agentleFS
iart-aiiart-ai/ad-video-skills/skills/ad-creative-video/SKILL.md

ad-creative-video

Skill22 starsChanged 5 months agoMITView on GitHub

This skill should be used when the user asks to "make a video ad", "create an animated ad", "build a performance/UGC-style ad", "batch-produce ad creative variations", "generate ad variants for A/B testing", "swap headline/offer/CTA across many ad versions", or "export one ad in multiple aspect ratios for Meta/TikTok/Reels". Covers ad hook structures, hook→CTA message-match, data-driven variant generation (1 template × CSV = N ads), multi-aspect export, and platform specs.

1 other project uses a version of this.

Adopted by

Most recently taken up by

  • Other versions1 project uses a version of this.

    Word for word · 1

    1. willlee1995/skills★ 1100% same

    Edited · 0

    None found.

    Wrote their own version · 0

    None found.

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

  • In this repository2 other files it gives its agents.

The file

We split this skill into the rules it gives an agent

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

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    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. Distinctive:

    Not for sale/discount countdown promos, and not for testimonial/review-driven ads — those are different structures.

    line 17
  2. Distinctive:

    Keep a column of 5–10 hook strings and let the batch render produce one ad per hook against the same body — that is the cleanest hook test …

    line 46
  3. Very specific:

    Keep critical content out of the 9:16 bottom third — that is where the platform stacks captions, the CTA button, and the username.

    line 109
  4. Very specific:

    Every animated value is a pure function of useCurrentFrame() — no CSS/library timers.

    line 115

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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

Ad Creative Video

Build ONE motion-graphics ad template, then mass-produce variants from a data table — different hooks, offers, CTAs, and products — so a performance marketer can A/B test dozens of versions across placements and aspect ratios. The craft is making the template fully data-driven and message-matched, so 40 variants stay on-brand and only the tested variable changes.

When to use

  • Performance/UGC-style video ads where the first 1.5–3s must stop the scroll.
  • Batch variant generation: one template rendered against many rows (headline/offer/CTA/product swaps) for A/B testing.
  • Multi-aspect export of the same ad for Meta Feed, Reels/Stories, TikTok, YouTube.

Not for sale/discount countdown promos, and not for testimonial/review-driven ads — those are different structures. Stay on the batch-variation, test-everything angle.

The two rules that make variants worth running

  1. Isolate one variable per variant. A test only teaches something if exactly one thing changes. Hold layout, motion, colors, and timing constant; swap only the field under test (hook, or offer, or CTA). Mixing two changes makes the winner uninterpretable.
  2. Message-match the hook to the CTA. The promise made in the first 3 seconds must be the promise the button pays off. A "stop wasting 2 hours a day" hook ends on "Save 2 hours — try free," not a generic "Shop now." The hook and CTA are a matched pair in every row of the data table.

Ad anatomy (15–30s)

BeatJobBudget
Hook (0–3s)Stop the scroll; state the problem or pattern-interrupt0–3s
ContextMake the viewer feel the problem / agitate3–8s
PayoffShow the product as the solution, one clear benefit8–18s
ProofOne concrete number, demo, or result18–25s
CTA (last 3s)Single action, message-matched to the hookhold ≥2s

Land the hook's emotional trigger before the 2-second mark — judgment forms in ~1.7s and scroll speed keeps rising. Test the hook FIRST; it moves CPA more than any other element.

Hook taxonomy (the field you vary most)

Hook typeTemplateWhen it wins
Problem-first"I didn't realize how much X was costing me until…"pain is felt but unnamed
Pattern interrupt"This is going to sound controversial, but…"crowded feed, generic category
Curiosity gap"Nobody talks about the one thing that…"educated, skeptical audience
Immediate benefit"Cut your X in half in 7 days"clear, quantifiable outcome
Direct callout"If you do X every morning, stop."sharp audience segment

Keep a column of 5–10 hook strings and let the batch render produce one ad per hook against the same body — that is the cleanest hook test possible.

Data-driven template (everything is a prop)

Hardcode nothing the marketer might A/B test. The composition reads a single variant object; the renderer never edits the component. Make every animated value a pure function of useCurrentFrame() so each frame renders deterministically (no CSS transitions, no library timers — they desync the render).

import {useCurrentFrame, interpolate, AbsoluteFill, useVideoConfig} from "remotion";

type Variant = {hook: string; benefit: string; proof: string; cta: string; bg: string; accent: string};

export const AdTemplate: React.FC<{v: Variant}> = ({v}) => {
  const frame = useCurrentFrame(); const {fps} = useVideoConfig();
  const hookIn = interpolate(frame, [0, 8], [40, 0], {extrapolateRight: "clamp"});
  const ctaIn  = interpolate(frame, [fps * 22, fps * 23], [0, 1], {extrapolateLeft: "clamp", extrapolateRight: "clamp"});
  return (
    <AbsoluteFill style={{background: v.bg, fontFamily: "Inter, sans-serif"}}>
      <h1 style={{transform: `translateY(${hookIn}px)`, opacity: interpolate(frame,[0,8],[0,1])}}>{v.hook}</h1>
      {frame > fps * 8  && <p>{v.benefit}</p>}
      {frame > fps * 18 && <strong>{v.proof}</strong>}
      <button style={{opacity: ctaIn, background: v.accent}}>{v.cta}</button>
    </AbsoluteFill>
  );
};

See references/batch-ad-pipeline.md for the complete template, the CSV→props parser with validation, and the brand-lock theme object.

1 template × CSV = N variants

The payoff: author the ad once, then let a CSV drive the matrix. Each row is one fully-formed, message-matched variant. Render once per row.

# variants.csv: id,hook,benefit,proof,cta,bg,accent  → one MP4 per row
node csv-to-props.js variants.csv ./props        # writes props/<id>.json per row
for f in props/*.json; do
  id=$(basename "$f" .json)
  npx remotion render AdTemplate "out/${id}.mp4" --props="$f"
done

To build a clean test matrix, hold every column constant and vary one: 5 hooks × 1 body = 5 ads (hook test); then take the winning hook × 3 CTAs (CTA test). references/batch-ad-pipeline.md has a matrix generator that expands variable lists into the CSV so combinatorial tests stay one-variable-at-a-time.

Multi-aspect export

Each placement wants a native aspect. Compose against a center safe zone so one master reframes cleanly to all of them — don't letterbox, re-center.

AspectPlacementsResolutionKeep-clear safe zone
9:16Reels, Stories, TikTok In-Feed1080×1920top ~14%, bottom ~20–35% (UI/caption), sides ~6%
4:5Meta Feed (best feed real estate)1080×1350center 90%; CTA out of bottom 10%
1:1Feed, broad reach1080×1080center square survives every crop
16:9YouTube, in-stream1920×1080center 90%

Design the hook text, product, and CTA inside the 1:1 center square so they survive every crop. Drive aspect from a prop and render each per variant:

for ar in 9x16 4x5 1x1; do
  npx remotion render AdTemplate "out/${id}-${ar}.mp4" --props="props/${id}.json" \
    --props-merge="{\"aspect\":\"${ar}\"}"
done

Keep critical content out of the 9:16 bottom third — that is where the platform stacks captions, the CTA button, and the username. references/platform-specs.md has exact pixel safe zones and per-platform duration/format limits.

Output checklist

  • One variable per variant; layout/motion/timing identical across a test.
  • Hook lands its trigger before 2s; hook and CTA are message-matched in every row.
  • Every animated value is a pure function of useCurrentFrame() — no CSS/library timers.
  • Dataset is a prop; nothing testable is hardcoded; one template renders the whole CSV.
  • Each variant exported 9:16 + 4:5 + 1:1 (+16:9 if needed), key content in the center square.
  • Brand stays locked across all variants via a single theme object.

Deliver & verify (rendered stills → MP4)

Packaged helper (scripts/): tile your stills with scripts/contact-sheet.sh sheet.png f-hook.png f-mid.png f-end.png, then assert the encode with scripts/probe-mp4.sh out.mp4 [WxH] [fps]. See scripts/README.md.

Output contract:

  • A Remotion ad template registered as <Composition> (+ zod schema + defaultProps), every animated value frame-driven (no CSS transitions / library timers / Date.now() / Math.random()).
  • Deliverable = the rendered out/*.mp4 per variant per aspect (plus the project + CSV, so the marketer re-renders on new rows).

Verify loop — render stills → inspect → encode. Cheap PNGs first, full encode only once they're clean. Render with the shipped props (the real row), not just defaultProps.

# Frame-exact stills across the hook→CTA arc, with a real variant's props
npx remotion still AdTemplate out/f-hook.png --frame=12  --props=props/v1.json   # hook readable < 2s
npx remotion still AdTemplate out/f-mid.png  --frame=300 --props=props/v1.json   # benefit/proof
npx remotion still AdTemplate out/f-cta.png  --frame=689 --props=props/v1.json   # CTA, message-matched
# inspect each: fidelity (hook / offer / proof / CTA text exact, brand bg+accent right)
#   AND artifacts (text overflow, off-canvas, CTA inside the 9:16 bottom third, missing font, wrong row binding)

Multi-aspect / batch — verify one variant in EACH aspect before batch-rendering the matrix. A layout bug repeats across every row × aspect; catch it once.

for ar in 9x16 4x5 1x1; do                                  # one representative variant, every target aspect
  npx remotion still AdTemplate "out/v1-${ar}.png" --frame=300 \
    --props=props/v1.json --props-merge="{\"aspect\":\"${ar}\"}"
done
# stills clean in all aspects? → then batch-render every row × aspect:
for f in props/*.json; do id=$(basename "$f" .json); for ar in 9x16 4x5 1x1; do
  npx remotion render AdTemplate "out/${id}-${ar}.mp4" --props="$f" --props-merge="{\"aspect\":\"${ar}\"}"
done; done
npx remotion render AdTemplate out/demo.gif --codec=gif --props=props/v1.json   # README demo

Before you finish:

  1. npx remotion still renders cleanly at hook / mid / CTA — no errors, no missing fonts/assets.
  2. Hook/offer/proof/CTA text exact and brand colors right; nothing in the 9:16 bottom third or outside the center safe zone.
  3. Frame-driven only — no CSS/library timers, Date.now(), or Math.random().
  4. Shipped row's props render correctly (not just defaultProps); one variable per variant holds.
  5. One variant verified in 9:16 + 4:5 + 1:1 before the batch; MP4s play; (optional) GIF for the README.

Reference files

  • references/batch-ad-pipeline.md — the full runnable Remotion ad template, the CSV→typed-props parser with validation, a one-variable test-matrix generator, the brand-lock theme object, and the complete batch + multi-aspect render script.
  • references/platform-specs.md — exact 2025/2026 pixel safe zones, durations, aspect ratios, and format limits for Meta Feed/Reels/Stories, TikTok In-Feed, and YouTube, plus a hook→CTA message-match worksheet.

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.