agentleFS
BenPremBenPrem/business-brain/.claude/skills/social-video-producer/SKILL.md

social-video-producer

Skill1 starsChanged 3 months agoMITView on GitHub

Full video production system across four engines — AI video generation via a CLI vendor (worked example Higgsfield; Seedance, Kling, Veo, Hailuo), Remotion motion graphics, FFmpeg real-footage editing, and cinematic multi-shot production with native audio. Includes auto-captions, safe-zone end cards, and platform export. Triggers on "video", "reel", "video ad", "TikTok/Reel/Short", "video for <CLIENT>". NOT for scroll-driven on-site animation — use interactive-website-builder.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

Show its rules and text

We split this skill into the rules it gives an agent

16 rules. About 0 are common, 10 are distinctive, and 6 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈10

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 6

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Create video content for clients using four production engines.

    line 8
  2. Very specific:

    Before every job: query the vendor for the model's current params and constraints (e.g.

    line 22
  3. Very specific:

    Ask for the target aspect ratio natively (--aspect-ratio 9:16).

    line 23
  4. Very specific:

    Auto-captions (Whisper): every reel gets captions unless the operator says otherwise — captions raise watch time materially.

    line 60
  5. Distinctive:

    Voiceover: only when the brief calls for narration.

    line 61
  6. Distinctive:

    Write as a shot list (visual, text overlay, audio, model, input mode per shot).

    line 87
  7. Distinctive:

    Use IMAGE-TO-VIDEO exclusively — never text-to-video.

    line 95
  8. Distinctive:

    Keep i2v prompts under 30 words describing only MOTION, LIGHTING, ATMOSPHERE — never re-describe what's already in the image.

    line 97
  9. Distinctive:

    After 3 failed attempts on one shot, stop retrying and switch models — don't burn credits on a model that's fighting you.

    line 117
  10. Very specific:

    Run on EVERY reel unless explicitly skipped.

    line 122
  11. Distinctive:

    Never generate video without the operator approving the script first.

    line 170
  12. Very specific:

    Never deliver without the operator reviewing the final render — extract frames, view them, check subject/framing/palette before calling it …

    line 171
  13. Distinctive:

    Always add captions unless explicitly skipped.

    line 172
  14. Very specific:

    Always check clients/<slug>/brand/ before creating anything.

    line 173
  15. Distinctive:

    Never open with the business logo — open with the customer's problem.

    line 174
  16. Distinctive:

    Every reel works standalone — assume the viewer has never heard of the brand.

    line 180

≈ “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

Social Video Producer

Create video content for clients using four production engines. Read this entire file before generating anything. The worked example vendor throughout is the Higgsfield CLI — swap in your own AI-video vendor; the engine routing and production doctrine are vendor-independent.

Routing Gate

This skill: reels, video ads, product demos, explainers, before/after reveals, testimonials, short films, website hero loop videos — any standalone video file. interactive-website-builder: video as scroll-driven animation embedded ON a website. content-repurposer: adapting written content into text posts (no video output).


The Four Engines

Engine 1 — AI Video Generation (CLI vendor)

Generates clips from text prompts or still images. Use when the client has NO raw footage: social reels, product demos, lifestyle B-roll, hero loops.

  • Before every job: query the vendor for the model's current params and constraints (e.g. higgsfield model get <job_type>), then price it before generating (e.g. higgsfield generate cost <job_type> --prompt "..." — free and exact). Model params drift; never assume.
  • Ask for the target aspect ratio natively (--aspect-ratio 9:16). Only pillarbox in FFmpeg when a model genuinely lacks the param.
  • Never bake text into AI-generated clips — all models struggle with text rendering. Generate clean video; overlay text in post (Step 6).

Engine 2 — Motion Graphics (Remotion)

Programmatic video from React code — text animations, data reveals, branded graphics. Use when content is primarily TEXT + GRAPHICS: stats reels, quote cards, service highlights, logo reveals. Cost: $0.

Engine 3 — Real Footage Editing (FFmpeg)

Edit, combine, and polish client-provided clips; add captions, overlays, transitions. Cost: $0.

Engine 4 — Cinematic Multi-Shot Production

Joint audio-video models (worked example: Seedance 2.0) generate synchronized sound + visuals in one pass — native lip-sync, ambient SFX, music-synced motion, no post-production audio work. Use for short films, brand films, multi-shot narratives, dialogue content, music videos, or character consistency across 5+ shots. For standard reels without audio-sync needs, Engine 1 is faster and cheaper.

  • Character consistency: prefer a trained reusable character reference (e.g. higgsfield soul-id) over ad-hoc reference packs when a character recurs across a campaign. Reference packs: max 3 stills per subject, consistent lighting.
  • Free-iterate complex shots on a vendor's free tier if one exists (watermarked); regenerate final picks on the paid path.

Engine Routing Tree

Does the client have raw footage?
├── YES → Engine 3 (FFmpeg)
│   └── Also need AI B-roll? → Engine 3 + Engine 1 hybrid
├── NO → Content primarily text/stats/graphics?
│   ├── YES → Engine 2 (Remotion) — free
│   └── NO → Multi-shot cinematic (5+ shots, character lock, native audio)?
│       ├── YES → Engine 4
│       └── NO → Engine 1
│           ├── Cinematic brand story → highest-quality model tier
│           ├── Product demo / lifestyle → mid-tier quality mode
│           ├── Quick social reel → fastest turbo model, 9:16 native
│           ├── Seamless product loop → same still as start AND end frame
│           ├── Animate existing illustrations → image-to-video (see Step 4)
│           └── Budget volume (10+ reels) → cheapest model; price the batch first

Shared Production Stack

Auto-captions (Whisper): every reel gets captions unless the operator says otherwise — captions raise watch time materially. Local whisper --word_timestamps True --output_format json is free. Voiceover: only when the brief calls for narration; confirm a TTS vendor is connected before promising narrated deliverables. Final assembly: FFmpeg, last step in every pipeline.

Platform Export Specs

PlatformRatioResolutionBest duration
Instagram Reel9:161080x192015-30s
TikTok9:161080x192015-60s
YouTube Short9:161080x192030-60s
YouTube long-form16:91920x10802-10min
Facebook9:16 or 1:11080x1920 / 1080x108015-60s
LinkedIn1:1 or 16:91080x1080 / 1920x108030-120s

Set aspect ratio at generation time — never crop after. All vertical output keeps text, faces, logos, and CTAs inside platform safe zones (UI chrome eats the top ~15% and bottom ~20% plus the right rail on 9:16 — keep critical content in the center safe area).


Production Workflow

Step 1 — Brief + budget check

Confirm: client (check clients/<slug>/brand/ first) · purpose · platforms · duration · message · does the client have footage (determines engine) · voiceover or text-only · style/mood vs brand · CTA · reference reels · character count and consistency needs · budget (estimate 2-3 attempts per AI shot; a 30s video ≈ 6 shots ≈ 15 generations — price the batch and check the account balance before large runs).

Step 2 — Script and storyboard

Structure: Hook (0-3s) — the customer's problem, never the business name/logo. Value (3-12s) — the solution or transformation. Proof (12-18s) — social proof, results, before/after. CTA (last 2-3s).

Write as a shot list (visual, text overlay, audio, model, input mode per shot). Present the script to the operator for approval BEFORE generating.

Step 3 — Character and location reference sheets (Engine 4 only)

Generate turnaround sheets per character (front full-body, profile, back, portrait close-up — consistent lighting, proportions, clothing, clean background) and per location (wide establishing, detail close-up, 3/4 angle, reverse wide — consistent time of day, no people). These become the reference pack; for recurring characters train a reusable reference instead.

Step 4 — Generate raw content

CRITICAL: does the client have existing illustrations or artwork? If YES (book illustrations, product photos, brand art):

  • Use IMAGE-TO-VIDEO exclusively — never text-to-video. The client's art IS the brand; never generate imagery that replaces it.
  • Flatten transparent PNGs onto the target background color before upload — transparency handling is unpredictable across models.
  • Keep i2v prompts under 30 words describing only MOTION, LIGHTING, ATMOSPHERE — never re-describe what's already in the image.

Otherwise, prompt with the six-step structure: Subject → Action → Environment → Camera → Style → Constraints. Always append a constraint block: "Face stable without deformation, normal human structure, natural smooth movements. 4K, cinematic texture, no blur, no ghosting, no flickering."

Engine 4 flow: build the reference pack → generate shot-by-shot with references attached and audio generation on → set aspect ratio per platform → review each shot → assemble in FFmpeg with crossfades.

Step 5 — Iterate and troubleshoot

Budget 2-3 attempts per shot. Common fixes:

ProblemQuick fixNuclear option
Face distortion"face stable" constraint, anchor imageSwitch to image-to-video with a reference photo
Jittery motionAdd speed descriptor ("slow", "gentle")ONE action verb, longer clip
Wrong camera moveExactly one camera verbRemove the camera instruction entirely
Character drift across shotsSame anchor image every shotReference pack at weight ~0.8
Hand/finger distortion"hands naturally at sides"Frame above the hands
Stray text/watermarks"No text, no watermarks, no signs"Regenerate
Prompt partially ignoredPrompt too long (120+ words)Trim to 60-80 words; move detail to anchor images
Motion/audio mismatchJoint audio-video modelGenerate video first, sync audio in FFmpeg

After 3 failed attempts on one shot, stop retrying and switch models — don't burn credits on a model that's fighting you. Past 2 retries, diagnose (extract frames, measure) before attempting a third.

Trim rule: cut the first and last 0.5s of every AI-generated clip before assembly — models produce boundary artifacts.

Step 6 — Captions

Run on EVERY reel unless explicitly skipped. Some Homebrew/minimal FFmpeg builds ship without the drawtext/subtitles/ass filters (ffmpeg -filters | grep drawtext). The portable method works everywhere:

  1. Transcribe: whisper audio.wav --word_timestamps True --output_format json
  2. Chunk words into caption cards: ≤4 words and ≤1.8s per card
  3. Render each card as a transparent PNG with Pillow — white bold text, black outline, max 2 lines
  4. Composite with the overlay filter, one input per card:
ffmpeg -i raw.mp4 -i cap1.png -i cap2.png -filter_complex \
  "[0:v][1:v]overlay=(W-w)/2:1450:enable='between(t,0.0,1.8)'[v1];\
   [v1][2:v]overlay=(W-w)/2:1450:enable='between(t,1.8,3.4)'[vout]" \
  -map "[vout]" -map 0:a -c:a copy captioned.mp4

Caption rules: bottom third but INSIDE the safe zone, below faces and held products, 1.5s+ per phrase.

Step 7 — Voiceover (if applicable)

ffmpeg -i video.mp4 -i vo.mp3 -i music.mp3 \
  -filter_complex "[1:a]volume=1.0[vo];[2:a]volume=0.12[bg];[vo][bg]amix=inputs=2[a]" \
  -map 0:v -map "[a]" -c:v copy narrated.mp4

Voiceover 100%, background music 10-15%. Loudness-normalize client speech with loudnorm=I=-16:TP=-1.5:LRA=11.

Step 8 — Platform export (every target in one pass)

ffmpeg -i final.mp4 -vf "scale=1080:1920:force_original_aspect_ratio=decrease,pad=1080:1920:(ow-iw)/2:(oh-ih)/2" -c:v libx264 -crf 23 -c:a aac reel_vertical.mp4
ffmpeg -i final.mp4 -vf "scale=1080:1080:force_original_aspect_ratio=decrease,pad=1080:1080:(ow-iw)/2:(oh-ih)/2" -c:v libx264 -crf 23 -c:a aac reel_square.mp4
ffmpeg -i final.mp4 -vf "scale=1920:1080:force_original_aspect_ratio=decrease,pad=1920:1080:(ow-iw)/2:(oh-ih)/2" -c:v libx264 -crf 18 -c:a aac video_horizontal.mp4

Step 9 — Deliver

Copy outputs to clients/<slug>/deliverables/video/. Log the deliverable in your CRM record and update the related task.


Content Templates

  • Before/after transformation (15-30s): Engine 1 or 3. Hook → before → transition → after → CTA.
  • Quick tips (10-15s): Engine 2. Hook → tips 1-3 → CTA.
  • Product/service demo (15-30s): Engine 1 (turbo model for speed, joint-audio model for quality). Problem → intro → demo → proof → CTA.
  • Testimonial (15-30s): Engine 3 + 2. Best quote as hook → client speaking → results → CTA.
  • Stats/data reveal (10-20s): Engine 2. Title → animated data → takeaway + CTA.
  • Day-in-the-life (15-60s): Engine 3 + 1. Hook → process quick cuts → result → CTA.
  • Cinematic brand film (30-120s): Engine 4, full narrative arc, reference-pack consistency.
  • Animated illustrated content (15-60s): Engine 1, image-to-video from the client's existing artwork.

Hard Rules

  1. Never generate video without the operator approving the script first.
  2. Never deliver without the operator reviewing the final render — extract frames, view them, check subject/framing/palette before calling it done.
  3. Always add captions unless explicitly skipped.
  4. Always check clients/<slug>/brand/ before creating anything.
  5. Never open with the business logo — open with the customer's problem.
  6. Match platform aspect ratio at generation time, never by cropping after.
  7. Hold on-screen text a minimum of 1.5 seconds.
  8. Export for every target platform in one pass.
  9. Log every deliverable in the CRM and task system.
  10. Route to the cheapest engine that meets the quality bar — Remotion and FFmpeg are free.
  11. Every reel works standalone — assume the viewer has never heard of the brand.
  12. Trim first/last 0.5s of every AI clip; budget 2-3 attempts per shot; switch models after 3 failures.
  13. Safe zones on all vertical output.
  14. Ad copy is grounded in the actual creative — extract frames and verify what the video shows before writing accompanying copy; never invent seasonal hooks or details not present in the footage.
  15. Any regulated topic in a client's vertical (financing, health claims, legal services) passes that client's compliance checklist BEFORE anyone sees the video.

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.