R54 — Creative Cloner

Clone viral videos with your product swapped in

Category: Content Creation Runtime: Claude Code AIOS Rating: 8/10

---

What It Does

Give it any viral video and your product images — it analyzes the video scene-by-scene using the SEALCaM framework, then recreates each scene with your product swapped in. Generates images, animates them into video clips, adds AI-generated music, and combines everything into a final video.

The pipeline has 4 mandatory human checkpoints so you approve every stage before spending money.

Architecture

Reference Video       Gemini AI analyzes        Scene
+ Product Images  →   each scene using      →   Breakdown
(inputs/ folder)      SEALCaM framework         (YAML)

  CHECKPOINT 1: "Does this scene breakdown look right?"

Agent writes          Prompts logged             Cost
image + video    →    to Airtable           →    Estimate
prompts per scene     (Scenes table)             Shown

  CHECKPOINT 2: "Ready to generate images? Cost: $X"

NanoBanana Pro        Images logged              Images
generates start  →    to Airtable           →    Shown to
image per scene       (start_image field)        User

  CHECKPOINT 3: "Ready to generate videos? Cost: $X"

Kling 2.6 animates    Videos logged              Videos
each image into  →    to Airtable           →    Shown to
5s video clip         (scene_video field)        User

  CHECKPOINT 4: "Ready to add music and combine?"

Suno V4 generates     FFmpeg concatenates        Final
background music →    all clips + adds      →    Video in
via Kie.ai            music with fade-out        outputs/

Tech Stack

Platform	Role	Cost
Kie.ai	Core engine — NanoBanana Pro (images), Kling 2.6 (video), Suno V4 (music)	Per-generation
Google Gemini	Video analysis — SEALCaM scene-by-scene breakdown	Free tier
Airtable	Scene logging — prompts, images, videos per scene	Free tier
FFmpeg	Post-production — concatenate clips + overlay music	Free (local)

Workflows

1. Analyze Reference Video

• Upload video to Gemini 2.0 Flash
• AI analyzes using SEALCaM (Subject, Environment, Action, Lighting, Camera, Metatokens)
• CHECKPOINT: Show scene breakdown, get approval

2. Write Prompts & Log to Airtable

• Agent writes image + video motion prompt per scene with your product
• Create records in Airtable Scenes table
• CHECKPOINT: Show prompts + cost estimate

3. Generate Images

• Upload reference images to Kie.ai (temporary 3-day URLs)
• NanoBanana Pro generates start image per scene (9:16, 2K)
• CHECKPOINT: Show images, get approval

4. Generate Videos

• Each start image animated into 5s video via Kling 2.6
• CHECKPOINT: Show videos, get approval

5. Add Music & Combine

• Suno V4 generates instrumental background music
• FFmpeg concatenates all clips + overlays music with 2-second fade-out
• Final video saved to outputs/

Project Structure

creative-cloner/
├── Getting_Started.md
├── .agent/
│   ├── .env
│   ├── requirements.txt
│   └── skills/creative-cloner/SKILL.md  (266 lines — brain)
├── tools/
│   ├── analyze_video.py
│   ├── generate_images.py
│   ├── generate_videos.py
│   ├── generate_music.py
│   └── combine_all.py
├── inputs/
│   ├── project1 - animals/          (animals.mp4 + tea.png)
│   └── project2 - surf/            (surf.mp4 + redboots.jpg + logo.png)
└── outputs/

Environment Variables

Variable	Source	Purpose
KIE_API_KEY	kie.ai dashboard	Image + Video + Music gen + File upload
GEMINI_API_KEY	aistudio.google.com/apikey	Video analysis with Gemini 2.0 Flash
AIRTABLE_TOKEN	airtable.com/create/tokens	Read/write scene records
AIRTABLE_BASE_ID	Airtable URL	Target base

Cost Estimates

Generation	Model	Cost	Time
Image (per scene)	NanoBanana Pro	$0.09	~30-60s
Video 5s (per scene)	Kling 2.6	$0.28	~2-4 min
Video 10s (per scene)	Kling 2.6	$0.56	~3-6 min
Music (per project)	Suno V4	~$0.10	~1-3 min

Example: 4 scenes with 5s videos = $1.58 (4 images $0.36 + 4 videos $1.12 + 1 music $0.10)

Key Rules

1. Stop at Every Checkpoint — NEVER auto-proceed between phases
2. Show Costs Before Generating — Always show per-unit + total before any generation
3. One Script at a Time — Run sequentially, never in parallel
4. Log Everything to Airtable — Every prompt, image, and video gets logged
5. NEVER Auto-Retry Errors — On failure, STOP and give user 3 options

Common Pitfalls

Problem	Cause	Fix
Kie.ai error but generation succeeded	Timeout/network issues	Check kie.ai dashboard before retrying
Bad scene breakdown	Low quality reference video	Use clear 5-15 second reference videos
Airtable logging fails	Wrong column names	Column names are case-sensitive

---

Built by Copyweb — architecting how marketing teams operate in an AI-first world.