← 返回 Skills 市场

Chanjing Video Compose

Name: Chanjing Video Compose
Author: iamzn1018

作者 IAMZn · GitHub ↗ · v1.0.3 · MIT-0

cross-platform ✓ 安全检测通过

213

总下载

当前安装

版本数

在 OpenClaw 中安装

/install chanjing-video-compose

功能描述

Use Chanjing video synthesis APIs to create digital human videos from text or audio, with optional background upload, task polling, and explicit download whe...

使用说明 (SKILL.md)

Chanjing Video Compose

技能包标识：chanjing-video-compose。

功能说明

调用蝉镜视频合成 Open API：列举公共/定制形象、上传素材、文本或音频驱动数字人、轮询任务；用户明确要求时用脚本下载成片。本 skill 脚本不依赖 ffmpeg/ffprobe（与一键成片编排不同）。机器可读凭据路径与敏感字段见篇首 metadata.openclaw.credentialModel。

运行依赖

python3 与同仓库 scripts/*.py
无 ffmpeg/ffprobe 门控

环境变量

CHANJING_CONFIG_DIR（可选）：credentials.json 所在目录，默认 ~/.chanjing
CHANJING_API_BASE（可选）：Open API 基址，默认 https://open-api.chanjing.cc

使用命令

ClawHub（slug 以注册表为准）：clawhub run chanjing-video-compose
本仓库：python skills/chanjing-video-compose/scripts/create_task.py …（见 Standard Workflow）

登记摘要（英文 · ClawHub / OpenClaw）

Primary credential (not primaryEnv): app_id + secret_key in ~/.chanjing/credentials.json (or $CHANJING_CONFIG_DIR/credentials.json); access_token / expiry read/written in the same file. primaryEnv is omitted on purpose (OpenClaw uses it for a single env-injected API key; this client uses file-based dual keys).

Required vs optional: CHANJING_API_BASE is optional (defaults to https://open-api.chanjing.cc). CHANJING_CONFIG_DIR is optional. No ffmpeg / ffprobe gate in this skill's metadata—orchestration-only use of this skill does not bundle local concat.

Purpose alignment: Chanjing digital-human video compose API client (list figures, upload assets, create/poll tasks, optional download_result.py). Trust: HTTPS to the API host and URLs returned in responses when saving media.

Chinese Q&A: 审阅四条对表 under Preconditions → 安全与凭据（登记摘要）；与 credential_hint、metadata.openclaw.credentialModel 一致。

When to Use This Skill

当用户要做这些事时使用本 Skill：

创建数字人视频合成任务
用文本驱动数字人出镜
用本地音频驱动数字人视频
查询公共数字人或定制数字人形象
轮询视频合成结果
在用户明确要求时下载最终视频到本地

如果需求更接近“上传一段真人视频做对口型驱动”，优先使用 chanjing-avatar，不要混用。

Preconditions

执行本 Skill 前，必须先通过 chanjing-credentials-guard 完成 AK/SK 与 Token 校验。

本 Skill 与 guard 共用：

~/.chanjing/credentials.json
https://open-api.chanjing.cc

无凭证时，脚本会自动打开蝉镜登录页，并提示配置命令。

审阅四条对表（Purpose / 主凭据 / 指令范围 / 持久化）

以下为对「Purpose & Capability / Instruction Scope / Credentials / Persistence」类评审话术的直接对表说明；与代码、篇首英文登记摘要、description frontmatter 一致。

#	关切	说明
1	名称/能力与实现一致？主凭据是否写进登记？	一致。本 skill 为蝉镜数字人视频合成 API 客户端（形象列表、上传、创建/轮询任务、用户明确要求时 `download_result.py`）。`CHANJING_API_BASE` 不是必填（有默认基址）。主凭据为 `credentials.json` 内 `app_id` / `secret_key` 及刷新后的 `access_token`；已在 `description`、`credential_hint`、`metadata.openclaw.credentialModel`、篇首英文摘要、下表安全与凭据写明。不声明 `primaryEnv`（双字段文件凭据，与单一 env Key 模型不符）。本 skill 不将 `ffmpeg`/`ffprobe` 写入 `metadata.requires.bins`（与一键成片不同）。
2	运行时指令范围与敏感数据、任意 URL	符合声明目的。脚本读写 `CHANJING_CONFIG_DIR/credentials.json`；缺凭证时可能浏览器或 `open_login_page.py`；向 Open API 上传/下载；用户明确要求落盘时 `download_result.py` 会请求接口返回的媒体 URL——请自行判断是否信任 API 主机与链接。若使用 `create_task.py --callback \x3CURL>`，蝉镜服务端可能向该 URL 推送任务结果——仅使用你信任且可接收回调的端点。
3	环境变量适合客户端，主凭据在文件	`CHANJING_CONFIG_DIR` / `CHANJING_API_BASE` 为可选配置。主凭据非单一注入式 env Key：`app_id` / `secret_key` 经 guard 等写入 `credentials.json`（路径见 `metadata.openclaw.credentialModel`）；`access_token` 持久化到磁盘，属敏感；勿提交版本库、勿在对话中回显完整密钥。
4	持久化与特权	凭据与 token 写入约定配置文件；可能交互式登录。`always: false`（默认）；不修改其它 skill 或全局 Agent 配置。

安全与凭据（登记摘要）

与 审阅四条对表、篇首英文登记摘要、description 一致。

维度	说明
主凭据	`app_id`、`secret_key` 存于 `credentials.json`（默认 `~/.chanjing/`，`CHANJING_CONFIG_DIR` 可改目录）。经 `chanjing-credentials-guard` 配置。非 OpenClaw `primaryEnv` 模型（非单一环境变量 API Key）。
Token	脚本将 `access_token` / `expire_in` 写回同一文件；与 `chanjing-credentials-guard` 及其它蝉镜子技能共用该文件（磁盘持久化、跨技能状态）。
回调（可选）	`create_task.py --callback`：若提供，API 可能向用户 URL 发送任务相关负载；属出站到你方端点的信任边界，请自行评估。
环境变量	`CHANJING_API_BASE`、`CHANJING_CONFIG_DIR` 可选。
网络与浏览器	HTTPS `https://open-api.chanjing.cc`；无凭证时可能 `webbrowser.open` 或通过 guard 打开登录页。
下载与信任	仅在用户明确要求落盘时调用 `download_result.py`，请求接口返回的媒体 URL；请信任蝉镜 API 与其返回链接。
持久化	`always: false`（默认）；不修改其它 skill 或全局 Agent 配置。

Standard Workflow

先让用户明确选择数字人来源：common（公共数字人）或 customised（定制数字人）
调用 list_figures.py --source \x3Ccommon|customised>（建议 --json，公共源可加大 --page-size 或翻页）获取可用形象；在候选内对比 name、各 figure 的 type 与分辨率、audio_man_id、audio_name（若有）与任务人设后再选定 person.id。禁止未比较就默认列表最前几项。
如果选择公共数字人，还要再确认 figure_type（与所选 figures[].type 一致），例如 sit_body / whole_body / circle_view。无用户特殊要求时，默认优先年轻、有活力的形象（名称/audio_name 偏青年、学生、元气等）；题材需要成熟或中老年气质时再改选。
若使用文本驱动，确定 audio_man_id
在创建任务前，必须明确询问用户字幕偏好：show（保留字幕）或 hide（隐藏字幕）
如果用户选择 show 但没有提出自定义样式或位置需求，直接使用官方文档推荐默认值；只有在用户明确想调整字幕位置或样式时，才继续追问 subtitle_config 参数
若用户要定制字幕位置，说明坐标以左上角为原点，再补充 subtitle_config 相关参数
若使用本地音频或背景图，先调用 upload_file.py 获取 file_id
调用 create_task.py 创建视频合成任务，得到 video_id
调用 poll_task.py 轮询直到成功，得到 video_url
只有在用户明确要求保存到本地时，才调用 download_result.py

Covered APIs

本 Skill 当前覆盖：

GET /open/v1/list_common_dp
POST /open/v1/list_customised_person
POST /open/v1/create_video
GET /open/v1/video
GET /open/v1/common/create_upload_url
GET /open/v1/common/file_detail

Scripts

脚本目录：

skills/chanjing-video-compose/scripts/

脚本	说明
`_auth.py`	读取凭证、获取或刷新 `access_token`
`list_figures.py`	按 `--source common
`upload_file.py`	上传音频或背景素材，轮询到文件可用后输出 `file_id`
`create_task.py`	创建视频合成任务；使用公共数字人时可补充 `--figure-type ...`，字幕支持 `--subtitle show
`poll_task.py`	轮询视频详情直到完成，默认输出 `video_url`
`download_result.py`	下载最终视频到 `outputs/video-compose/`

Usage Examples

示例 1：公共数字人文本驱动

# 1. 先列公共数字人
python skills/chanjing-video-compose/scripts/list_figures.py --source common

# 2. 用公共数字人创建文本驱动视频
VIDEO_ID=$(python skills/chanjing-video-compose/scripts/create_task.py \
  --person-id "C-ef91f3a6db3144ffb5d6c581ff13c7ec" \
  --figure-type "sit_body" \
  --audio-man "C-0ae461135d8a4eb2b59c853162ea9848" \
  --subtitle "show" \
  --subtitle-x 31 \
  --subtitle-y 1521 \
  --subtitle-width 1000 \
  --subtitle-height 200 \
  --subtitle-font-size 64 \
  --subtitle-stroke-width 7 \
  --text "你好，这是一个蝉镜视频合成测试。")

# 3. 轮询到完成，拿到 video_url
python skills/chanjing-video-compose/scripts/poll_task.py --id "$VIDEO_ID"

示例 2：定制数字人上传本地音频驱动

python skills/chanjing-video-compose/scripts/list_figures.py --source customised

AUDIO_FILE_ID=$(python skills/chanjing-video-compose/scripts/upload_file.py \
  --service make_video_audio \
  --file ./input.wav)

VIDEO_ID=$(python skills/chanjing-video-compose/scripts/create_task.py \
  --person-id "C-ef91f3a6db3144ffb5d6c581ff13c7ec" \
  --subtitle "hide" \
  --audio-file-id "$AUDIO_FILE_ID")

python skills/chanjing-video-compose/scripts/poll_task.py --id "$VIDEO_ID"

示例 3：显式下载最终视频

python skills/chanjing-video-compose/scripts/download_result.py \
  --url "https://example.com/output.mp4"

Download Rule

下载是显式动作，不是默认动作：

poll_task.py 成功后应先返回 video_url
不要自动下载结果文件
只有当用户明确表达“下载到本地”“保存到 outputs”“帮我落盘”时，才执行 download_result.py

Figure Selection Rule

选择数字人时遵循这条规则：

如果用户要用平台已有人物库，先走公共数字人：list_figures.py --source common
如果用户要用自己训练或上传生成的人物，先走定制数字人：list_figures.py --source customised
使用公共数字人创建视频时，可按所选形态传 --figure-type \x3Ctype>
使用定制数字人时，不需要 figure_type

Subtitle Rule

字幕遵循这条规则：

不要默认假设用户要字幕或不要字幕
创建任务前，必须先明确询问用户选择：show 或 hide
若由 chanjing-one-click-video-creation 的 run_render.py 调用 create_task.py，以当次 workflow.json 根级 subtitle_required 为准（默认 false → --subtitle hide；true → show 及推荐样式），无需为该一键成片路径再单独追问字幕开关，除非用户在需求里明确要求改字幕策略
用户选择保留字幕时，调用 create_task.py --subtitle show
若用户未指定字幕位置或样式，直接使用官方推荐默认值；create_task.py 在未传 --subtitle-color 时默认白字 color=#FFFFFF：1080p 为 x=31 y=1521 width=1000 height=200 font_size=64 stroke_width=7 asr_type=0；4K 画布为 x=80 y=2840 width=2000 height=1000 font_size=150 stroke_width=7 asr_type=0（两组均含 color=#FFFFFF）
用户选择隐藏字幕时，调用 create_task.py --subtitle hide 或兼容旧用法 --hide-subtitle
若用户要求调整字幕位置或样式，可继续传 --subtitle-x / --subtitle-y / --subtitle-width / --subtitle-height / --subtitle-font-size / --subtitle-color / --subtitle-stroke-color / --subtitle-stroke-width / --subtitle-font-id / --subtitle-asr-type
坐标基于左上角原点；字幕区域不能超出 screen_width / screen_height
如果用户只说“要字幕”但没指定位置，不必再追问具体数值；除非用户明确要调位置，否则直接走默认值

Output Convention

默认本地输出目录：

outputs/video-compose/

Additional Resources

更多接口细节见：

skills/chanjing-video-compose/reference.md
skills/chanjing-video-compose/examples.md

安全使用建议

This skill is coherent with its description, but review and accept these points before installing: (1) it reads and writes credentials.json (app_id/secret_key) in ~/.chanjing (or $CHANJING_CONFIG_DIR) and persists access_token to disk — do not commit that file to source control; (2) only run the included scripts if you trust https://open-api.chanjing.cc (the API host) because the skill will upload files and fetch media URLs returned by the service; (3) be careful when using --callback: providing a callback URL lets the service POST task results to that endpoint (only use endpoints you control/trust); (4) download_result will fetch and write arbitrary URLs you provide — only download from trusted URLs; (5) inspect the credentials-guard companion before using it if you are concerned about how app_id/secret_key are obtained or stored.

功能分析

Type: OpenClaw Skill Name: chanjing-video-compose Version: 1.0.3 The chanjing-video-compose skill is a legitimate integration for the Chanjing video synthesis API. It includes scripts for listing digital human figures, uploading media assets, creating synthesis tasks, and downloading results. Credential management (app_id/secret_key) is handled via a local JSON file (~/.chanjing/credentials.json) as explicitly declared in the OpenClaw metadata. The code uses standard Python libraries (urllib, json) and follows the stated purpose without any evidence of malicious intent, data exfiltration, or unauthorized execution.

能力评估

✓ Purpose & Capability

Name/description, the included scripts, and declared metadata all describe a Chanjing video-composition client. The scripts only call the documented Open API endpoints (list, upload, create_video, video status, file_detail) and require the app_id/secret_key/access_token stored in credentials.json — this is coherent with the stated purpose.

✓ Instruction Scope

Runtime instructions and scripts read/write ~/.chanjing/credentials.json (or $CHANJING_CONFIG_DIR) and call the Chanjing API. They may open a browser to the login page if credentials are missing. create_task supports an optional --callback (server-side callback to a user-supplied URL) and download_result downloads a user-provided URL; both behaviors are documented and require explicit user action. The scripts do not access other system secrets or unexpected paths.

✓ Install Mechanism

No install spec; this is instruction+scripts only. No network install or archive extraction is performed by the skill metadata. The scripts are plain Python and depend on the presence of python3, which is documented.

ℹ Credentials

The skill intentionally uses file-based dual credentials (app_id/secret_key in credentials.json) rather than a single environment API key; CHANJING_CONFIG_DIR and CHANJING_API_BASE are optional. Persisting access_token/expire_in to disk is explicit in the metadata and scripts — this is expected for the chosen credential model, but users should be aware that sensitive tokens are stored on disk and shared across sibling Chanjing skills.

✓ Persistence & Privilege

always:false and the skill does not modify other skills or global agent configuration. It writes/reads its own credentials file (~/.chanjing/credentials.json) which is expected for this client and documented; no elevated platform privileges are requested.

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 chanjing-video-compose
安装完成后，直接呼叫该 Skill 的名称或使用 /chanjing-video-compose 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

v1.0.3

**Chanjing Video Compose v1.0.3** - Clarified credential usage: added explicit `credential_hint` and machine-readable `metadata.openclaw.credentialModel`, detailing the file path, optional env variables, and sensitive fields to never commit. - Documented new behavior: usage of `open_login_page.py` when credentials are missing. - Noted new optional `--callback` argument for `create_task.py` to enable API-side payload POSTs to user endpoints (trust boundary clarified). - Expanded/clarified credential sharing and persistence policies, including how access tokens persist to disk and are shared with `chanjing-credentials-guard` and sibling skills. - Updated English and Chinese credential/policy descriptions for audit clarity and precise OpenClaw/ClawHub registration. - No script or runtime behavior changes; changelog applies to documentation and manifest/metadata only.

v1.0.2

chanjing-video-compose 1.0.2 - Added four new script entrypoints: create_task.py, download_result.py, list_figures.py, poll_task.py, and upload_file.py. - Migrated all script filenames and documentation references from old names to `.py` suffix (e.g., `poll_task.py` instead of `poll_task`). - Updated documentation (SKILL.md) for clearer workflow, more concise structure, and explicit Python script usage. - Expanded metadata to include author, binaries, env, category, and tags fields for clarity. - No changes to required dependencies or API coverage.

v1.0.1

**Expanded credential and selection guidance, public/common & subtitle logic** - Clarified credential usage in description and frontmatter: now specifies credentials.json, dual key, access_token persistence, and optional environment variables; explicitly notes no ffmpeg/ffprobe dependency. - Expanded workflow and rules for selecting digital humans: users must explicitly pick from 'common' or 'customised' sources, and compare candidate models before proceeding. - Added detailed, stepwise guidance for subtitle handling: must ask user about subtitle preference before creation; defaults and customizations are delineated. - Enhanced security and credential documentation: includes English/Chinese Q&A; reviewed for alignment with ClawHub/OpenClaw practices. - Updated covered APIs and scripts table to reflect support for common figures (`GET /open/v1/list_common_dp`), expanded arguments, and clarified distinctions between public and customized figure creation.

v1.0.0

Initial release of chanjing-video-compose skill. - Enables creation of digital human videos via Chanjing API using text or local audio. - Supports background uploads, polling for video synthesis results, and explicit download on user request. - Requires prior credential verification with chanjing-credentials-guard. - Covers key APIs: list figures, create video, poll video status, upload/download files. - Adds structured scripts for each workflow step. - Download of final videos is only performed when explicitly requested by the user.

元数据

Slug chanjing-video-compose

版本 1.0.3

许可证 MIT-0

累计安装 3

当前安装数 3

历史版本数 4

常见问题