freebeatfit-ai（freebeatfit AI 搜索）

Freebeat MCP Workflows

Use this skill for reliable Freebeat MCP task execution with correct tool order and async handling.

Quick Start

1. Confirm the user goal:

- generate_effect (template-based effect generation) - generate_music_video (MV generation from uploaded audio)

1. Gather required inputs.
2. Execute the correct workflow.
3. Poll task status until terminal state.
4. Fetch result only after completion.

Tool Map

• upload_audio: Upload local audio or import from URL. Returns music_id.
• upload_image: Upload one or two images. Returns image_urls.
• list_effects: Discover templates and defaults for effect generation.
• generate_effect: Start async effect generation task.
• generate_music_video: Start async MV generation task.
• get_task_status: Poll async task state.
• get_task_result: Fetch output only when status is completed.

Workflow A: Generate Effect

Preferred order:

1. Call list_effects.
2. Select effect_id and collect defaults:

- default_music_id - image_url (default image)

1. Optional overrides:

- If user provides custom music, call upload_audio and use returned music_id. - If user provides custom image, call upload_image with one image and use image_urls[0].

1. Call generate_effect with:

- effect_id (required) - music_id (required) - prompt (required, trimmed length 1..2000) - reference_image_urls (required, exactly one URL) - watermark (optional, default false)

1. Poll with get_task_status until completed or failed.
2. On completed, call get_task_result and return video_url and cover_url.

Workflow B: Generate Music Video

Preferred order:

1. Call upload_audio first and capture music_id.
2. Optional image references:

- Call upload_image and pass returned image_urls to generate_music_video.reference_image_urls.

1. Call generate_music_video with:

- music_id (required) - prompt (required, trimmed length 1..2000) - Optional: mv_type, style, aspect_ratio, resolution, watermark, start_ms, end_ms, reference_image_urls

1. Poll with get_task_status until terminal state.
2. On completed, call get_task_result and return video_url and cover_url.

Async Handling Rules

• Treat task submission as asynchronous.
• Continue polling while status indicates in-progress work (for example pending).
• Status values may be lowercase (pending, completed, failed) and some older systems may use uppercase variants.
• Do not call get_task_result before status is completed.
• If get_task_result returns TASK_NOT_COMPLETED, continue polling.
• On failed, report error_message, error_code, and any backend message from status response.

Input Validation Checklist

Before generation calls, verify:

• Prompt is present and trimmed length is within 1..2000.
• For generate_effect, reference_image_urls contains exactly one image URL.
• For generate_effect, music_id is either default_music_id from list_effects or from upload_audio.
• For generate_music_video, music_id comes from upload_audio.
• For upload_audio, provide exactly one of file_path or url.

Response Format Guidelines

When reporting progress or final outputs, include:

• Task type (effect_generation or music_video_generation)
• task_id
• Current status
• On completion: video_url and cover_url
• On failure: failure code/message and next fix step

Defaults and Practical Guidance

• Use list_effects defaults unless the user explicitly asks to override.
• For effect generation, default image is list_effects.image_url.
• For MV generation, default values are acceptable unless user specifies otherwise.
• Keep polling cadence steady and avoid tight loops.

Example Decision Logic

If user asks "make an effect from template":

• Use Workflow A.

If user asks "make a music video from this song":

• Use Workflow B.

If user provides only a prompt without media:

• For effect: use list_effects defaults.
• For MV: request audio input (or URL) and run upload_audio first.