agentleFS
yyert123yyert123/miragea-skills/ugc-video-ads/SKILL.md

ugc-video-ads

Skill0 starsChanged 2 days agoMITView on GitHub

UGC-style video ads for TikTok, Instagram Reels, YouTube Shorts and Meta ads, made by Claude Code, Codex or Cursor without a video model. The agent writes 3-5 hooks and a direct-response body (problem, demo, proof, CTA) from a product page; the script voices every line in the creator's or founder's own cloned voice, renders every hook x body combination as a captioned 1080x1920 MP4 (plus 4:5, 1:1 or 16:9), keeps captions, price overlays and the end card inside the platform safe zones, trims dead air, hits -14 LUFS, measures each hook against a 3-second budget, lints risky ad claims (guarantees, health, income, personal attributes, "free"), and writes a manifest for Ads Manager naming. Lines are cached, so a new hook re-bills only that hook. Use when the user asks for a UGC ad, a TikTok/Reels/Meta video ad, product video ad, ad creative variations, creative testing, a faceless or founder-voice product ad, or "make ads from my product photos". Voice and speech recognition by Miragea; stdlib Python + ffmpeg.

  • Reads credentials

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 0 are distinctive, and 3 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.

  • 3

    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), ffmpeg -version works, and a voice is chosen: MIRAGEA_VOICE_ID, or MIRAGEA_REF_URL + …

    line 24
  2. Very specific:

    Write ad.json next to the media (see examples/brewly.ad.json):

    line 29
  3. Very specific:

    No AI actor or lip-sync: this is a voice-first ad over your product footage.

    line 64

≈ “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

UGC video ads (voice-first, no video model)

Product photos, clips or screen recordings in; a matrix of ready-to-upload ad videos out. You (the agent) are the copywriter and creative strategist; the script is the editor: voice, timing, captions, safe zones, loudness, export and naming.

When to use

  • "Make a TikTok / Reels / Meta ad for my product" or "UGC-style ad from these photos".
  • "Give me 3 hook variations of this ad to test."
  • "Turn this product page into a 20-second video ad in my voice."
  • "Export the ad in 9:16 and 4:5 for Instagram feed."

Steps

  1. Check MIRAGEA_API_KEY is set (never print it), ffmpeg -version works, and a voice is chosen: MIRAGEA_VOICE_ID, or MIRAGEA_REF_URL + MIRAGEA_REF_TEXT. Only a voice the user owns or may use; founder-voice ads are the honest default.
  2. Gather the facts: product name, price, offer, who it is for, the one problem it solves, real proof (reviews, numbers the user can back up). Never invent a testimonial, statistic or result.
  3. Write ad.json next to the media (see examples/brewly.ad.json):
    • hooks: 3-5, each under ~3 s spoken (7-8 words). Mix angles: price shock, POV, call-out ("Stop paying..."), demo-first, curiosity. Optional overlay (on-screen text, <= 28 chars) and media.
    • bodies: one or more {id, lines}; 2-4 lines: problem or setup -> demo -> proof/benefit. Short spoken sentences; write numbers the way they should be heard if it matters.
    • cta: one spoken line (what to tap + why now). end_card: title, up to 3 lines, button, seconds, media.
    • platform: tiktok | reels | shorts | meta | any (length advice). label: e.g. "AI voice" when the voice is synthetic or first-person lines are not the speaker's own words.
    • media: image (slow push-in) or clip (cover-cropped, looped/trimmed). Omit it for a gradient.
  4. python scripts/ads.py check ad.json (free). ! must be fixed (missing media, over the platform's max length). ~ is advice: long hooks, length outside the usual sweet spot, risky claims, missing CTA/end card, testimonial disclosure. Rewrite and re-check until clean or each ~ is a conscious choice.
  5. python scripts/ads.py render ad.json --dry-run (free) shows the variant matrix, estimated lengths and how many unique lines will be voiced.
  6. Render: python scripts/ads.py render ad.json -o ads_out --ratios 9:16,4:5 Options: --only price-demo (or a hook/body id), --max N, --music bed.mp3 (ducked), --gap 0.12 (pace), --hook-budget 3, --lowercase, --no-captions.
  7. Report: the files, their lengths, any hook over budget (hooks_over_budget), and the advice you kept. Suggest testing all hooks with the same body first, then the winning hook with new bodies.

Output (in ads_out/)

  • <brand>_<hook>-<body>_<9x16|4x5|1x1|16x9>.mp4: H.264/AAC, 30 fps, -14 LUFS, captions burned in.
  • <brand>_<hook>-<body>.srt: the captions (upload as a separate track where supported).
  • manifest.csv / manifest.json: file, variant, hook, body, ratio, seconds, measured hook seconds, over-budget flag, plus the check advice.

How it works

  • Each line is voiced once with Miragea and cached by text + voice; edge silence is trimmed; lines are joined with a short UGC-pace gap, so every scene change lands on a line change.
  • Captions use your script's exact spelling (brand names stay right) placed on Miragea speech-recognition timing, 1-3 words per caption, breaking at commas and sentence ends.
  • Safe zones (9:16): captions centred at ~60% height and shifted left of the like/share rail; overlays below the top bar; end card text inside the middle band.

Honest limits

  • No AI actor or lip-sync: this is a voice-first ad over your product footage. Pair it with real creator clips when you have them (put them in media).
  • The claim lint is a checklist, not legal review. Platform policies change; read the current ad policy.
  • Speech recognition and voice are cloud-based (Miragea); only the generated voice lines are uploaded.

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.