Viral Producer

Viral Producer — Faceless Reel Production Pipeline

Purpose

Receive a complete production spec from the viral-ig-marketer orchestrator and render it into a ready-to-post Instagram Reel. This skill owns the production pipeline from validated spec to final .mp4. It does not select content angles, score virality, or write captions — those decisions are made upstream.

The input is a spec. The output is content.

Scope

• Platform: Instagram Reels (9:16 vertical, 1080x1920, h264+aac, .mp4)
• Content type: Faceless only — no face-on-camera content
• Format tiers: T1-T4 (niche formats) and P1-P3 (premium formats)
• Rendering engine: Remotion (programmatic React-based video)
• Asset generation: fal.ai (AI imagery), stock footage, Figma/SVG design assets
• Output: Rendered .mp4 package at output/reels/<slug>/, ready for posting via Postiz or manually

Prerequisites

• Remotion — installed and configured. See references/remotion-guide.md for setup.
• fal.ai API key — required for AI image/video generation. Set in references/production-config.json
• ffmpeg — required for video verification and post-processing
• Node.js — required by Remotion
• Research brief — a completed brief from the viral-research skill, or at minimum a content angle with a specified format tier

Read references/production-config.json before every production session. It defines the niche, brand voice, visual system, and tool configuration.

Read ../viral-ig-marketer/references/soul.md before writing any text content (hooks, captions, CTAs). All copy must align with the brand voice defined there.

Production Workflow

Phase 1: Spec Validation

Goal: Validate the incoming production spec before starting work.

Spec contract: See ../viral-ig-marketer/references/production-spec.md for the full contract (required fields, optional fields, example). This is the single source of truth for the spec interface.

Process:

1. Read the production spec (passed as input from viral-ig-marketer)
2. Validate all required fields are present and non-empty
3. Verify format_tier is a recognized tier from references/format-templates.md
4. Read references/production-config.json for brand colors, fal.ai settings, Remotion composition mapping
5. If any required field is missing → error immediately with a clear message listing missing fields. Do not proceed.

Output: Validated spec ready for asset preparation.

Phase 2: Asset Preparation

Goal: Generate or source all visual and text assets needed for the Reel.

Read references/format-templates.md for the exact asset requirements per format tier.

Audio is sourced from ../viral-ig-marketer/audio/ using the production spec's audio_track field. Treat audio_track as the selected filename to resolve from that directory.

For T1 (Meme Reel):

• Source stock footage that visually contradicts the text message
• Write 1-2 text overlays (setup text + optional label)
• No custom assets needed

For T2 (Quote Card):

• Generate background image via fal.ai or source stock photo
• Write the quote text (read ../viral-ig-marketer/references/soul.md for voice)
• Style: bold, stacked, center-aligned

For T3 (Text Card):

• Source subtle background footage (timelapse, nature, gradient animation)
• Write both text sections (myth list + reality list, or equivalent comparison)
• Design text hierarchy (bold headers, lighter body)

For T4 (Truth Bomb):

• Source contemplative footage (real timelapse preferred over AI)
• Write the 3-part argument (fact → irony → conclusion)
• Choose clean serif typography

For P1 (Functional Loop):

• Design the interactive element (breathing circle, day counter, progress ring)
• Choose monochromatic brand palette from references/production-config.json
• Plan the loop point (last frame must match first frame)
• Text: 1-2 words maximum

For P2 (Data Viz Humor):

• Design the data element (gauge, counter, EKG, progress bar)
• Plan the visual metaphor or gamification hook
• Dark UI aesthetic (black + single accent color)
• Plan the destruction/escalation/resolution arc

For P3 (Cultural Edutainment):

• Source premium footage (drone aerial, cinematic nature — real, not AI)
• Design the frosted glass dictionary card
• Write: term, pronunciation, part of speech, definition, "see also" (humorous)
• Choose monochromatic palette aligned to footage

Asset generation via fal.ai:

# Example: generate a background image
curl -X POST "https://fal.run/fal-ai/flux/dev" \
  -H "Authorization: Key $FAL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "aerial drone shot of ocean waves at sunset, cinematic, 9:16 vertical",
    "image_size": {"width": 720, "height": 1280}
  }'

Respect the budget ceiling in references/production-config.json → fal.budgetCeilingUSD. Fall back to stock footage or solid/gradient backgrounds if the budget is hit.

Output: All assets saved to output/assets/<slug>/ — images, footage clips, SVGs, text files.

Phase 3: Remotion Rendering

Goal: Render the final .mp4 using Remotion.

Read references/remotion-guide.md for detailed rendering instructions per format tier.

General process:

1. Prepare the Remotion input props (JSON file with all content and asset paths):

{
  "format": "T2",
  "topic": "sobriety regret quote",
  "hook": "I would've been 1 year sober today...",
  "backgroundImage": "output/assets/slug/background.png",
  "textOverlays": [
    { "text": "I would've been", "style": "bold", "position": "center" },
    { "text": "1 year sober today.", "style": "bold", "position": "center" },
    { "text": "If I quit drinking", "style": "bold", "position": "center" },
    { "text": "1 year ago.", "style": "bold", "position": "center" }
  ],
  "duration": 6,
  "fps": 30
}

2. Pre-render validation (T2 motion check):

   Before invoking the render, verify the composition contains actual motion — not a single static frame:

   • If format is T2: inspect the Remotion input props and composition code. The composition must include at least one animation sequence:
     • Text animation: fade-in, typewriter, staggered reveal, or equivalent text motion
     • Background motion: Ken Burns zoom/pan, parallax shift, subtle particle effect, or slow camera drift
   • If both are missing (i.e., the composition would render a single static image for the entire duration), fail the render with error: "T2 motion check failed: composition contains no animation sequences. T2 Quote Card requires text animation + background motion. Add at least one text reveal and one camera/background movement before rendering."
   • For all other format tiers (T1/T3/T4/P1-P3): motion is inherent to their composition templates — no additional check needed.

3. Invoke the Remotion render:

npx remotion render <composition-id> \
  --props=<props-file> \
  --output=output/reels/<slug>.mp4 \
  --codec=h264

4. Verify the output:

ffprobe -v quiet \
  -show_entries stream=codec_name,width,height,duration \
  -show_entries format=duration \
  output/reels/<slug>.mp4

Expected: 1080x1920, h264 video + aac audio, duration matches target ±0.5s.

If Remotion is not set up or render prerequisites are missing, fail the cycle loudly. Do not silently degrade to a static/manual package.

Required behavior:

• Stop before claiming the reel is ready.
• Emit a clear error explaining why automated render could not run.
• Mark the cycle/output as render_failed, not ready-for-review and not ready-for-manual-render.
• If you generated useful inputs (backgrounds, props, caption drafts), keep them as debugging artifacts only — not as the primary deliverable.
• Never treat render-instructions.md as a successful reel output.

Output: Rendered .mp4 in output/reels/<slug>.mp4 or an explicit render failure.

Phase 4: Caption Save

Goal: Save the caption provided in the production spec.

The caption is written upstream by viral-ig-marketer. This phase saves it to the output package.

1. Extract the caption field from the production spec
2. Save to output/reels/<slug>/caption.txt
3. Verify the caption is non-empty

Output: Caption text saved to output/reels/<slug>/caption.txt

Phase 5: Output Package

Goal: Deliver a complete, ready-to-post package.

For each produced Reel, the final output includes:

output/reels/<slug>/
├── <slug>.mp4          — rendered Reel (1080x1920, h264+aac)
├── caption.txt         — full caption with CTA and hashtags
├── metadata.json       — production metadata (format tier, topic, hook type,
│                         virality score, assets used, render settings, audio)
└── thumbnail.png       — first frame extracted for preview

Extract the thumbnail:

ffmpeg -i output/reels/<slug>.mp4 -vf "select=eq(n\,0)" -vsync 0 \
  output/reels/<slug>/thumbnail.png -y

Write metadata.json:

{
  "slug": "<slug>",
  "format": "T2",
  "topic": "sobriety regret quote",
  "hookType": "urgency",
  "emotionalTrigger": "regret",
  "viralityScore": 4,
  "duration": 6.0,
  "resolution": "1080x1920",
  "productionDate": "YYYY-MM-DD",
  "assetsUsed": {
    "background": "fal.ai generated / stock / custom",
    "typography": "font name, weight, color",
    "audio": "track filename from production spec"
  },
  "caption": "first line of caption...",
  "hashtags": ["#tag1", "#tag2"],
  "status": "ready-to-post"
}

Present the package to the human for review before posting. The human makes the final publish decision — this skill never posts directly.

Operating Principles

• Research first, produce second. Never produce content without a research brief or at minimum a validated content angle. Uninformed content is wasted effort.
• Spec is the contract. The production spec from viral-ig-marketer defines what to produce. Do not second-guess creative decisions (topic, hook, CTA, caption) — produce exactly what the spec says. If the spec is incomplete, error with missing fields.
• Virality gate is upstream. The orchestrator (viral-ig-marketer) runs the virality gate before delegating to this skill. Viral-producer produces what it's told — it does not re-score.
• Faceless only. All content must be producible without a human appearing on camera. If a content angle requires face-on-camera, adapt the narrative structure to a faceless format (text overlay, animation, stock footage) or discard it.
• Under 10 seconds by default. The research data consistently shows sub-10-second Reels get higher completion rates and more algorithmic distribution. Only exceed 10 seconds for P2 (Data Viz) formats where the escalation arc requires it.
• Soul for visual tone. Read ../viral-ig-marketer/references/soul.md for brand voice context. Use it to guide visual tone and typography choices — not to rewrite copy (that's upstream).
• Fail loudly on Remotion unavailability. If Remotion isn't available, record an explicit render failure instead of shipping an instructions-only package as the deliverable. If fal.ai budget is hit, fall back to stock or gradient backgrounds, but the pipeline must still produce a real .mp4 to count as complete.
• Loop everything possible. Seamless loops (last frame = first frame) maximize watch time, which is the algorithm's #1 ranking signal. For static formats (T1, T2), the loop is inherent. For animated formats (P1, P2), engineer the loop point intentionally.
• All Reels MUST contain motion — no static frames. Every rendered Reel must include at least one animation sequence. For T2 Quote Card specifically: require text animation (fade-in, typewriter, or staggered reveal) + slow background motion (Ken Burns zoom/pan, parallax, or subtle particle effect). Duration 5-7s. T1/T3/T4/P1-P3 inherently contain motion. Never render a single static frame as a Reel — Instagram's algorithm deprioritizes static images published as Reels vs actual video content.

Memory

All paths relative to this skill's directory:

• Production config: references/production-config.json
• Brand voice: ../viral-ig-marketer/references/soul.md (single source of truth — no local copy)
• Format templates: references/format-templates.md
• Remotion guide: references/remotion-guide.md
• Caption guide: ../viral-ig-marketer/references/caption-guide.md (owned by viral-ig-marketer — creative decisions live upstream)
• Output directory: output/ (created on first run if missing)
• Asset directory: output/assets/ (generated assets per production run)
• Reel directory: output/reels/ (final rendered Reels)