agentleFS
yyert123yyert123/miragea-skills/auto-broll/SKILL.md

auto-broll

Skill0 starsChanged 2 days agoMITView on GitHub

Automatic B-roll for talking-head videos, podcasts, tutorials and faceless YouTube/TikTok videos, for Claude Code, Codex and Cursor. The agent reads a timestamped transcript and picks a cutaway for each concrete moment (things, actions, numbers, places); the script lands every cut on the exact spoken word, keeps the speaker's audio untouched, covers the frame (16:9 or 9:16), adds a slow Ken Burns push on stills, warns about flash cuts, covered hooks and too much B-roll, and exports the MP4 plus a CMX 3600 EDL for Premiere Pro, DaVinci Resolve and Avid. Uses your own clips and images, or free Pexels stock. Faceless mode voices a script in your own cloned voice first, then covers it end to end with B-roll. No extra LLM key. Use when the user asks to add B-roll, cutaways or stock footage to a video, make a talking head less boring, find B-roll for a script, auto-edit a podcast clip, or make a faceless video with B-roll and voiceover. Speech recognition and voice by Miragea; stdlib Python + ffmpeg.

  • Reads credentials

The file

We split this skill into the rules it gives an agent

2 rules. About 0 are common, 0 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Check MIRAGEA_API_KEY is set (never print it) and ffmpeg -version works.

    line 24
  2. Very specific:

    Check for free: python scripts/broll.py check talk.broll.json !

    line 42

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Auto B-roll (cut on the word)

A video (or a script) in, the same video with B-roll cut in on the words that call for it. You (the agent) are the editor: you read each line with its word timings and decide what to show. The script does the frame-accurate work: word anchoring, cover-cropping, timing rules, mixing, export.

When to use

  • "Add B-roll to my talking-head video / podcast clip / course lesson."
  • "Find cutaways for this script and drop them on the right words."
  • "Make a faceless video from this script with my voice and B-roll."
  • "Give me an EDL of the B-roll so I can finish in Premiere / Resolve."

Steps

  1. Check MIRAGEA_API_KEY is set (never print it) and ffmpeg -version works.
  2. Prepare the plan:
    • Talking head / podcast / screen recording: python scripts/broll.py prep talk.mp4 --library broll/
    • Faceless (no footage of a speaker): python scripts/broll.py prep --script script.txt --library broll/ --vertical voices the script in the creator's Miragea voice (MIRAGEA_VOICE_ID, or MIRAGEA_REF_URL + MIRAGEA_REF_TEXT; only a voice they own or may use) and writes script.narration.wav. Either way it writes <name>.broll.json: sentence lines with per-word start times, and the library of the user's clips (with lengths). The transcript is cached and shared with the video-captions, video-to-shorts and video-dubbing skills, so it is billed once.
  3. Plan the cutaways: add one object per shot to "cutaways": {"line": 3, "on": "espresso machine", "clip": "broll/espresso.mp4", "dur": 2.5}
    • on: the word or phrase the cut lands on; it must be in that line (numbers match either way: "ninety three" = "93"). clip: a library file, any path, or "pexels:<search words>". dur optional (default: until the line ends, 1.2-4 s). clip_start: in-point inside a clip.
    • Show what is said: concrete nouns, actions, numbers, places, products, screens. Do not illustrate opinions or filler. Open the clips' first frames if unsure what they show.
    • Talking head: cover 20-40% of the runtime, never the first 2 s (the hook), keep at least 1 s of speaker between shots. Faceless: a new shot every 2-4 s; each runs until the next one.
  4. Check for free: python scripts/broll.py check talk.broll.json ! = must fix (word not found, missing clip, stock not fetched). ~ = advice (flash cut, covered hook, back-to-back cutaways, clip shorter than its slot). Re-run until it exits 0.
  5. Stock (optional): python scripts/broll.py fetch talk.broll.json downloads each pexels: query at the right orientation (free key from pexels.com/api) and records credits in broll_stock/CREDITS.txt.
  6. Render: python scripts/broll.py render talk.mp4 talk.broll.json -o broll_out Options: --fade 0.12 (soft edges instead of a hard cut), --music bed.mp3 (ducked under the voice), --lufs -14 (talking head audio is left untouched unless set; faceless defaults to -14), --vertical (faceless 1080x1920), --skip-bad, --dry-run.
  7. Report: the MP4, the EDL, the number of cutaways, coverage, and any ~ advice you chose to keep.

Output (in broll_out/)

  • talk.broll.mp4: the original picture with cutaways on their words; the speaker's audio is the original track, never re-timed.
  • talk.broll.edl: CMX 3600 EDL of the cutaways at their timeline positions (File > Import in Premiere Pro / Resolve), with the source clip name and anchor word as comments.
  • talk.broll.manifest.json: every cutaway's start, length, word, clip and warnings.

Honest limits

  • The agent chooses footage from what it can see (file names, first frames, your descriptions); it does not generate video. Pair it with your own clips, screen captures or a stock library.
  • One B-roll layer, full frame. No picture-in-picture or split screen yet.
  • Speech recognition is cloud-based (Miragea); only a 16 kHz mono audio track is uploaded, never the video. Stock clips come from Pexels under the Pexels license.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.