Kre Video Translator｜视频翻译 Skill

AI video translation and subtitle export / AI 视频翻译

Kre Video Translator

Use the bundled CLI with the production API https://kretrans.com/api. Set SKILL_DIR to this installed skill directory and replace <locale> with the conversation locale in every command:

• Product and registration: https://kretrans.com/<locale>/

• Agent Skill overview: https://kretrans.com/<locale>/video-translation-agent-skill/

• ClawHub listing: https://clawhub.ai/skz7s/skills/kre-video-translator

• Chinese conversation: --locale zh

• Other languages: --locale en

Access

Check local access:

bash

python3 "$SKILL_DIR/scripts/account.py" --locale <locale> auth detect

When no Key is configured, show the user:

当前还没有配置 Kre Video Translator API Key。

请先完成以下步骤：

1. 注册或登录 Kre Video Translator
2. 创建 API Key
3. 确认 Key 包含：account:read、task:write、task:read

注册地址：注册或登录

创建 Key：创建 API Key

完成后请回复“已创建”。

Use the English text and /en/ URLs for other conversations. After the user confirms, run:

English onboarding text:

Kre Video Translator API key is not configured yet. Please register or sign in, create a Key with account:read, task:write, and task:read, then use Registration and API Keys. Reply “Created” when finished.

bash

python3 "$SKILL_DIR/scripts/account.py" --locale <locale> auth launch

This opens a visible terminal on macOS, Windows, or a Linux desktop and starts Key-only setup. It does not open a browser. The user pastes the Key into the hidden prompt and presses Enter. An existing Key is verified by pressing Enter; a pasted Key replaces it. The terminal setup asks only for the Key, verification, and local save.

Read the secret-free result_file returned by auth launch. Continue after its status is verified or account_verified. A manual_required result includes an absolute command for the user's terminal. auth verify checks an existing configuration. Account, sign-in, billing, redemption, and Key management use the official production console.

Translate

Collect the source, source language, target languages, and any user-provided model, prompt, or transcript reference. The task title defaults to the source filename, including its extension; pass --title only when the user explicitly requests a custom title. Submit with the frontend-compatible fields:

bash

python3 "$SKILL_DIR/scripts/translate.py" submit INPUT_OR_URL \
  --locale <locale> --source-language en \
  --target-language zh --state-file "/path/task.state.json" \
  --confirm-upload

Repeat --target-language for multiple targets. Add supplied values with --translation-model, --custom-prompt, and --source-subtitle (SRT/TXT, up to 1 MiB). The CLI prepares the upload as mono MP3. It checks FFmpeg and Python dependencies without installing them by default. For a URL, it checks yt-dlp only when the URL workflow starts:

bash

ffmpeg -y -i INPUT \
  -map 0:a:0 -vn -sn -dn \
  -b:a 64k -acodec mp3 -ar 44100 -ac 1 OUTPUT.mp3

The multipart request contains media, optional source_subtitle, title, source_language, target_languages, translation_model, custom_prompt, original_filename, and source_is_video, plus Idempotency-Key. Use the original filename, including its extension, as title unless the user explicitly asks for a custom title. For a URL, the CLI downloads a readable filename from the URL metadata. Save the returned task_id and idempotency_key; reuse the same idempotency key when submission outcome is uncertain. The source audio, filename, language settings, and supplied prompt/reference are uploaded to the production service.

If KRE_VIDEO_TRANSLATOR_ERROR_JSON.next_action is ask_user_to_approve_tool_install, tell the user which local tool is missing and ask whether it may be installed. After approval, rerun the same command with --allow-install-tools. This flag installs only tools required by that invocation. A local file never requires yt-dlp; URL input may require it. If the user declines, provide the install_command from the error JSON and stop.

Progress

Submission and processing are asynchronous. Submit returns immediately. Resume and save the final result:

bash

python3 "$SKILL_DIR/scripts/translate.py" resume --locale <locale> \
  --task-id TASK_ID --state-file "/path/task.state.json" \
  --result-json "/path/task.result.json"

Status checks start at 20 seconds, back off to 120 seconds, and follow HTTP 429 Retry-After. Relay changed stage and percentage values from KRE_VIDEO_TRANSLATOR_PROGRESS_JSON; combine unchanged intervals into one user update. A timeout preserves the remote task and its resume command.

Errors

Read KRE_VIDEO_TRANSLATOR_ERROR_JSON and route errors as follows:

• API unavailable: for connection failures, HTTP 502/503/504, or service-unavailable errors, tell the user the production API is temporarily unavailable. Check whether the installed Skill is current. If it was installed from ClawHub, run clawhub update kre-video-translator when an update is available, then retry the same command once. Reuse the existing task ID and idempotency key when available. Do not treat HTTP 429 as an outage; follow the CLI backoff and Retry-After.
• Insufficient balance or quota: for insufficient_quota, insufficient_balance, quota_exceeded, insufficient_entitlement, no_available_minutes, or equivalent messages, do not retry or update the Skill. Tell the user to check balance and packages at https://kretrans.com/<locale>/app/account/packages, then add or redeem minutes.
• Invalid or insufficient API Key permissions: guide the user to create or replace a Key containing account:read, task:write, and task:read. Never request the Key in chat.
• Local FFmpeg/export failure: preserve the completed JSON/SRT result and report the concrete FFmpeg/libass action. Do not claim video export success unless the encode succeeds.

Review And Export

The completed result contains result.cc_list with stable segment IDs, cue timing, source text, each target translation, and optional word timestamps. Review translation completeness, meaning, cue timing, and line breaks. Use word timestamps to make local translation or line-break corrections when the evidence is clear, preserving segment IDs and timing structure in the reviewed JSON.

Ask whether the user wants subtitle files or a video with subtitles.

Subtitle files:

bash

python3 "$SKILL_DIR/scripts/translate.py" export --locale <locale> \
  --result-json "/path/task.reviewed.json" --format srt \
  --output "/path/subtitles/"

Burned-in video:

bash

python3 "$SKILL_DIR/scripts/translate.py" export ORIGINAL_VIDEO --locale <locale> \
  --result-json "/path/task.reviewed.json" --format video \
  --language zh --output "/path/video.zh.mp4"

Video export checks for FFmpeg's subtitles/libass filter before encoding. On macOS it can use Homebrew's ffmpeg-full build when the standard build lacks that filter. Other platforms need an FFmpeg build with libass enabled. When the capability is unavailable, report that the subtitle translation is complete, provide the subtitle-file result, and give the platform-specific FFmpeg installation action. The CLI removes temporary subtitle files and incomplete video outputs after a failed encode.

If video export reports a missing subtitles/libass capability, ask before installing a compatible FFmpeg build and rerun export with --allow-install-tools. The reviewed subtitle file remains available while video export is waiting for approval.

SRT export is local. Video export requires the original local video, one target language, and a successful FFmpeg encode. Return artifact paths only after the export command succeeds. Keep the API Key in the local credential store and result files free of credentials.