/product-reel-generator

name: product-reel-generator
description: Generates Instagram-ready product reels from any e-commerce product page URL. Scrapes product images, classifies by type, generates AI-animated clips via Higgsfield API, creates text overlays with style presets, and composes a 15-20 second reel with music. Supports model-based and product-only reels.
user-invocable: true
allowed-tools: Bash, Read, Write, Edit, Grep, Glob, WebSearch
argument-hint: [product-page-url]

Product Reel Generator

You are a video production skill that takes an e-commerce product page URL and produces an Instagram-ready reel. The reel features AI-animated model clips (or Ken Burns product showcases), text overlays, and background music.

---

Requirements

• **FFmpeg** installed and available in PATH (`brew install ffmpeg` on macOS, `apt install ffmpeg` on Linux)
• **Python 3** with `Pillow` and `python-dotenv` packages (`pip install Pillow python-dotenv`)
• **Higgsfield API credentials** — `HIGGSFIELD_API_KEY_ID` and `HIGGSFIELD_API_KEY_SECRET` in a `.env` file (project root or any parent directory)

**Before starting:** Verify dependencies are available. If FFmpeg or Python packages are missing, instruct the user to install them before proceeding.

---

Input

The user provides: 1. **Product page URL** (required) — any e-commerce product page (Shopify, Zara, DaMENSCH, etc.) 2. **Music file** (optional) — path to an MP3. If not provided, use a royalty-free track. 3. **Style preset** (optional) — one of: `minimal`, `luxury`, `bold`, `editorial`, `clean`. Defaults to auto-detect based on brand. 4. **Brand name** (optional) — for watermark. If not provided, extract from the page.

---

Pipeline

Step 1: Scrape Product Images

Try these methods in order until one works:

1. **Shopify JSON** — append `.json` to the product URL and extract images from the response 2. **HTML scraping with referrer** — `curl` with `-H "Referer: <site-domain>"` and a browser user-agent 3. **Chrome DevTools** — navigate to page, extract image URLs via JavaScript, download

For each image, download at the highest available resolution.

Step 2: Classify Images (Heuristic)

Use image position on the product page as the primary signal:

| Position | Likely Type | Use In Reel | |----------|-------------|------------| | Image 1 (first on page) | Hero / front-facing model | Walk forward (AI) | | Image 2 | Alternate angle (side/back) | Turn or side walk (AI) | | Image 3-4 | Close-up or detail | Detail insert (Ken Burns) | | Last image | Size guide or back view | Back turn (AI) or product card |

**Model detection heuristic:** If image height > 1.5× width AND file size > 100KB → likely a model photo → use AI animation pipeline. Otherwise → product-only → use Ken Burns pipeline.

Step 3: Generate AI Video Clips

Use the Higgsfield API via this skill's `scripts/higgsfield_video.py` script or direct `curl` calls.

**API details:**

• Base URL: `https://platform.higgsfield.ai`
• Auth header: `Authorization: Key {HIGGSFIELD_API_KEY_ID}:{HIGGSFIELD_API_KEY_SECRET}`
• Always set `"aspect_ratio": "9:16"` for Instagram Reels

**Model selection:**

• **Seedance** (`bytedance/seedance/v1/pro/image-to-video`) — for hero/walk scenes. Higher quality, ~45 credits. Use for the most important clip.
• **Kling** (`kling-video/v2.1/pro/image-to-video`) — for secondary scenes. Good quality, ~6 credits. Use for turns, side angles.

**Prompt guidelines:**

• Always mention the clothing color and type in the prompt
• Specify direction of motion ("walks forward", "turns from front to side")
• Add "clean studio background" or describe the actual background
• Add "smooth cinematic motion" at the end
• For walk prompts, add "does not stop or turn around" to prevent reversal

**Duration:** Use `"duration": 5` for each clip. Kling only supports 5 or 10.

**Polling:** After submission, poll `GET /requests/{request_id}/status` every 15 seconds until `status: "completed"`. Then download the video from `response.video.url`.

Step 4: Create Ken Burns Scenes

For detail/texture shots where AI animation adds no value, use FFmpeg Ken Burns:

ffmpeg -y -loop 1 -i "detail.jpg" \
  -vf "scale=2160:3840,zoompan=z='1+0.06*in/75':x='iw/2-(iw/zoom/2)':y='ih/2-(ih/zoom/2)':d=75:s=1080x1920:fps=25" \
  -t 3 -c:v libx264 -pix_fmt yuv420p -r 25 "scene-detail.mp4"

Vary the zoom type: zoom-in, zoom-out, pan-left, pan-right, pan-up, pan-down.

Step 5: Create Text Overlays

Use Python Pillow to generate transparent PNG overlays, then composite with FFmpeg.

**IMPORTANT:** Many FFmpeg installations do NOT have the `drawtext` filter. Always use Pillow to create PNG text images, then overlay with:

ffmpeg -y -i video.mp4 -loop 1 -t <duration> -i overlay.png \
  -filter_complex "[1:v]format=rgba[txt];[0:v][txt]overlay=0:0" \
  -t <duration> -c:v libx264 -pix_fmt yuv420p -r 25 output.mp4

Style Presets

Fonts are provided as shared files in the pack's `fonts/` directory (copied into each skill on install). Fall back to system fonts if custom fonts are not found.

| Preset | Title Font | Body Font | Text Color | Treatment | |--------|-----------|-----------|------------|-----------| | **minimal** | Montserrat-Light.ttf | Montserrat-Light.ttf | White (255,255,255) | No background, subtle shadow | | **luxury** | System Didot (/System/Library/Fonts/Supplemental/Didot.ttc) | Cormorant-Regular.ttf | Cream (245,235,210) | Thin gold stroke | | **bold** | System Futura (/System/Library/Fonts/Supplemental/Futura.ttc) | Montserrat-Bold.ttf | White | Dark backdrop bar, uppercase | | **editorial** | Cormorant-Italic.ttf | Cormorant-Regular.ttf | White | Minimal, italic titles | | **clean** | System Helvetica (/System/Library/Fonts/Helvetica.ttc) | System Helvetica | White | Simple shadow, professional |

**Overlays to create:** 1. **Brand watermark** — small text, top-right corner, persistent on all video scenes 2. **Product info card** — product name, price, CT