# Youtube Broll Maker > B-Roll Maker - Skill: `rrijssenbeek/youtube-broll-maker` (Agent Skill) - Install (CLI): `npx skillmds@latest add rrijssenbeek/youtube-broll-maker` - Raw SKILL.md: https://api.skillmd.com/api/skills/rrijssenbeek/youtube-broll-maker/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Coding & Dev Tools - Author: rrijssenbeek (https://skillmd.com/u/rrijssenbeek) - Updated: 2026-09-22 - Page: https://skillmd.com/skills/rrijssenbeek/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 "" "" ``` (or run the curl command provided in the response verbatim). - `media_confirm` with `media_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 ``), then call `generate_video`: ``` params: model: "seedance_2_0" prompt: "" duration: 10 aspect_ratio: "16:9" resolution: "" genre: "" count: 1 medias: - { value: "", 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 | # | Name | Vibe | `genre` | `resolution` | Prompt template (substitute ``) | |---|------|------|---------|--------------|------------------------------------------| | 1 | **Cinematic Anamorphic** | Moody, shallow DOF, teal-and-orange film grade, slow dolly-in | `drama` | `1080p` | `Cinematic anamorphic shot, shallow depth of field, teal-and-orange color grade, soft volumetric backlight, slow dolly-in. The person . 35mm film texture, premium feature-film feel, identity preserved.` | | 2 | **Documentary Vlog** | Natural light, handheld, observational vérité | `auto` | `720p` | `Observational documentary B-roll, natural window light, handheld with subtle camera sway, true-to-life colors. The person . Candid, unstaged, authentic vlog feel, identity preserved.` | | 3 | **Tech Studio** | Dark room, RGB / neon accents, glossy desk, smooth slider | `action` | `1080p` | `Modern tech-creator studio B-roll, dark room with RGB and neon accents, glossy desk, glass and metal surfaces, smooth motorized slider move, crisp focus. The person . High-end YouTube tech-channel aesthetic, identity preserved.` | | 4 | **Golden Hour Lifestyle** | Warm sunlight, lens flare, gentle bokeh, Apple-ad clean | `auto` | `1080p` | `Golden hour lifestyle B-roll, warm low-angle sunlight through window, soft lens flare, creamy bokeh background, gentle slider movement. The person . Aspirational, clean, Apple-ad aesthetic, identity preserved.` | | 5 | **Moody Mono Film** | High-contrast black & white, 35mm grain, contemplative | `noir` | `1080p` | `High-contrast black-and-white 35mm film B-roll, deep shadows, soft key light, visible film grain, slow contemplative push-in. The person . Editorial, introspective, festival-short-film tone, identity preserved.` | ## 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 — "" · preset: Prompt used: "" (Higgsfield job: · ~10s · 16:9 · ) ``` Use `$ARGUMENTS` for the photo path + action when present.