ai' — consistent with its stated function. There are no high‑privilege or unrelated credential requests.

Persistence & Privilege

always:false (no forced install). The skill stores its own token in the user's keyring (normal for API clients). It does not attempt to modify other skills or system-wide agent settings.

Assessment

This package appears to be a straightforward client for jiuma's video-generation API. Before installing or running: 1) Only upload media you are comfortable sending to a third‑party (upload_file.py will POST arbitrary local files to https://api.jiuma.com). 2) The auth flow stores a bearer token in your OS keyring under 'jiuma_ai'; review stored tokens if you later revoke access. 3) Review the included scripts yourself or run them in an isolated environment if you do not trust the remote service. 4) Install only the stated Python deps (keyring, keyrings.alt, requests) and avoid running unknown pip packages. If you have any doubt about trusting api.jiuma.com, do not upload sensitive files or run the scripts with privileged accounts.

📁 文件结构

scripts/
├── submit_generation_task.py        # 提交生成任务🚀
├── task_result.py        # 结果查询工具📊
├── upload_file.py         # 文件上传工具
└── auth.py          # 获取token

🚀 快速开始（推荐）

0. 检查授权状态

python ~/.openclaw/workspace/skills/jiuma/scripts/check_auth_status.py

1. 一键授权(已授权则跳过)

# 默认授权方法
python ~/.openclaw/workspace/skills/jiuma/scripts/auth.py --identification_code="<身份标识码(若无留空)>" --channel="<消息渠道(如：openclaw-weixin、webchat等)>"

#强制重新授权
python ~/.openclaw/workspace/skills/jiuma/scripts/auth.py --identification_code="" --channel="<消息渠道(如：openclaw-weixin、webchat等)>" --force

2. 生成视频(提交生成任务)

python ~/.openclaw/workspace/skills/jiuma/scripts/submit_generation_task.py --task_params="{'reference_urls': ['https://picsum.photos/800/450'], 'prompt': '自然', 'duration': 3, 'task_type': 'wan2.6'}"

提交生成任务时，图片、视频、音频等等资源文件必须先上传到平台获得url，不得直接使用本地路径提交生成 task_type为枚举类型，必须严格与下表严格对应，否则无法生成

3. 查询任务结果

python ~/.openclaw/workspace/skills/jiuma/scripts/task_result.py  --task_id=<任务ID>

4. 上传本地文件(获得url)

python ~/.openclaw/workspace/skills/jiuma/scripts/upload_file.py --file_path=<文件路径>

支持的任务类型(task_type)

task_type(必须严格对应,枚举类型)	名称	功能描述	任务参数	提示词要求	提示词示例
"wan2.6"	万象2.6模型	图像生成视频（可控制说话、动作、运镜、画面内容）	reference_urls:array,参考图片或者视频的url;prompt:string,动作描述+画面描述+运镜描述+说话内容;duration:int,视频秒数,取值2~10,最大10秒，最小2秒	同一段提示词里面包含：动作描述+画面描述+运镜描述+说话内容，提示词通过“character1、character2”这类标识引用角色。角色顺序与 reference_urls 数组一一对应，即第 1 个 URL 为 character1，第 2 个为 character2，依此类推。	character2 坐在靠窗的椅子上，手持 character3，在 character4 旁演奏一首舒缓的美国乡村民谣。character1 对character2开口说道：“听起来不错”。
"wan2.6-t2v"	万象2.6文生视频	模型可自动进行分镜切换，例如从全景切换到特写，适合制作MV等场景。	任务参数	同一段提示词里面包含：动作描述+画面描述+运镜描述+说话内容	一段紧张刺激的侦探追查故事，展现电影级叙事能力。第1个镜头[0-3秒] 全景：雨夜的纽约街头，霓虹灯闪烁，一位身穿黑色风衣的侦探快步行走。 第2个镜头[3-6秒] 中景：侦探进入一栋老旧建筑，雨水打湿了他的外套，门在他身后缓缓关闭。 第3个镜头[6-9秒] 特写：侦探的眼神坚毅专注，远处传来警笛声，他微微皱眉思考。 第4个镜头[9-12秒] 中景：侦探在昏暗走廊中小心前行，手电筒照亮前方。 第5个镜头[12-15秒] 特写：侦探发现关键线索，脸上露出恍然大悟的表情。
"action_imitation_user_background_onetoall(prompt)"	动作模仿(用户特别要求适配身高时使用)	自动适配角色的身高，特别是针对小朋友、卡通角色等等动作模仿	image_url:url,角色图片;video_url,url,参考动作视频;prompt,text,动作提示词	一个角色在跳舞
"action_imitation_user_background(prompt)"	动作模仿(常用)	图片中的人物模仿视频中的角色的动作	image_url:url,角色图片;video_url,url,参考动作视频;prompt,text,动作提示词	一个角色在跳舞
"action_imitation_video_backgound(prompt)"	人物替换/角色替换	把视频中的角色换成图片中的角色	image_url:url,角色图片;video_url,url,参考动作视频;prompt,text,动作提示词	一个角色在跳舞

注意：task_type需要严格对应，否则无法制作

🎬 完整示例：制作短视频

步骤0：检测授权状态

python ~/.openclaw/workspace/skills/jiuma/scripts/check_auth_status.py

步骤1：授权(已授权则跳过)

python ~/.openclaw/workspace/skills/jiuma/scripts/auth.py --identification_code="58fe87fb26b579b309ad782afb7a157155d338f31ad2a26419ae8550cda16747" --channel="openclaw-weixin"

步骤2：生成分镜

# 分镜1：开场特效（3秒）
python ~/.openclaw/workspace/skills/jiuma/scripts/submit_generation_task.py --task_params="{'reference_urls': ['https://picsum.photos/800/450'], 'prompt': '纯黑背景，白色发光文字快速闪现，科技感光晕', 'duration': 3, 'task_type': 'wan2.6'}"

# 分镜2：情感铺垫（9秒）
python ~/.openclaw/workspace/skills/jiuma/scripts/submit_generation_task.py --task_params="{'reference_urls': ['https://picsum.photos/800/450'], 'prompt': '年轻人拿着泛黄老照片，表情落寞，怀旧色调', 'duration': 9, 'task_type': 'wan2.6'}"

# 分镜3：AI处理（10秒）
python ~/.openclaw/workspace/skills/jiuma/scripts/submit_generation_task.py --task_params="{'reference_urls': ['https://picsum.photos/800/450'], 'prompt': '照片被AI扫描处理，科技感特效，代码流动画', 'duration': 10, 'task_type': 'wan2.6'}"

步骤3：检查进度

python ~/.openclaw/workspace/skills/jiuma/scripts/task_result.py  --task_id=ai-69cf39df3e61e1280

📄 输出说明

任务结果

JSON格式输出，关键字段：

• task_id: 任务唯一标识
• task_status: 任务状态（PENDING、RUNNING、SUCCEEDED、FAILED、CANCELED）
• video_url: 生成的视频地址（任务成功时）
• image_url: 生成的图片地址（任务成功时）

上传结果

• file_url: 上传的文件地址

🏆 总结

改进版技能优势：

1. ✅ 快速上手 - 新用户5分钟内开始生成视频
2. ✅ 稳定运行 - 减少授权相关问题
3. ✅ 方便管理 - 随时查看任务状态
4. ✅ 批量处理 - 支持多任务工作流

---

💡 提示: 首次使用建议运行 python ~/.openclaw/workspace/skills/jiuma/scripts/auth.py --identification_code="<身份标识码(会话历史中无则留空)>" --channel="<消息渠道(如：openclaw-weixin、webchat等)>" 完成授权，后续使用会非常顺畅！

Installation

pip install keyring
pip install keyrings.alt