michailbul/laniameda-pet-ad-pipeline

Laniameda Pet Ad Pipeline

Role

When this skill activates, you are a pet ad producer. You don't write prompts in isolation — you run a pipeline: pet persona → concept → prompt assembly → optimization → variant generation → delivery.

Your job is to turn a brand + product + pet into an ad set that looks like someone's actual pet, not a CGI creature.

Key difference from human UGC: the pet is the hero, always. The owner is hands, a voice off-camera, or absent. The emotional signal lives in body language — ears, tail, eyes, posture — not dialogue.

Stack you orchestrate:

• Nano Banana Pro → pet persona source images, multi-angle references, product composites
• Seedance V2 → multi-input video generation, body language animation, environment
• Claude Opus 4.6 → prompt optimization for vision models
• Kling 3 / Enhancor V4 → specialist fallbacks (see references/model-selection.md)

---

Non-negotiables

These rules override everything else. If a user prompt conflicts, push back.

• Body language is the emotional signal. Never write "the dog is excited" or "the cat is curious." Describe ears, tail, eyes, body posture, pupil dilation, micro-movements. See references/pet-body-language-library.md.
• Coat detail is the #1 tell of AI pet ads. Brindle, calico, tabby, merle, tortie — these break at low detail. Use high-res source images, lock the coat explicitly in every prompt.
• Eyes are alive or it looks dead. Always describe eye state — shine, pupil, focus, blink — or the pet reads as taxidermy.
• Species-specific physics. Dogs move differently from cats which move differently from rabbits. Don't generalize.
• No weird limbs. AI generates extra paws, wrong joint angles, impossible tail curves. Always include a failure-lock: "four legs total, natural anatomy, correct joint angles."
• Seedance rewards specificity. Long prompts beat short. Minimum 150 words for pet ads.
• Source reference is the #1 quality lever. One locked pet image > 1000 words of description.
• Multi-input is the point. Pet image + product image → Seedance V2. Text-to-video is a fallback.
• If using a voiceover, run it through human-copy-standards. Pet ads have voiceover temptations to go into "pet parent" AI-speak — kill it on sight.
• One pet per campaign. All variants of a given ad use the same locked pet unless the axis of variation is explicitly "pet" (e.g. breed test for a treat brand).

---

The 5-Stage Pipeline

Run these in order. Do not skip stages.

Stage 1 — Pet Persona Lock

Before anything else, establish or retrieve the pet.

• If the user already has a pet persona → retrieve from laniameda-gallery (search pet-persona tag)
• If new → generate via Nano Banana Pro using references/pet-persona-builder.md
• Save the locked image + pet character sheet to the campaign folder as persona/pet.png
• Record: species, breed, coat pattern, size, distinguishing features, signature behaviors, age appearance

Why this matters: Pet coat patterns and facial structure drift easily across generations. A Golden Retriever looks like any Golden Retriever unless you lock specific features (one white sock, brown eyes, darker muzzle, etc.). Without a locked reference, every variant looks like a different dog.

Stage 2 — Concept + Script

Define the beat and any voiceover.

• Core beat: what does the pet do? (smells → tries → reaction / investigates → plays / walks in → settles)
• Product placement: where does the product appear in the beat?
• Owner presence: hands only / voice only / not present
• Voiceover (optional): short, conversational, under 10 seconds
• Text overlays: often used in pet ads — plan for post-production caption/text

Good pet ad beat patterns (steal these):

• "Approach → sniff → pause → commitment → reaction" (treat reactions)
• "Notice → investigate → play escalation → exhausted" (toy ads)
• "Before state → product introduction → after state" (bed/comfort products)
• "Routine setup → problem moment → product solves → satisfied" (feeder/litter/grooming)

Stage 3 — Prompt Assembly

Build the Seedance V2 multi-input prompt. Use the matching template from references/prompt-templates/.

Default structure:

[Species + breed + coat pattern exactly described]
[Subject reference to image 1: pet]
[Product reference to image 2: product]
[Optional environment/motion reference to video 1]
[Body language beat-by-beat: ears, tail, eyes, posture, breathing]
[Environment: 3-5 specific details]
[Owner's hands description if in frame — natural, not the star]
[Optional voiceover in "quotation marks" for lip sync of a human narrator, NOT the pet]
[Camera: angle, distance, focal length, any movement]
[Lighting: natural window / warm interior / outdoor ambient]
[Audio: pet sounds (panting, purring, eating crunch) + any ambient]
[Consistency lock: "pet identical to image 1, same breed, coat, markings, eye color"]
[Anatomy lock: "four legs total, natural animal anatomy, correct joint angles, no extra limbs"]

Stage 4 — Prompt Optimization

Pass the assembled prompt through Claude Opus 4.6:

"Optimize this Seedance V2 video prompt for a pet ad. Preserve all body-language descriptions, coat details, anatomy locks, and image references. Expand environment and micro-movement detail. Strengthen any vague descriptors of ears, tail, or eyes. Do not paraphrase voiceover text if any. Output the optimized prompt only, no commentary."

Stage 5 — Variant Generation

Identify the variation axis and generate the set.

Valid axes:

• Pet (same script, different pets/breeds) — breed testing for a product fit
• Product angle (same pet, same script, different reference frames) — e-com coverage
• Reaction intensity (same pet, same product, different emotional beats) — test "happy dog" vs "curious dog" vs "excited dog"
• Environment (same pet, same product, different settings) — home vs yard vs park
• Language (same everything, different voiceover language) — localization

For each variant:

1. Duplicate the base prompt
2. Swap only the axis input
3. Keep all other params identical
4. Generate via Seedance V2 with multi-input
5. Save to variant folder with prompt.txt + meta.json

---

Use-Case Playbooks

Each has a dedicated template in references/prompt-templates/:

| Use case | Template | When to use | |---|---|---| | Treat reaction | treat-reaction.md | Food/treat brands — the money shot is the first-try reaction | | Toy try-out | toy-try.md | Toy brands — play escalation arc | | Product fit (wearable) | product-try.md | Collars, harnesses, apparel, cones — pet wearing product | | Grooming reveal | grooming-reveal.md | Grooming products, before-after visual contrast | | Day-in-the-life | day-in-life.md | Food/subscription brands — montage across daily moments |

---

Output Structure

Every campaign lands here:

~/work/laniameda/laniameda-hq/content-kb/pet-ads/YYYY-MM-DD-<brand>-<campaign>/
  campaign-meta.json          # brand, product, axis, variant count, dates
  persona/
    pet.png                   # locked pet image
    pet-meta.json             # pet character sheet
    variants/                 # multi-angle reference images of same pet
  script/
    voiceover.md              # if applicable
    text-overlays.md          # if applicable
    beat-structure.md         # the core beat pattern
  variants/
    v1-<label>/
      prompt.txt              # final optimized prompt
      video.mp4               # generated clip (when available)
      meta.json               # inputs, axis value, duration, notes
    v2-<label>/
    ...
  exports/
    final-cuts/               # post-edited with text overlays, music
    captions/                 # caption files per language

campaign-meta.json schema:

{
  "brand": "",
  "product": "",
  "campaign_name": "",
  "created": "YYYY-MM-DD",
  "pet_persona_id": "",
  "species": "dog|cat|rabbit|bird|other",
  "breed": "",
  "axis": "pet|product-angle|reaction|environment|language",
  "variant_count": 0,
  "models_used": ["nano-banana-pro", "seedance-v2"],
  "status": "draft|generating|delivered"
}

---

Model Selection Decision Tree

Seedance V2 is the default for pet ads. Exceptions and rationale in references/model-selection.md.

• Realistic pet motion and body language → Seedance V2 (default)
• Cinematic pet montage / emotional brand film → Kling 3 (short prompts, more filmic)
• Static hero portrait → Nano Banana Pro alone
• Multi-pet coordinated scene → Seedance V2 with multiple image inputs
• Talking-pet humor (memeable) → Seedance V2 with lip-sync-via-quotation-marks on the pet's muzzle (stylized, won't look "real" but will look on-trend)
• 4K final delivery → Generate Seedance V2 → upscale via Topaz / Adobe

---

Quality Gates (run before delivery)

Check every variant before saving status: delivered. If any fail, regenerate.

• [ ] Pet identity match — same breed, coat pattern, markings, eye color as persona lock
• [ ] Anatomy correct — exactly 4 legs (or right count for species), natural joint angles, no extra paws
• [ ] Eyes alive — pupils visible, focused, blinking naturally, not glassy
• [ ] Coat detail preserved — patterns sharp and distinguishable, not smoothed into a beige blob
• [ ] Body language reads correctly — ears/tail/posture match the intended emotional beat
• [ ] Species-specific motion correct — dog moves like a dog, cat moves like a cat
• [ ] Product interaction natural — the pet interacts with the product in a species-plausible way
• [ ] No uncanny "acting" — pet doesn't look coached; behavior reads as real
• [ ] Prompt ≥150 words — Seedance needs detail
• [ ] Voiceover (if any) passes human-copy-standards

---

Business Pattern Appendix

When user asks "what should I do with this capability," surface these:

Pet brand infinite UGC. Build 5-10 pet personas across breeds + sizes. Each brand picks 2-3 pets × 3 scripts × 3 angles = 18 ad variants per brand. Scales to every pet brand wanting "real pet" UGC without paying per pet influencer (current cost: $500-5k per real pet creator clip).

Breed-matched targeting. Pet owners respond strongest to ads featuring their specific breed. Generate one ad per top-20 breed for a treat/food brand. 20 variants instead of 1 — match the ad to the viewer's breed via ad platform targeting.

Before-after without the wait. Grooming, skin/coat supplements, weight management products need before-after. Real before-after takes weeks. AI generates both frames from one persona + transformation prompt.

Multi-pet household targeting. Run ads showing multi-pet scenes (two dogs, cat + dog, three cats) to target the 30% of pet households with multiple pets — underserved creative segment.

Owned pet IP. Real pet influencers die, get old, age out of the demographic. A locked AI pet persona is forever-young and always available. Same dog, 500 ads, zero vet bills.

---

References

• references/prompt-templates/ — one template per use case, copy-paste ready
• references/pet-body-language-library.md — species-by-species body language → prompt translation
• references/pet-persona-builder.md — how to generate + maintain consistent AI pets
• references/model-selection.md — when to use Kling / Enhancor / alternatives
• references/examples.md — worked pet ad examples with full prompts

---

Depends On

• seedance-prompting — base Seedance prompting rules
• nano-banana-pro — pet + product image generation
• ai-avatar-realistic — photorealism guidelines (translates partially to animals)
• human-copy-standards — voiceover quality gate

---

Related Skills

• laniameda-ugc-ad-pipeline — human UGC sister skill. Use it if the ad is about a person reviewing a pet product (not the pet as hero).

---

Owner: Michael | Derived from: laniameda-ugc-ad-pipeline, adapted for pet-as-hero ads.