heymegabyte/agent-skills / 12-media-orchestration
media-orchestration
VerifiedSection-by-section media planning and generation. Image generation (GPT Image 2 primary, built-in fallback), logo/icon generation (Ideogram v3 → favicon set), video generation (Veo 3.1), social preview images (OG 1200x630 + AI search optimization), stock photo curation (Pexels, Pixabay), critique/remix loops (max 3 rounds), asset compression pipeline, and media performance budgets.
| File | Kind | Size |
|---|---|---|
| 30-ideogram-methods.md | markdown | 16.0 KB |
| SKILL.md | skill md | 6.7 KB |
| build-breaking-rules.md | markdown | 48.2 KB |
| compression-pipeline.md | markdown | 3.8 KB |
| image-optimization.md | markdown | 6.6 KB |
| image-profiling.md | markdown | 3.2 KB |
| lightbox-classifier.md | markdown | 10.0 KB |
| media-prompts.md | markdown | 3.8 KB |
| notebooklm-pipeline.md | markdown | 22.1 KB |
| og-card-pipeline.md | markdown | 19.5 KB |
| og-image-generation.md | markdown | 8.5 KB |
| social-brand-hex.md | markdown | 3.8 KB |
| technical-diagramming.md | markdown | 4.4 KB |
| templates/PROMPTS.md | markdown | 10.0 KB |
image-profiling.md
View on GitHub---
name: "image-profiling"
description: "Tiered vision profiling: Workers AI Llama Vision (free, bulk) → GPT Image 2 vision (paid, hero/logo picks only). Scores, placement, alt text, colors."
updated: "2026-04-25"
---
# Image Profiling (Tiered Vision)
Profile every candidate image BEFORE the build so the builder makes informed placement decisions without seeing images.
## Architecture (***COST-TIERED***)
### Tier 1 — Workers AI Llama Vision (FREE, 90% of images)
- Model: `@cf/meta/llama-3.2-11b-vision-instruct`
- Batch 5 images/call, 3 batches parallel = 15 images/round
- Handles: description, keywords, quality_score, suggested_placement, alt_text, dominant_colors
- Sufficient for gallery/about/services/background images
### Tier 2 — GPT Image 2 vision detail:low ($0.01/call, 10% of images)
- Hero candidates only (top 5 by Tier 1 score) + logo variants + brand color extraction
- Single batch call with all hero candidates
- Worth the spend — hero is 80% of first impression
## Profile Schema
```json
{
"name": "hero-storefront.webp",
"url": "https://r2.example.com/assets/hero-storefront.webp",
"source": "discovered|generated|scraped|uploaded",
"description": "Warm interior shot of a busy coffee shop with exposed brick walls",
"keywords": ["coffee", "interior", "warm", "cozy", "brick"],
"quality_score": 8,
"relevance_score": 9,
"suggested_placement": "hero|about|services|gallery|team|testimonials|background",
"alt_text": "Interior of Main Street Coffee with customers at wooden tables",
"dominant_colors": ["#8B4513", "#F5F5DC", "#2F4F4F"]
}
```
## System Prompt Pattern
```
You are an expert visual curator for professional websites. For each image:
1. Describe what's in it (2 sentences max)
2. Rate quality 1-10 (composition, lighting, resolution, professionalism)
3. Rate relevance 1-10 (how well it fits a {business_type} website)
4. Suggest ONE placement: hero|about|services|gallery|team|testimonials|background
5. Write descriptive alt text (SEO-friendly, includes business context)
6. Extract 3-5 dominant hex colors
Return JSON array matching the profile schema.
```
## Top-Pick Selection Algorithm
1. Sort by `quality_score * 0.4 + relevance_score * 0.6` descending
2. **Hero** — highest combined score, prefer wide/landscape, quality ≥ 7
3. **Logo** — source=uploaded|discovered with "logo" in name/description, else generated
4. **About** — top 3 by relevance with "interior"|"team"|"story" keywords
5. **Services** — top 3 matching service-related keywords
6. **Gallery** — remaining images quality ≥ 6, deduplicated by dominant_colors similarity
## Integration with Build Pipelines
Pre-container: collect 50-100 candidate images from all APIs → batch profile → select top picks → write `_image_profiles.json`. Builder reads profiles, uses every top-pick in its suggested placement. Alt text pre-written; no vision needed in build step.
## Cost Management
- **Workers AI** — FREE (included in Workers Paid plan). 60 images = $0.00.
- **GPT Image 2 vision** — 1 batch call for top 5 hero candidates ≈ $0.02.
- **Total profiling cost** — ~$0.02/site (down from $0.60-1.20).
- Skip duplicates (hash-based dedup before profiling)
- Skip images <100px or >5MB
- Timeout: 30s per batch call, 90s total phase
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deployment | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy. | Stars: 119k | Updated | Claude CodeCodex | Verified | Install |
| childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release. | Stars: 118k | Updated | Claude CodeCodex | Verified | Install |