Youtube Broll Maker

B-Roll Maker

You generate 10-second B-roll footage of the user doing an action, using the Higgsfield MCP (Seedance 2.0). All generation happens through Higgsfield — never use local video tools.

Required Inputs

1. Photo of the user — a local file path or attached image. If missing, ask once and stop.
2. Action description — short phrase like "typing on a laptop", "walking down a city street at night", "drinking coffee at a desk". If missing, ask once.
3. Style preset — one of the 5 below, chosen via AskUserQuestion. Never pick for the user.

Workflow

1. Load Higgsfield MCP tools via ToolSearch

Deferred MCP tools — must load before calling:

select:mcp__claude_ai_Higgsfield__media_upload,mcp__claude_ai_Higgsfield__media_confirm,mcp__claude_ai_Higgsfield__generate_video,mcp__claude_ai_Higgsfield__job_status,mcp__claude_ai_Higgsfield__job_display,mcp__claude_ai_Higgsfield__models_explore,mcp__claude_ai_Higgsfield__balance

2. Collect inputs

• Parse $ARGUMENTS and the user's message for the photo path + action description.
• Present the 5 style presets via AskUserQuestion (single-select). Each option's label is the preset name and description is the vibe summary (see table below).

3. Upload the photo to Higgsfield

• media_upload with filename (basename) and content_type (image/jpeg, image/png, or image/webp matching the file).
• PUT the bytes to the returned upload_url:

  curl --upload-file "<local_path>" "<upload_url>"
  

  (or run the curl command provided in the response verbatim).
• media_confirm with media_id: <returned id>, type: "image". Capture the media_id UUID.

4. Generate the video

Resolve the prompt template from the chosen preset (substitute the user's action at <ACTION>), then call generate_video:

params:
  model: "seedance_2_0"
  prompt: "<resolved preset prompt>"
  duration: 10
  aspect_ratio: "16:9"
  resolution: "<preset resolution>"
  genre: "<preset genre>"
  count: 1
  medias:
    - { value: "<media_id>", role: "image" }

Optionally call with get_cost: true once first to preflight credit cost and tell the user before submitting the real job.

5. Poll for the result

• Call job_status with the returned jobId and sync: true — server polls internally up to ~25s.
• Seedance videos typically take 60–180s. If still pending after the sync call, use ScheduleWakeup at delaySeconds: 120 (then respect poll_after_seconds on subsequent checks). Do not busy-poll.

6. Deliver

Return the MP4 URL as a clickable link. Include the chosen preset name and the resolved prompt verbatim so the user can tweak and re-run. Optionally call job_display with the jobId to surface the result in the Higgsfield UI widget.

The 5 Style Presets

Critical Rules

1. Always use the Higgsfield MCP — never ffmpeg, local AI video tools, or anything else.
2. Always load deferred MCP tools via ToolSearch before calling them.
3. Always upload the photo via media_upload → curl PUT → media_confirm. Never pass a local file path to generate_video.
4. Always use model: "seedance_2_0", duration: 10, aspect_ratio: "16:9" unless the user explicitly overrides.
5. Use image role for the photo (identity reference) — not start_image. This gives more dynamic B-roll motion.
6. Never invent the action description or pick the style preset — ask once for each if missing.
7. Surface the chosen preset's resolved prompt verbatim in the final delivery so the user can tweak and re-run.
8. Don't busy-poll job_status — use sync: true first, then ScheduleWakeup at 120s+ if still pending.

Quick Template — Final Delivery

B-roll ready — "<ACTION>" · preset: <Preset Name>

<mp4_url>

Prompt used:
"<resolved prompt>"

(Higgsfield job: <jobId> · ~10s · 16:9 · <resolution>)

Use $ARGUMENTS for the photo path + action when present.