Chanjing Digital Human

Chanjing Platform for FrameVideo

When To Use

Use this skill for:

• 数字人视频 — 生成蝉镜平台数字人主持视频
• AI Creation 资产 — 使用蝉镜 AIGC 生成图片/视频素材
• 平台音乐/音效 — 下载蝉镜平台背景音乐和音效
• OAuth 认证 — 检查/设置蝉镜平台认证
• 资产集成 — 将蝉镜生成的资产接入 FrameVideo composition

Do NOT Use

Avoid this skill for:

• 本地 TTS — 使用 framevideo-media (Kokoro TTS)
• 本地音频处理 — 使用 framevideo-media (transcribe, remove-background)
• Composition HTML 编写 — 使用 framevideo
• SSML 语音标记 — 使用 framevideo-voiceover-ssml

Quick Start

场景 1: 生成数字人视频

# 1. 检查认证
npx framevideo auth status

# 2. 列出可用数字人和声音
npx framevideo chanjing people list --json
npx framevideo chanjing voices list --json

# 3. 创建数字人视频项目
# (详见 references/website-project-payload.md)

# 4. 下载生成的视频
# 下载到 assets/digital-humans/

场景 2: 下载平台音乐

# 1. 认证
npx framevideo auth login

# 2. 浏览音乐分类
npx framevideo chanjing music categories --json

# 3. 列出音乐
npx framevideo chanjing music list --category <id> --compact

# 4. 下载
npx framevideo chanjing music download --id <id> --output assets/music/bg.mp3

场景 3: AI Creation 生图/生视频

# 使用 AI Creation API 生成图片或视频
# (详见 references/ai-creation.md)

Routing

Read only what the request needs:

For local-only Kokoro TTS, Whisper transcription, or background removal, use framevideo-media instead.

Platform Rules

• Use OAuth Bearer tokens from the shared Chanjing credential store. Never print, echo, save, or commit tokens.
• Prefer the existing FrameVideo CLI, Studio routes, and ChanjingOpenApiClient in packages/cli/src/tts/chanjingOpenapi.ts; do not copy standalone credential scripts.
• Do not use app_id, secret_key, CHANJING_OPENAPI_ACCESS_TOKEN, or dp_open_app data for agent-authored workflows.
• If OAuth is missing and preview/Studio is available, prefer the Studio login UX instead of stopping at env-var instructions.
• Download every generated or selected platform asset before referencing it in composition HTML. Do not use render-time remote Chanjing/OSS URLs.

Asset Locations

Use these local project paths:

Digital-Human Workflow

1. Check auth. If missing, use Studio login flow from studio-routes.md.
2. Resolve person_id and voice_id.
3. Fetch the selected person's real metadata before layout.
4. Build a website-project payload from website-project-payload.md.
5. Save the website project and verify project_id.
6. Submit video synthesis and verify task_id.
7. Poll using api-enums.md.
8. Download the finished video to assets/digital-humans/.
9. Wire it into FrameVideo using composition-wiring.md.

ID Sources

Do not invent ids or reuse fixture ids. If a user-provided person id cannot be fetched, report that error instead of falling back to another person.

AI Creation Workflow

Use this for non-presenter AIGC images/videos such as B-roll, generated backgrounds, concept images, product display clips, transition clips, or abstract visuals.

1. Read ai-creation.md.
2. Fetch active models with framevideo chanjing ai-models ...; do not read backend constants or copied model lists.
3. Build payloads from the selected model's params_config.fields; do not hard-code model options.
4. Create .framevideo/ai-creation-tasks/*.json task metadata before submission.
5. Apply local and remote idempotency checks before submitting billable tasks.
6. Submit only missing tasks.
7. Use short sync runs to poll/download.
8. Compose only with local files under assets/ai-creation/images/ or assets/ai-creation/videos/.

Music and SFX Workflow

Use Chanjing platform music/SFX when the user asks for platform BGM, soundtrack, chorus extraction, whooshes, hits, clicks, or transition sounds.

npx framevideo chanjing music categories --json
npx framevideo chanjing music list --compact
npx framevideo chanjing music download --id <music-id> --chorus --duration 10 --json

npx framevideo chanjing sound-effect list --compact
npx framevideo chanjing sfx download --id <effect-id> --volume 0.8 --json

Keep BGM under assets/music/ and SFX under assets/sfx/. Use low BGM volume under speech or digital-human narration.

Composition Wiring

Read composition-wiring.md before writing HTML that uses Chanjing assets.

Core rules:

• Visual video clips are muted.
• Add a separate <audio> clip when the generated video contains narration.
• Keep captions clear of faces, hands, presenter body, CTA, product UI, and platform subtitles.
• For styled FrameVideo captions, prefer synthesis timing fields when available; otherwise use framevideo-media to transcribe the downloaded local video/audio.
• For transparent presenter overlays, use framevideo-media remove-background after download.

Hard Stops

• Do not pass a digital-human id as a website projectId.
• Do not hand-author workspace_v2.
• Do not call compatibility stubs such as direct digital-human generation routes that return 501.
• Do not insert remote output_url or Chanjing/OSS URLs into composition HTML.
• Do not create duplicate AI Creation tasks when matching local or recent remote task metadata exists.

Validation

After wiring Chanjing assets into a FrameVideo project:

npx framevideo lint
npx framevideo validate
npx framevideo inspect

For visual overlap, caption safety, presenter placement, and B-roll readability, use framevideo-visual-qa.