功能描述

Use Chanjing text-to-digital-person APIs for AI portraits, talking videos, optional LoRA training, polling, and explicit downloads when requested.

使用说明 (SKILL.md)

Chanjing Text To Digital Person

Name: Chanjing Text To Digital Person
Author: binkes

功能说明

文生图、图生说话视频、可选 LoRA 训练与轮询；用户明确要求时下载生成物。凭据与权限见 manifest.yaml。脚本不依赖 ffmpeg/ffprobe。

运行依赖

python3 与同仓库 scripts/*.py（含 _auth.py、_task_api.py）
无 ffmpeg/ffprobe 门控

环境变量与机器可读声明

环境变量键名与说明：manifest.yaml（environment 段）及本文
变量、凭据模型、合规 permissions、clientPermissions、agentPolicy：manifest.yaml

使用命令

ClawHub（slug 以注册表为准）：clawhub run chanjing-text-to-digital-person
本仓库：python skills/chanjing-text-to-digital-person/scripts/create_photo_task.py …（见 Standard Workflow）

登记与审稿（单一事实来源）

路径、primaryEnv 省略、persistAccessTokenOnDisk、敏感字段、agentPolicy、可选 env 等：以 manifest.yaml 为准。实现上由 _auth.py、_task_api.py 与各 CLI 脚本承担；本篇从 When to Use 起写流程。

When to Use This Skill

当用户要做这些事时使用本 Skill：

根据人物提示词生成数字人形象图
把生成的人物图转成会说话的短视频
查询文生图 / 图生视频 / LoRA 任务状态
在用户明确要求时，把生成图片或视频下载到本地

如果需求是“上传真人素材训练定制数字人”，优先使用 chanjing-customised-person。
如果需求是“拿已有数字人做口播视频合成”，优先使用 chanjing-video-compose。

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（若同仓库存在则执行 chanjing-credentials-guard/scripts/open_login_page.py，否则 webbrowser.open），并提示本地执行 chanjing_config.py。

审阅与安全（凭据）

与 Purpose / Credentials / Persistence 相关的逐项说明见 manifest.yaml（缺凭证时可能子进程调用 guard 的 open_login_page.py 等行为见 clientPermissions）。

Standard Workflow

主流程通常分两段，且都是异步任务：

调用 create_photo_task.py 创建文生图任务，得到 photo_unique_id
调用 poll_photo_task.py 轮询到成功，选一张 photo_path
调用 create_motion_task.py 创建图生视频任务，得到 motion_unique_id
调用 poll_motion_task.py 轮询到成功，得到最终 video_url
只有在用户明确要求保存到本地时，才调用 download_result.py

可选扩展：

若用户想做 LoRA 训练，调用 create_lora_task.py 和 poll_lora_task.py
poll_lora_task.py 成功后会返回一条 photo_task_id，可继续用 poll_photo_task.py 拿图

Covered APIs

本 Skill 当前覆盖：

POST /open/v1/aigc/photo
GET /open/v1/aigc/photo/task
GET /open/v1/aigc/photo/task/page
POST /open/v1/aigc/motion
GET /open/v1/aigc/motion/task
POST /open/v1/aigc/lora/task/create
GET /open/v1/aigc/lora/task

Scripts

脚本目录：

skills/chanjing-text-to-digital-person/scripts/

本仓库随附文件（勿与仅含 `_auth.py` 的精简包混淆）

完整包内含 _auth.py、_task_api.py（供任务脚本复用）及下列 .py CLI；请用 python3 \x3C路径>/\x3C脚本名>.py 调用（与仓库内其它蝉镜 skill 约定一致）。

文件名（仓库内）	说明
`_auth.py`	读 `credentials.json`、刷新并写回 `access_token` / `expire_in`；缺 AK/SK 时尝试 `open_login_page.py`
`_task_api.py`	任务 API 共用逻辑（由各 CLI import）
`create_photo_task.py`	创建文生图任务 → `photo_unique_id`
`get_photo_task.py`	单个文生图任务详情
`list_tasks.py`	任务列表（`type=1` photo，`type=2` motion）
`poll_photo_task.py`	轮询文生图至完成 → 默认首张图 URL
`create_motion_task.py`	创建图生视频 → `motion_unique_id`
`get_motion_task.py`	单个图生视频任务详情
`poll_motion_task.py`	轮询图生视频至完成 → 默认视频 URL
`create_lora_task.py`	创建 LoRA 训练 → `lora_id`
`get_lora_task.py`	LoRA 任务详情
`poll_lora_task.py`	轮询 LoRA 至完成 → 默认首条 `photo_task_id`
`download_result.py`	仅在需要落盘时：下载到 `outputs/text-to-digital-person/`（或 `--output`）

若环境中缺少上表任一入口或 _task_api.py，属于 分发/打包不完整。

Usage Examples

示例 1：文生图后直接图生视频

PHOTO_TASK_ID=$(python3 skills/chanjing-text-to-digital-person/scripts/create_photo_task.py \
  --age "Young adult" \
  --gender Female \
  --number-of-images 1 \
  --industry "教育培训" \
  --background "现代直播间背景" \
  --detail "短发，亲和力强，职业装" \
  --talking-pose "上半身特写，站立讲解")

PHOTO_URL=$(python3 skills/chanjing-text-to-digital-person/scripts/poll_photo_task.py \
  --unique-id "$PHOTO_TASK_ID")

MOTION_TASK_ID=$(python3 skills/chanjing-text-to-digital-person/scripts/create_motion_task.py \
  --photo-unique-id "$PHOTO_TASK_ID" \
  --photo-path "$PHOTO_URL" \
  --emotion "自然播报，语气清晰自信" \
  --gesture)

python3 skills/chanjing-text-to-digital-person/scripts/poll_motion_task.py \
  --unique-id "$MOTION_TASK_ID"

示例 2：LoRA 训练

LORA_ID=$(python3 skills/chanjing-text-to-digital-person/scripts/create_lora_task.py \
  --name "演示LoRA" \
  --photo-url https://example.com/1.jpg \
  --photo-url https://example.com/2.jpg \
  --photo-url https://example.com/3.jpg \
  --photo-url https://example.com/4.jpg \
  --photo-url https://example.com/5.jpg)

python3 skills/chanjing-text-to-digital-person/scripts/poll_lora_task.py \
  --lora-id "$LORA_ID"

Download Rule

下载是显式动作，不是默认动作：

poll_photo_task.py 和 poll_motion_task.py 成功后应先返回远端 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

Output Convention

默认本地输出目录：

outputs/text-to-digital-person/

Additional Resources

更多接口细节见：

skills/chanjing-text-to-digital-person/reference.md
skills/chanjing-text-to-digital-person/examples.md

安全使用建议

This skill appears to do what it says: call Chanjing APIs, poll tasks, and optionally download outputs. Before installing: (1) Confirm you trust the Chanjing service (open-api.chanjing.cc) because API requests and output URLs go there; (2) be aware credentials (app_id/secret_key) are stored in ~/.chanjing/credentials.json and the skill will persist access_token to that file — keep that file out of version control (manifest already notes doNotCommitToVcs); (3) when asked to download a URL, the downloader will fetch any URL you provide — only request downloads from trusted URLs; (4) the skill may open your browser or run a local credentials-guard script if credentials are missing, which is documented but worth confirming the guard script you have is the expected one; (5) avoid uploading/pointing to sensitive personal photos unless you accept the provider's privacy/usage terms. Overall the package is internally consistent and documented.

功能分析

Type: OpenClaw Skill Name: chanjing-text-to-digital-person Version: 1.0.6 The skill bundle provides a legitimate interface for the Chanjing Text-to-Digital-Person API, allowing users to generate AI portraits and videos. It manages credentials locally in `~/.chanjing/credentials.json` and communicates exclusively with documented endpoints (`open-api.chanjing.cc`). While `_auth.py` contains logic to execute a script from a sibling directory (`chanjing-credentials-guard`), this behavior is explicitly documented as a dependency for credential management. The code uses standard libraries (`urllib`, `subprocess`) for its stated purposes, and no evidence of data exfiltration, malicious execution, or prompt injection was found.

能力评估

✓ Purpose & Capability

Name/description match the included scripts and manifest: the package implements API calls for photo/motion/LoRA tasks against open-api.chanjing.cc, and the credential model (credentials.json with app_id/secret_key/access_token) is consistent with that purpose. Allowed hosts and allowed command (python3) in manifest align with functionality.

✓ Instruction Scope

SKILL.md and scripts limit actions to reading/writing the declared credentials.json, calling the documented Chanjing endpoints, polling task status, and downloading output only when explicitly requested. The only out-of-skill behaviors are opening a browser or invoking a local credentials-guard script when credentials are missing—this behavior is documented in SKILL.md/manifest.

✓ Install Mechanism

No install spec or remote downloads; the skill is delivered as scripts that run under python3. There are no third-party installers or archive extraction steps that would write or execute arbitrary code from external URLs.

✓ Credentials

No unexpected environment variables or unrelated credentials are requested. The skill uses a local credentials.json (default ~/.chanjing/credentials.json) to store app_id/secret_key and persists access_token to disk — this persistence is declared in the manifest and is proportionate to the need to use the API.

ℹ Persistence & Privilege

The skill persists access_token/expire_in back to credentials.json (persistAccessTokenOnDisk: true), and will open a browser or call a local guard script if AK/SK are missing. alwaysSkill is false and the skill does not modify other skills or global agent settings per manifest.

版本历史

v1.0.6

**Summary:** This release formalizes environment variables and credential handling, adds a machine-readable manifest, and aligns documentation to the manifest model. - Added manifest.yaml for machine-readable skill metadata and environment declarations. - Deprecated SKILL.md-specific credential/environment documentation in favor of single-source manifest.yaml. - Updated environment variable names to CHANJING_OPENAPI_CREDENTIALS_DIR and CHANJING_OPENAPI_BASE_URL (legacy names still accepted for backward compatibility). - Clarified in documentation that all credential, permission, and agentPolicy details are defined in manifest.yaml. - No changes to core script logic; scripts/_auth.py and scripts/_task_api.py are unchanged in function. - ffmpeg/ffprobe requirements remain unchanged (not required).

v1.0.5

- Added top-level skill metadata: author, binaries, env, category, and tags. - Clarified core features, runtime dependencies, and usage commands in a concise "功能说明" section. - Provided explicit environment variable documentation and standard usage instructions. - Consolidated and streamlined introductory content for easier understanding and onboarding. - No functional or API-related changes; documentation only.

v1.0.4

chanjing-text-to-digital-person 1.0.4 - Added full set of CLI scripts for all task types: create, get, poll, and download for photo, motion, and LoRA (11 new files, e.g., create_lora_task.py, poll_motion_task.py, etc.). - Expanded and clarified documentation to formally declare credentials file handling, out-of-band config paths, registry metadata, and security policy—now aligns registry and runtime behavior. - Standardized script interface: all scripts now accept explicit parameters and consistent output conventions. - Usage examples and API coverage sections updated for all available tasks and flows. - Registry metadata in SKILL.md now explicitly lists paths, sensitive fields, and credential persistence settings. - No fundamental workflow changes—improvements are in packaging, documentation, and script completeness.

v1.0.3

- Internal changes were made to scripts/_auth.py. - No changes to user-facing features or behavior. - Documentation and usage remain the same.

v1.0.2

- Added a metadata block to SKILL.md for improved discoverability and registry format. - Updated and clarified the "Preconditions" section to require use of chanjing-credentials-guard for AK/SK and Token validation before running the skill. - Expanded credential and security notes, with "安全与凭据（登记摘要）" outlining details on configuration file, API base, and token handling. - Minor refinements to descriptions and script table to align with new credential guard and documentation conventions. - General documentation improvements for clarity and compliance with updated best practices.

v1.0.1

- Improved skill isolation: authentication and config now fully handled internally, not dependent on external guard scripts. - SKILL.md updated to clarify use of CHANJING_CONFIG_DIR and CHANJING_API_BASE environment variables. - Added documentation for new scripts: chanjing-config and chanjing-get-token. - Removed requirement to run chanjing-credentials-guard before using this skill. - Minor enhancements to local credentials and token refresh logic.

v1.0.0

- Initial release of chanjing-text-to-digital-person skill. - Provides scripts to generate digital human portraits, convert them into talking videos, optionally run LoRA training, and manage tasks. - Supports polling task status for photo, video, and LoRA tasks. - Implements explicit asset downloads only when requested by the user. - Integrates with chanjing-credentials-guard for authentication.

元数据

Slug chanjing-text-to-digital-person

版本 1.0.6

许可证 MIT-0

累计安装 1

当前安装数 1

历史版本数 7

常见问题