Video Translation Skill

One-click video translation powered by Alibaba Cloud IMS, supporting subtitle-level and speech-level translation.

Input Format Requirements

IMPORTANT: Different APIs use different address formats!

API Address Format Reference

Key: Subtitle extraction uses oss://, video translation uses HTTP URL!

User Input Handling

Format Conversion Rules

oss:// format ⇄ HTTP URL format

oss://my-bucket/videos/test.mp4
    ⇄
https://my-bucket.oss-cn-shanghai.aliyuncs.com/videos/test.mp4

Conversion Formula:

• oss://\x3Cbucket>/\x3Cpath> → https://\x3Cbucket>.oss-\x3Cregion>.aliyuncs.com/\x3Cpath>
• HTTP URL does not require signing, use Bucket domain format directly

Local Video Processing Flow

User provides local video path
    │
    ├─ AskUserQuestion: "Please provide OSS upload path (format: oss://\x3Cbucket>/\x3Cpath>/\x3Cfilename>.mp4)"
    │
    ├─ User specifies upload path
    │   ├─ Check if Bucket exists
    │   ├─ Upload file: aliyun oss cp \x3Clocal_path> \x3Coss_path>
    │   ├─ Save oss:// format → for subtitle extraction
    │   └─ Save HTTP URL format → for video translation
    │
    └─ User does not specify path → STOP, user MUST provide upload path

Upload Command:

aliyun oss cp \x3Clocal_path> oss://\x3Cbucket>/\x3Cpath>/\x3Cfilename>.mp4

Save both formats after upload:

Local: /Users/demo/videos/test.mp4
Uploaded to: oss://my-bucket/videos/test.mp4
    ├─ oss:// format: oss://my-bucket/videos/test.mp4 (for subtitle extraction)
    └─ HTTP URL: https://my-bucket.oss-cn-shanghai.aliyuncs.com/videos/test.mp4 (for video translation)

Execution Gate Checklist

Strict Requirement: Agent MUST execute in phase order, cannot proceed without passing current phase!

Phase 0: Environment and Credential Check (HARD-GATE)

HARD-GATE: Cannot proceed with any subsequent operations without passing!

Phase 1: Translation Mode Confirmation (BLOCKING)

AskUserQuestion: "Do you need subtitle translation (translate subtitles only) or speech translation (translate subtitles + replace voiceover)?"

┌─ Subtitle translation → NeedSpeechTranslate: false
└─ Speech translation → NeedSpeechTranslate: true

⚠️ No reply received → STOP, cannot proceed!

DO NOT infer translation mode from input type!

Phase 2: Subtitle Processing Confirmation (BLOCKING)

AskUserQuestion: "Do you need to erase original subtitles from the video? Do you need to burn-in translated subtitles?"

⚠️ No reply received → STOP, cannot proceed!

Parameter Mapping:

Phase 3: Output Path Confirmation (Non-blocking)

Default Output Rules:

• Bucket: Same bucket as input video
• Directory: Same directory as input video
• Filename: {source}_translated_{random8}.mp4
• Example: oss://bucket/videos/demo.mp4 → oss://bucket/videos/demo_translated_a1b2c3d4.mp4

DO NOT use shell variables, use Python: python3 -c "import random; print(''.join(random.choices('abcdefghijkmnpqrstuvwxyz23456789', k=8)))"

Phase 4: Subtitle Review Confirmation (Conditional Blocking)

CRITICAL: After subtitle extraction, MUST output content as-is for user review, DO NOT change format!

Scenario Entry Selector

Key Points:

1. When user inputs local video, MUST first upload to OSS and get HTTP URL
2. When user does not provide subtitle, MUST ask if subtitle extraction and review is needed

User inputs video
    │
    ├─ Local video?
    │   └─ Yes → AskUserQuestion: "Please provide OSS upload path"
    │       ├─ User provides path → Upload to OSS → Convert to HTTP URL → Continue
    │       └─ User does not provide → STOP
    │
    ├─ oss:// format?
    │   └─ Yes → Inform user to convert to HTTP URL format
    │
    └─ HTTP URL format? → Continue
        │
        ├─ User provides SRT file?
        │   ├─ Yes → Input type = with_subtitle
        │   │   ├─ Translation mode = speech → 【Scenario 4】 ⚠️ MUST ask CustomSrtType
        │   │   └─ Translation mode = subtitle → 【Scenario 3】
        │   │
        │   └─ No → Input type = only_video ⚠️ MUST ask if review needed
        │       │
        │       ├─ AskUserQuestion: "Do you need to extract subtitles for review first, or translate directly?"
        │       │
        │       ├─ Need review → 【Scenario 2】 ⚠️ Phase 4 blocking
        │       │
        │       └─ Direct translation → 【Scenario 1】 (TextSource=OCR_ASR)

Scenario 0 (Local video) detailed flow:

1. AskUserQuestion: "Please provide OSS upload path (format: oss://\x3Cbucket>/\x3Cpath>/\x3Cfilename>.mp4)"
2. After user specifies path, execute aliyun oss cp \x3Clocal_path> \x3Coss_path>
3. Convert to HTTP URL: https://\x3Cbucket>.oss-\x3Cregion>.aliyuncs.com/\x3Cpath>/\x3Cfilename>.mp4
4. Continue with subsequent scenario flow

Scenario 2 detailed flow:

1. Ask for subtitle detection region (roi parameter)
2. Call CaptionExtraction to extract subtitles, input and output use oss:// format
3. Output subtitle content as-is for user review
4. After user confirmation, use reviewed SRT to submit translation

Parameter Decision Table

Decision Rules: Clearly define handling for each parameter, DO NOT assume arbitrarily!

TextSource Scenario Mapping:

CustomSrtType Trigger Rules:

Failure Protection Mechanism

HARD-GATE: After speech translation fails, DO NOT auto-switch to subtitle translation!

API Error Handling

SRT Format Repair Flow

Detect empty subtitle entries → Delete empty entries → Renumber → Upload repaired file → Inform user

See troubleshooting.md for details.

CLI Command Templates

IMPORTANT: Before submitting API, MUST reference api-parameters.md to confirm parameter format!

See cli-commands.md for details.

Core Commands:

# Register media asset
aliyun ice register-media-info --input-url "oss://\x3Cbucket>/\x3Cobject>" --media-type video --user-agent AlibabaCloud-Agent-Skills

# Submit subtitle extraction (use OSS path)
aliyun ice submit-iproduction-job \
  --function-name CaptionExtraction \
  --input "Media=oss://\x3Cbucket>/\x3Cobject> Type=OSS" \
  --biz-output "Media=oss://\x3Cbucket>/\x3Coutput>.srt Type=OSS" \
  --job-params '{"lang":"ch","roi":[[0.5,1],[0,1]]}' \
  --force \
  --user-agent AlibabaCloud-Agent-Skills

# Submit video translation
aliyun ice submit-video-translation-job \
  --user-agent AlibabaCloud-Agent-Skills

CLI Format Key Points:

• Subtitle extraction uses command name submit-iproduction-job (lowercase, - separator)
• --input and --biz-output format: space-separated string "Media=... Type=OSS", NOT JSON
• --job-params format: JSON string
• MUST add --force to skip plugin parameter validation
• All ICE commands MUST add --user-agent AlibabaCloud-Agent-Skills

Documentation Reference

Key Constraints

• Before submitting API, MUST reference api-parameters.md to confirm parameter format
• All ICE CLI commands MUST add --user-agent AlibabaCloud-Agent-Skills
• Subtitle extraction (SubmitIProductionJob) uses oss:// format
• Video translation (SubmitVideoTranslationJob) uses HTTP URL format, no signing needed
• Local videos MUST first be uploaded to OSS, user MUST provide upload path
• NeedFaceTranslate MUST be false
• SpeechTranslate and SubtitleTranslate are mutually exclusive
• InputConfig.Subtitle MUST use HTTPS format, DO NOT use oss://
• Speech translation + SRT input requires SpeechTranslate.CustomSrtType
• DO NOT infer translation mode from input type

Task Polling

Mandatory: MUST continuously poll task status until completion (State=Finished) or failure (State=Failed), DO NOT exit early!

Polling Logic:

Loop polling until:
  - State == "Finished" → Return result
  - State == "Failed" → Report error
  - Exceeds 30 minutes → Report TimeoutError

Prohibited: Return after single query / Skip polling and return JobId directly

Time Reference (3-minute video):

• Subtitle-level translation: 3-5 minutes
• Speech-level translation: 10-20 minutes

Result Retrieval

# Get media asset info
aliyun ice get-media-info --media-id "\x3CMediaId>"

# Generate signed URL (for private Bucket)
aliyun oss sign oss://\x3Cbucket>/\x3Cobject> --timeout 3600

End of Document