AI Subtitle Remover

550W AI subtitle and watermark removal

Use the highest-level matching tool. The global edition defaults to English; its account setup uses Eraser, while media operations use the shared 550W Open API. Respond in the user's language when possible; tool messages fall back to English except for Chinese. Language never changes the service region.

Credentials and data

• If credentials are missing, direct the user to the localized Eraser API page at https://eraser.550wai.com/{locale}/api/; use https://eraser.550wai.com/api/ for English or unsupported locales. They need the User ID and API Key from the same account.
• In WorkBuddy, ask the user to fill in both fields in the connector settings. Otherwise use SUBTITLE_REMOVER_USER_NO and SUBTITLE_REMOVER_API_KEY, or the configureCredentials action. Never repeat a complete API Key.
• Before the first media operation, explain that the selected file or public link is sent to the 550W Open API. Use only a file the user explicitly selected or named. Privacy and terms: https://eraser.550wai.com/privacy/ and https://eraser.550wai.com/terms/; localized pages are available.

Capabilities

• Remove hardcoded subtitles from a local MP4/MOV video or a public direct video-file URL: remove_video_subtitles (CLI action workflow). The tool derives local-video metadata. removeAudio is only an optional parameter when explicitly requested.
• Remove a short-video watermark: ask the user to copy the public video share link inside the platform app and provide that link, or share text containing one link. TikTok and other service-supported platforms are examples. Call remove_video_watermark (CLI action removeVideoWatermark) with the share link; do not download the video first. Use a stable operationId when available.
• Remove watermarks or unwanted text from a user-selected local image: remove_image_watermark (CLI action removeImageWatermark). Process multiple images serially.
• Check subtitle or image task status, subtitle task history, and credit balance: get_subtitle_task, get_image_task, list_subtitle_tasks, query_credits.
• Delete an owned subtitle task: delete_subtitle_task. Explain that this schedules media cleanup and does not refund credits; confirm immediately before deletion.
• Ask one question if the input or intended billed operation is ambiguous.

If the host has no registered tools, send one JSON request on standard input to node {baseDir}/dist/550w-skill.cjs. Use an absolute params.filePath for a local file and parse the single JSON response.

Processing and billing

• Subtitle removal uses full-frame erasing. Never request or send erase coordinates. For a public direct video URL, ffprobe obtains actual width, height, and duration unless all three are supplied. Reject private-network or credential-bearing URLs; never invent metadata.
• Use the same subUserId for a child user's submission, task lookup, list filtering, and deletion. The subtitle workflow polls for up to 10 minutes; on timeout or repeated query failure, return the task ID for later lookup.
• Image removal is synchronous by default. For processing, preserve the task ID and query it later. Only expose a result URL on success; explain failed and expired states.
• Subtitle removal is billed by duration and resolution. Share-link video watermark removal costs one credit per successful result; image removal costs 10 credits per successful image. State the unit before a batch and include only selected items.
• A timeout does not prove submission failed. Query the task first; never blindly retry a billed operation. Reuse an idempotencyKey only with identical subtitle inputs. Report the service's refundStatus on subtitle failure without assuming a refund.
• Return the result or actionable error to the user in every case.

Read API contract for limits, statuses, billing formulas, and errors. Read agent compatibility for installation outside WorkBuddy.