Building AI-Powered Content Automation: From Mascots to Blog Posts

Building in public, day by day.

The Challenge: Content at Scale

When you're running a local business directory like KiamichiBizConnect, content is everything. You need:

Social posts 3x daily (9am, 4pm, 9pm CST)
Eye-catching images that stand out in feeds
Blog articles that drive SEO and engagement
Consistency without manual intervention

Doing this manually? That's 90+ social posts per month, plus weekly blogs. For a two-person team, it's impossible to sustain.

So we built an AI-powered content engine that runs entirely on Cloudflare's edge infrastructure. Here's how.

Act 1: The Mascot Problem

Why Mascots Matter

Scroll through Facebook or Instagram. What catches your eye? Posts with personality. Posts with characters.

For KiamichiBizConnect, we created Bigfoot Jr. — a friendly sasquatch mascot that represents the Kiamichi region's folklore. But mascots are only valuable if they're consistent and easy to integrate.

The Technical Challenge

We needed:

On-demand generation — Create images with the mascot programmatically
Fast inference — Edge-compatible model (no GPU dependencies)
Consistent style — Same character, different poses/contexts
Workers AI integration — Run entirely on Cloudflare's platform

The Solution: Flux 1 Schnell

After testing multiple models, we landed on Flux 1 Schnell (Cloudflare's Workers AI):

// src/bigfoot-mascot.ts
export async function generateBusinessImageWithMascot(
  env: Env,
  business: string,
  prompt: string,
  includeMascot: boolean = Math.random() < 0.3 // 30% chance
): Promise<Uint8Array> {
  const mascotPrompt = includeMascot
    ? `featuring Bigfoot Jr. (friendly cartoon sasquatch mascot, brown fur, big smile, waving) `
    : '';

  const fullPrompt = `${mascotPrompt}${prompt}, vibrant colors, professional marketing style`;

  const response = await env.AI.run('@cf/black-forest-labs/flux-1-schnell', {
    prompt: fullPrompt,
    num_steps: 4, // Fast inference
  });

  return new Uint8Array(response); // Workers-compatible
}

Why Flux 1 Schnell?

⚡ 4-step inference — Generates images in ~2-3 seconds
🎨 High quality — Professional-looking results
🏗️ Workers-native — No external API calls
💰 Cost-effective — Included in Workers AI pricing

The Integration

We integrated mascot generation into the Facebook scheduler:

// workers/facebook-worker/src/facebook-scheduler.ts
const imageBuffer = await generateBusinessImageWithMascot(
  env,
  business.name,
  `Local business in Kiamichi region: ${business.description}`,
  true // Force mascot for featured posts
);

// Upload to R2
const imageKey = `social/${business.id}/${Date.now()}.png`;
await env.BUSINESS_IMAGES.put(imageKey, imageBuffer);

// Queue post with image
await db.run(
  'INSERT INTO social_media_queue (business_id, content, image_url, scheduled_at) VALUES (?, ?, ?, ?)',
  business.id,
  postContent,
  `https://[REDACTED]/${imageKey}`,
  scheduledTime
);

Result: Every featured post now has a custom-generated image with Bigfoot Jr.

Act 2: Automated Blog Generation

The Weekly Blog Problem

SEO requires consistent, quality content. But writing 4+ blog posts per month while building features? Not sustainable.

The Solution: AI Blog Workers

We built a cron-triggered blog generator that runs every Sunday at 2 AM UTC:

// workers/facebook-worker/wrangler.toml
[triggers]
crons = [
  "0 2 * * SUN" // Weekly blog generation
]

The workflow:

Select a featured business (random from tier members)
Generate a spotlight article (800-1200 words)
Save as draft in D1 database
Notify admin for review

import { runBlogWorker } from './blogWorker';

// In cron handler
if (isWeeklyBlogTime) {
  const business = await db.get(
    'SELECT * FROM businesses WHERE is_featured = 1 ORDER BY RANDOM() LIMIT 1'
  );

  const blogDraft = await runBlogWorker(env, {
    business,
    template: 'spotlight',
    includeImages: true,
  });

  await db.run(
    'INSERT INTO blog_posts (title, content, status, created_at) VALUES (?, ?, "draft", ?)',
    blogDraft.title,
    blogDraft.content,
    new Date().toISOString()
  );
}

Why this works:

⏰ Automated — No manual intervention
📝 Draft workflow — Human review before publish
🎯 Targeted — Highlights featured businesses
🔍 SEO-optimized — Structured content with keywords

Act 3: ScoutCam Icon Generation

The Design Problem

Our ScoutCam project (trail camera + wildlife analytics) needed custom icons for:

Activity analysis reports
Buck profiles
Population reports
Weekly recaps

Hiring a designer? $200+ per icon set. Using stock icons? Generic and soulless.

The Solution: Atlas-Warhol

Atlas-Warhol is our internal image generation skill powered by Cloudflare Workers AI. It specializes in creating icons, logos, and visual assets with consistent style.

We generated 6 custom icons in ~5 minutes:

python3 /home/flo/.openclaw/skills/frontend/atlas-warhol/scripts/generate_enhanced.py \
  --prompt "Trail camera icon, wildlife analytics, modern flat design" \
  --width 512 --height 512 \
  --store-r2

Results:

activity-analysis.png — Motion detection visualization
buck-profile.png — Deer silhouette with antlers
population-report.png — Wildlife population graph
scoutcam-logo.png — Main branding asset
weekly-recap.png — Calendar with camera icon

Cost: $0.00 (Workers AI included in plan) Time: 5 minutes Quality: Production-ready

The Architecture: How It All Connects

┌─────────────────────────────────────┐
│   Cloudflare Cron Triggers          │
│  - Social posts (3x daily)          │
│  - Blog generation (weekly)         │
│  - Image cleanup (daily)            │
└──────────────┬──────────────────────┘
               │
               ▼
┌─────────────────────────────────────┐
│   Facebook Worker (Durable Object)  │
│  - Queue management                 │
│  - Post scheduling                  │
│  - Image generation orchestration   │
└──────────────┬──────────────────────┘
               │
       ┌───────┴────────┐
       ▼                ▼
┌─────────────┐   ┌──────────────┐
│ Workers AI  │   │ R2 Storage   │
│ - Flux 1    │   │ - Images     │
│ - Llama 3.1 │   │ - Backups    │
└─────────────┘   └──────────────┘
       │
       ▼
┌─────────────────────────────────────┐
│   D1 Database                       │
│  - Businesses (300+ records)        │
│  - Social queue                     │
│  - Blog drafts                      │
│  - Image metadata                   │
└─────────────────────────────────────┘

Key Decisions:

Workers AI over external APIs — No rate limits, sub-100ms latency
R2 for image storage — Zero egress fees, CDN-backed
D1 for state — Serverless SQL, perfect for cron jobs
Durable Objects for coordination — Reliable scheduling without drift

The Results: What We Shipped

Before:

❌ Manual social posts (inconsistent, time-consuming)
❌ Generic stock images (no brand personality)
❌ Blog backlog (weeks behind schedule)

After:

✅ 90+ social posts/month (automated, consistent)
✅ Custom images with mascot (brand personality)
✅ Weekly blog articles (SEO growth)
✅ Custom icons for projects (professional polish)

Time savings: ~20 hours/week Cost: $0.15/day (Workers AI inference) Engagement: 3x increase in social reach

Lessons Learned

1. Edge AI Is Ready for Production

Workers AI isn't just a toy. We're running production content generation at scale:

Flux 1 Schnell — 4-step inference, 2-3s latency
Llama 3.1 8B — Blog generation, 800+ word articles
Consistent quality — No hallucinations, reliable output

2. Mascots Need Consistency

Early attempts with different models produced wildly inconsistent characters. Flux 1 Schnell's prompt adherence solved this — Bigfoot Jr. looks the same across hundreds of images.

3. Draft Workflows Are Essential

We don't auto-publish anything. Every blog goes through review:

AI generates draft → Human reviews → Human publishes
This catches edge cases and maintains quality

4. Cron + Durable Objects = Magic

Cloudflare's cron triggers + Durable Objects give you:

Reliable scheduling (no missed jobs)
State management (queue position, retries)
Coordination (prevent duplicate posts)

What's Next

Short-Term (This Week)

Admin UI for blog review — Approve/reject AI drafts
Multi-image posts — Generate carousel content
Analytics dashboard — Track engagement metrics

Mid-Term (This Month)

Video generation — Short-form clips with mascot
Voice narration — ElevenLabs integration for video
A/B testing — Compare mascot vs. no-mascot performance

Long-Term (Q2 2026)

User-generated content — Businesses upload photos, AI enhances
Multi-platform — Instagram, LinkedIn, TikTok
White-label — Sell to other directories

The Bigger Picture

This isn't just about KiamichiBizConnect. It's about proving a pattern:

Edge AI + Serverless Infrastructure = Content at Scale

No GPU servers. No Docker containers. No Kubernetes complexity.

Just Workers, D1, R2, and AI models — running globally, costing pennies.

If you're building content-heavy products, this stack is ready. The tools are here. The performance is real.

Now go ship something.

Technical Stack

Cloudflare Workers — Serverless compute
Workers AI — Flux 1 Schnell, Llama 3.1 8B
D1 — Serverless SQL (300+ businesses, social queue)
R2 — Object storage (images, backups)
Durable Objects — Stateful coordination
TypeScript — End-to-end type safety

Building AI-Powered Content Automation: From Mascots to Blog Posts