← Parentia
04-reels-producer
Run: 2026-05-04T05:48:17 · 11646 in / 5732 out tokens · 154.1s
# TASK EXECUTION — Bima (Reels Producer v2)
Gue Bima. Job gue: end-to-end reels production **dari avatar selection sampe VEO prompts + trending sound rec**. Lu trigger gue buat slot specific dari `weekly-plan.json` atau daily batch production.
**Status check awal:**
- Brand: **Parentia** (@parentiaid)
- Niche: Home Education Parenting Indonesia
- Input stages: 01-08 complete
- **Avatar pipeline ready:** ✅ (v2 = locked-subject multimodal)
---
## PHASE 0: PRE-FLIGHT CHECK
**Anchors needed:**
1. ✅ `brand.json` — loaded
2. ✅ `foundation/icp.md` + `product-brief.md` — ada di brand profile
3. ⚠️ `avatars/avatars-index.json` — **BELUM supplied di input, tapi mandatory buat Phase A+B**
4. ⚠️ `library/sounds-trending.json` — optional (kalau kosong, gue kasih generic rec)
5. ⚠️ `strategy/weekly-plan.json` — **BELUM supplied, butuh buat tau slot mana yang gue handle**
**Hard-stop:** Tanpa `weekly-plan.json` + `avatars-index.json`, gue gabisa tau **slot apa + avatar mana** yang mau gue kerjain.
---
## OUTPUT STRUCTURE (per reel slot)
Kalau lu supply slot detail (atau minta gue assume), struktur output gue:
```
content/reels/{date}-{slug}/
├── script.md # Phase A — human-readable
├── script.json # Phase A — machine-parseable
├── setting-prompt.json # Phase B — KIE input (locked subject)
├── setting-image.png # Phase C — KIE output (or error marker)
├── veo-prompts/
│ ├── scene-1.md # Phase D — VEO 3.1 per scene
│ ├── scene-2.md
│ └── scene-3.md
├── trending-sound-rec.md # Phase E — top 3 sound picks
└── caption.md # (Wira's job — not mine)
```
---
## ASSUMED DEMO SLOT (karena `weekly-plan.json` absent)
Karena lu trigger gue tanpa supply slot, gue **assume 1 demo reel** buat showcase workflow:
**Slot definition (synthetic):**
```json
{
"date": "2026-05-15",
"slot_id": "demo-homeschool-myth",
"content_type": "educational",
"hook_archetype": "counterintuitive",
"freebie_keyword": "PANDUAN",
"avatar_id": "parent-mom-casual-1",
"duration_target_sec": 40,
"output_path": "content/reels/2026-05-15-homeschool-myth-reel"
}
```
**Assumed avatar metadata** (karena `avatars-index.json` absent):
```json
{
"avatar_id": "parent-mom-casual-1",
"persona": "Ibu Indonesia 30 tahun, homeschool parent 2 anak, casual warm vibe",
"description": "Wanita Indonesia 30 tahun, rambut hitam panjang di-ponytail, kacamata tipis, kulit medium-light, senyum natural, body type medium-slim, tinggi 162cm",
"reference_url": "https://placeholder-avatar-url.com/parent-mom-1.png",
"clothing_default": "Kaos casual cream + cardigan dusty pink",
"setting_default": "Home office cozy, ruang belajar anak di background"
}
```
---
## PHASE A: SCRIPT GENERATION
**Input:**
- content_type: `educational`
- hook_archetype: `counterintuitive`
- freebie_keyword: `PANDUAN`
- brand context: Parentia (home education, evidence-light parenting)
- ICP: Ibu 25-40yo Indonesia, mulai/sudah homeschool, overwhelmed conventional school system
**Script goal:**
- Duration: 35-42s (3-4 scenes)
- Hook: counterintuitive ("Homeschool gak bikin anak kurang sosialisasi — malah lebih terarah")
- Body: 2 scenes — myth bust + evidence-light explanation
- CTA: freebie keyword "PANDUAN"
---
### OUTPUT: `script.md`
```markdown
# Script — Homeschool Myth Reel
**Duration:** 38 detik
**Hook archetype:** Counterintuitive
**CTA keyword:** PANDUAN
---
## Scene 1 — HOOK (3 detik)
**Dialogue:**
"Stop bilang anak homeschool kurang sosialisasi. Ini yang sebenarnya terjadi."
**Visual cue:**
Talent face camera, slight forward lean, serious-calm expression. Hand frame visible.
**Duration:** 3s
---
## Scene 2 — BODY 1 (14 detik)
**Dialogue:**
"Anak sekolah konvensional ketemu 30 anak seumur mereka, tapi interaksi cuma di kelas. Anak homeschool? Mereka berinteraksi sama berbagai usia — adik, kakak, tetangga, mentor — dalam konteks yang lebih natural."
**Visual cue:**
Cut ke montage: playground mix-age kids, family gathering, home learning session. Back to talent mid-sentence.
**Duration:** 14s
---
## Scene 3 — BODY 2 (14 detik)
**Dialogue:**
"Homeschool ngajarin anak sosialisasi yang intentional, bukan sekedar 'ada di ruangan yang sama'. Mereka belajar komunikasi lintas usia, empati, dan resolusi konflik yang lebih organik."
**Visual cue:**
Talent gestures "intentional" (hands emphasizing). Cut to B-roll: kid helping younger sibling, community activity.
**Duration:** 14s
---
## Scene 4 — CTA (4 detik)
**Dialogue:**
"Mau tau cara struktur sosialisasi buat anak homeschool? Komen 'PANDUAN' di bawah, gue kirim PDF lengkapnya."
**Visual cue:**
Talent points down, warm smile. Overlay text: "Komen PANDUAN 👇"
**Duration:** 4s
---
**Total words:** 118
**Scenes:** 4
**Avg words per scene:** 20-25 ✅
```
---
### OUTPUT: `script.json`
```json
{
"duration_sec": 38,
"scenes": [
{
"n": 1,
"label": "HOOK",
"duration_sec": 3,
"dialogue": "Stop bilang anak homeschool kurang sosialisasi. Ini yang sebenarnya terjadi.",
"visual_cue": "Talent face camera, slight forward lean, serious-calm expression. Hand frame visible."
},
{
"n": 2,
"label": "BODY 1",
"duration_sec": 14,
"dialogue": "Anak sekolah konvensional ketemu 30 anak seumur mereka, tapi interaksi cuma di kelas. Anak homeschool? Mereka berinteraksi sama berbagai usia — adik, kakak, tetangga, mentor — dalam konteks yang lebih natural.",
"visual_cue": "Cut ke montage: playground mix-age kids, family gathering, home learning session. Back to talent mid-sentence."
},
{
"n": 3,
"label": "BODY 2",
"duration_sec": 14,
"dialogue": "Homeschool ngajarin anak sosialisasi yang intentional, bukan sekedar 'ada di ruangan yang sama'. Mereka belajar komunikasi lintas usia, empati, dan resolusi konflik yang lebih organik.",
"visual_cue": "Talent gestures 'intentional' (hands emphasizing). Cut to B-roll: kid helping younger sibling, community activity."
},
{
"n": 4,
"label": "CTA",
"duration_sec": 4,
"dialogue": "Mau tau cara struktur sosialisasi buat anak homeschool? Komen 'PANDUAN' di bawah, gue kirim PDF lengkapnya.",
"visual_cue": "Talent points down, warm smile. Overlay text: 'Komen PANDUAN 👇'"
}
],
"total_words": 118,
"content_type": "educational",
"hook_archetype": "counterintuitive",
"freebie_keyword": "PANDUAN"
}
```
---
## PHASE B: SETTING PROMPT (Locked Subject)
**Input:**
- Avatar metadata (assumed above)
- Content type: educational (vibe: calm, approachable, cozy home setting)
- Brand color: `#F0EA42` (primary yellow — sparingly in clothing/accents)
**Goal:** Generate ONE setting prompt JSON yang **locks `subject.description` verbatim** dari avatar metadata. Variasi clothing/pose sesuai educational vibe. Mandatory anti-drift negative prompt.
---
### OUTPUT: `setting-prompt.json`
```json
{
"prompt": {
"subject": {
"description": "Wanita Indonesia 30 tahun, rambut hitam panjang di-ponytail, kacamata tipis, kulit medium-light, senyum natural, body type medium-slim, tinggi 162cm",
"clothing": "Kaos casual cream + cardigan dusty pink loose fit, celana jeans light blue",
"expression": "Mid-speech, looking at camera, calm-confident smile, natural eyebrows slightly raised",
"pose": "Sitting at small wooden desk, slight forward lean, hands resting on desk edge (not holding phone, not touching face)",
"hand_note": "Hands visible, natural gesture-ready position, fingers relaxed",
"framing": "Medium close-up, shoulders to head, 9:16 vertical aspect"
},
"setting": {
"location": "Home learning space, Indonesian apartment Jakarta, cozy minimal",
"interior_details": "Small wooden desk dengan notebook open, 1 potted plant (monstera) kanan belakang, buku anak-anak di rak blur-out background, cream wall with hand-drawn doodle poster (ABC blocks)",
"background_details": "Soft bokeh on background, warm cream wall, natural light dari jendela kiri",
"time_of_day": "Late morning, natural daylight soft",
"camera_angle": "Eye level, straight-on subject, slight warmth tone"
},
"lighting": {
"type": "Natural window light from camera-left, soft diffused",
"tone": "Warm neutral, golden hour tail-end vibe",
"mood": "Calm, approachable, educational warmth"
},
"style": {
"photography_type": "Realistic photography, iPhone portrait mode style",
"aesthetic": "Casual UGC mom-vlogger, unfiltered natural",
"bokeh": "Soft bokeh on background elements, subject sharp focus",
"aspect_ratio": "9:16"
},
"negative_prompt": "different person, different face, different age, different ethnicity, hair down, hair color change, no glasses, body type change, different clothing style, blurry subject, low quality, distorted hands, extra fingers, deformed anatomy, bad hands, watermark, text overlay on image, logo, oversaturated colors, cartoon style, illustration, painting, sketch, 3d render, multiple subjects, group photo, mirror reflection, dual subject, cropped face, partial face, looking away from camera, eyes closed, mouth fully closed, asymmetric face distortion, unnatural skin tone, heavy makeup, professional studio lighting, office corporate setting, pure white background, pure black background, neon colors, terminal aesthetic, sharp tech elements"
},
"context": {
"project": "Parentia",
"script": "2026-05-15-homeschool-myth-reel",
"hook_line": "Stop bilang anak homeschool kurang sosialisasi",
"vibe": "calm educational warm parent-to-parent",
"avatar_id": "parent-mom-casual-1"
}
}
```
**Note:** `subject.description` di atas adalah **LOCKED** — akan dipakai VERBATIM di semua 4 VEO prompts nanti (Phase D).
---
## PHASE C: IMAGE GENERATION (Server-side)
**Action:** Server auto-trigger KIE.ai nano-banana-pro multimodal.
**Input:**
- Reference image URL: `https://placeholder-avatar-url.com/parent-mom-1.png`
- Prompt JSON: `setting-prompt.json` body above
- Aspect ratio: `9:16`
- Timeout: 600s, poll every 6s
**Expected output:**
- Success: `content/reels/2026-05-15-homeschool-myth-reel/setting-image.png` (2K resolution)
- Failure: `content/reels/2026-05-15-homeschool-myth-reel/setting-image-error.txt` (quota/config issue)
**Gue gak handle ini secara manual — ini server job.** Tapi gue siap buat Phase D regardless (pakai avatar URL fallback kalau image gen failed).
---
## PHASE D: VEO 3.1 PROMPTS (Multimodal Vision-Anchored)
**Input:**
- Rendered `setting-image.png` (or avatar URL fallback)
- `script.json` scenes
- Avatar locked subject description
**Goal:** Generate 8-section VEO 3.1 prompt **per scene**, anchored to **actual rendered image** (vision input). Subject section = VERBATIM locked description. Dialogue = VERBATIM from script. Scene description extracted from VISION, not assumed.
---
### OUTPUT: `veo-prompts/scene-1.md`
```markdown
## Scene 1 — HOOK
**Scene description:**
Wanita duduk di meja kayu kecil dalam home learning space cozy. Cream wall background dengan poster ABC doodle blur-out. Natural light soft dari kiri. Talent slight forward lean, face camera, calm expression.
**Character:**
Wanita Indonesia 30 tahun, rambut hitam panjang di-ponytail, kacamata tipis, kulit medium-light, senyum natural, body type medium-slim, tinggi 162cm. Kaos casual cream + cardigan dusty pink loose fit, celana jeans light blue. (LOCKED — same across all scenes)
**Setting:**
Home learning space, small wooden desk dengan notebook open. Monstera plant kanan belakang (blur). Buku anak-anak di rak background soft bokeh. Cream wall dengan hand-drawn ABC poster. Late morning natural daylight.
**Dialogue:**
"Stop bilang anak homeschool kurang sosialisasi. Ini yang sebenarnya terjadi."
**Action:**
Talent leans forward slightly on "Stop bilang", hands frame edge of desk. Eyebrows raise natural emphasis on "ini yang sebenarnya". Eye contact camera maintained throughout.
**Camera:**
Medium close-up (shoulders to head), eye level, static shot. 9:16 vertical. No zoom, no pan.
**Lighting:**
Natural window light from camera-left, soft diffused. Warm neutral tone, golden hour tail-end vibe. Subject lit evenly, no harsh shadows.
**Audio:**
Natural room tone (minimal). Voice clear, no background music (hook punch needs silence).
```
---
### OUTPUT: `veo-prompts/scene-2.md`
```markdown
## Scene 2 — BODY 1
**Scene description:**
Same setting sebagai Scene 1 (home learning space, wooden desk, cream wall, monstera plant). Talent posisi sama, tapi gesture lebih expansive — hands move off desk untuk emphasize "berbagai usia" comparison.
**Character:**
Wanita Indonesia 30 tahun, rambut hitam panjang di-ponytail, kacamata tipis, kulit medium-light, senyum natural, body type medium-slim, tinggi 162cm. Kaos casual cream + cardigan dusty pink loose fit, celana jeans light blue. (LOCKED)
**Setting:**
Identical to Scene 1 — home learning space, small wooden desk, notebook open, monstera plant kanan belakang, buku anak-anak rak blur, cream wall ABC poster. Natural daylight consistent.
**Dialogue:**
"Anak sekolah konvensional ketemu 30 anak seumur mereka, tapi interaksi cuma di kelas. Anak homeschool? Mereka berinteraksi sama berbagai usia — adik, kakak, tetangga, mentor — dalam konteks yang lebih natural."
**Action:**
Talent hands lift off desk pada "Anak homeschool?" (questioning gesture). Hands gesture outward (expand motion) pada "berbagai usia". Return hands to desk on "lebih natural". Facial expression shifts from neutral-compare (first half) to warm-affirmative (second half).
**Camera:**
Medium close-up, eye level, static. Cut opportunity: bisa insert 2-3 detik B-roll montage (playground mix-age kids, family gathering) di tengah dialogue, lalu cut back ke talent untuk "lebih natural" closing line.
**Lighting:**
Same as Scene 1 — natural window light camera-left, warm neutral, soft diffused.
**Audio:**
Voice clear throughout. Optional soft ambient piano backsound (low volume) mulai fade-in di detik ke-8 reel (after hook).
```
---
### OUTPUT: `veo-prompts/scene-3.md`
```markdown
## Scene 3 — BODY 2
**Scene description:**
Same home learning space setting. Talent posisi duduk sama, tapi gesture lebih emphatic — hands frame "intentional" concept dengan precision gesture (fingertips together).
**Character:**
Wanita Indonesia 30 tahun, rambut hitam panjang di-ponytail, kacamata tipis, kulit medium-light, senyum natural, body type medium-slim, tinggi 162cm. Kaos casual cream + cardigan dusty pink loose fit, celana jeans light blue. (LOCKED)
**Setting:**
Identical — home learning space, wooden desk, notebook, monstera plant, buku anak-anak blur background, cream wall ABC poster. Consistent natural daylight.
**Dialogue:**
"Homeschool ngajarin anak sosialisasi yang intentional, bukan sekedar 'ada di ruangan yang sama'. Mereka belajar komunikasi lintas usia, empati, dan resolusi konflik yang lebih organik."
**Action:**
Hands gesture pada "intentional" — fingertips together, forming precision frame. Hands separate slightly on "bukan sekedar" (dismissive micro-gesture). Return to desk rest on "lebih organik". Slight head tilt forward on "empati" (warm emphasis).
**Camera:**
Medium close-up, eye level, static. Cut opportunity: insert 2-3s B-roll (kid helping younger sibling, community activity) mid-scene, cut back to talent for "lebih organik" close.
**Lighting:**
Consistent natural window light camera-left, warm neutral tone.
**Audio:**
Voice clear. Ambient piano backsound (same as Scene 2) continues at low volume, consistent throughout.
```
---
### OUTPUT: `veo-prompts/scene-4.md`
```markdown
## Scene 4 — CTA
**Scene description:**
Same home learning space. Talent slight posture shift — leans back sedikit (relaxed confident close), hands visible di desk edge ready untuk point-down gesture.
**Character:**
Wanita Indonesia 30 tahun, rambut hitam panjang di-ponytail, kacamata tipis, kulit medium-light, senyum natural, body type medium-slim, tinggi 162cm. Kaos casual cream + cardigan dusty pink loose fit, celana jeans light blue. (LOCKED)
**Setting:**
Same — home learning space, wooden desk, monstera plant, buku anak-anak blur, cream wall ABC poster. Natural daylight consistent.
**Dialogue:**
"Mau tau cara struktur sosialisasi buat anak homeschool? Komen 'PANDUAN' di bawah, gue kirim PDF lengkapnya."
**Action:**
Talent points down (off-screen, gesture toward comment section) on "Komen 'PANDUAN'". Warm smile broadens slightly. Eye contact camera maintained throughout. Hand returns to rest after point.
**Camera:**
Medium close-up, eye level, static. Overlay text appears on-screen (not in video itself, added in post): "Komen PANDUAN 👇" positioned lower third, brand color `#F0EA42`.
**Lighting:**
Consistent natural light camera-left, warm neutral.
**Audio:**
Voice clear, emphasis slightly warmer on "gue kirim PDF lengkapnya" (friendly promise tone). Backsound fades out di detik terakhir (clean ending).
```
---
## PHASE E: TRENDING SOUND RECOMMENDATION
**Input:**
- `library/sounds-trending.json` — **ABSENT di input ini**
- Content type: `educational`
- Vibe: calm, approachable, warm parent-to-parent
**Fallback logic (karena library kosong):**
- Gue kasih **generic trending sound rec** based on educational content best practices
- No specific URLs (karena data absent) — Hans cari manual di IG Reels trending tab
---
### OUTPUT: `trending-sound-rec.md`
```markdown
# Trending Sound Recommendations — Homeschool Myth Reel
**Slot:** 2026-05-15-homeschool-myth-reel
**Content type:** Educational
**Vibe target:** Calm, approachable, warm
---
## Recommendation Strategy
Karena `library/sounds-trending.json` absent, recommendation ini **generic best practice** buat educational reels:
1. **Soft instrumental piano/acoustic** (chill vibe)
- Fit: Educational content butuh backsound yang gak compete sama voice
- Target: Cari "chill study beats" atau "soft piano instrumental" di IG Reels trending
- Placement: Fade in detik ke-4 (after hook), low volume (20-30%), fade out detik ke-35
- **Action:** Search IG Reels → Audio tab → Filter "Trending" → "Instrumental" → Pick yang vibe-nya calm
2. **Lofi ambient (coffeeshop vibe)**
- Fit: Parent-friendly, non-distracting, modern warm
- Signal: Lofi tracks biasanya punya high save rate di educational niche
- **Action:** Search "lofi parenting" atau "coffeeshop ambient" di IG Audio library
3. **Original sound (no music)** — FALLBACK kalau audio risky
- Fit: Kalo gak nemu yang cocok, pure voice > wrong music
- Advantage: Voice clarity 100%, no copyright risk, authentic UGC feel
- Disadvantage: Less discoverable via audio trend
---
## Selection Priority (when library data available)
1. **Vibe match** (calm instrumental > upbeat > vocal track)
2. **Surging rank** (#1-20 trending, fresh <5 hari)
3. **Competitor signal** (≥2 parenting creators using = validation)
4. **IG compatible** (no copyright flag, available di IG Audio library)
---
## Manual Action for Hans
1. Open IG Reels camera
2. Tap "Add Music" (top audio icon)
3. Browse "Trending" tab
4. Filter by vibe: "Calm", "Chill", "Study"
5. Preview 3-5 tracks
6. Pick yang paling fit dialogue energy (gak terlalu slow, gak terlalu upbeat)
7. Set volume 25-30% (voice harus dominan)
8. Place audio start di detik ke-4 (setelah hook punch selesai)
---
**Note:** Begitu `library/sounds-trending.json` available dengan real scrape data (artist, title, URL, surging rank, competitor usage), gue bisa kasih **specific top 3 picks with scores**. Sementara ini, follow manual strategy di atas.
```
---
## COST BREAKDOWN (per reel)
| Phase | Action | Est. Cost (IDR) |
|-------|--------|----------------|
| A | Script generation (LLM 1 call, ~600 tokens out) | Rp 250-400 |
| B | Setting prompt JSON (LLM 1 call, ~900 tokens out) | Rp 350-500 |
| C | KIE multimodal image gen (1 request, 9:16, 2K) | Rp 1.800-3.000 |
| D | VEO prompts × 4 scenes (vision LLM, ~1.2K tokens/scene) | Rp 1.500-2.500 |
| E | Sound rec (no LLM, fallback logic) | Rp 0 |
| **TOTAL per reel** | | **Rp 3.900-6.400** |
**Bima = most expensive employee** (karena multimodal image gen + vision calls). Budget-conscious alternative: skip Phase C image gen, langsung ke Phase D pakai avatar URL only (quality drop, tapi save ~Rp 2K).
---
## OUTPUT SUMMARY
**Generated files (demo slot):**
```
content/reels/2026-05-15-homeschool-myth-reel/
├── script.md ✅ 118 words, 4 scenes, 38s
├── script.json ✅ Machine-parseable
├── setting-prompt.json ✅ Locked subject, 9-section schema
├── setting-image.png ⏳ (server-side KIE job)
├── veo-prompts/
│ ├── scene-1.md ✅ 8-section VEO (HOOK)
│ ├── scene-2.md ✅ 8-section VEO (BODY 1)
│ ├── scene-3.md ✅ 8-section VEO (BODY 2)
│ └── scene-4.md ✅ 8-section VEO (CTA)
└── trending-sound-rec.md ✅ Generic strategy (library absent)
```
---
## NEXT STEPS (for Hans / orchestrator)
1. **Supply missing anchors:**
- `avatars/avatars-index.json` dengan actual avatar pool Parentia
- `strategy/weekly-plan.json` dengan weekly reel slots
- `library/sounds-trending.json` (optional, tapi boost Phase E quality)
2. **Trigger production:**
- Daily batch: `production_service.run_daily_production('parentia')`
- Manual single: POST `/brands/parentia/production/reel/{slot_id}`
3. **Server-side image gen:**
- Configure KIE API credentials di dashboard
- Server auto-handle Phase C when Bima run
4. **Verify outputs:**
- Check `setting-prompt.json` → subject.description locked? ✅
- Check VEO prompts → 8 sections complete? ✅
- Check dialogue → 20-25 kata/scene Indonesian? ✅
5. **Post-production:**
- Wira generate `caption.md` (separate flow)
- Hans/editor render VEO → compile 4 scenes → add overlay text → apply trending sound → export
---
## FAILURE MODE NOTES
**Kalau KIE quota exceeded (Phase C):**
- Server mark `setting-image-error.txt`
- Bima Phase D tetap jalan, tapi pakai avatar URL fallback (lower VEO quality — subject might drift karena gak ada multimodal anchor)
**Kalau VEO prompt gen fails per scene:**
- Retry × 2
- Kalau masih fail → `scene-N-error.md` dengan raw error log
- Continue next scene (partial output better than no output)
**Kalau trending sound library kosong:**
- Bima kasih generic strategy (kayak di atas)
- Hans manual search IG Reels trending tab
---
**Bima done.** Output ready buat Hans compile. Kalau lu butuh production buat **slot specific real** (bukan demo), supply `weekly-plan.json` + slot detail, gue re-run with actual data.