Slop — Open-Source Ad Creative Intelligence

100% free. 100% local. No paid APIs.

Replicates the full "Eddie" ad system from the viral thread using only open-source tools:

Paid tool	Free/Open-Source replacement
Apify ($49+/mo)	Meta Ad Library API (free token)
OpenAI Whisper API (pay per minute)	openai-whisper — runs locally on your CPU
GPT-4 / Claude API (pay per token)	Ollama — runs locally, zero cost, forever
Arcads ($500+/mo)	Script output → $15-50 UGC creators
Singular MMP ($$$)	SQLite + CSV import from any ad platform

Total cost: $0/month. You just need a computer.

---

How It Works

research → generate → run ads → import results → optimize → repeat
   │           │                      │               │
   ▼           ▼                      ▼               ▼
 Meta API   Ollama +              CSV import     Winners feed
 + Whisper  brain/*.md            into SQLite    next batch
 (free)     (free)                (free)         (free)

1. Research — Scrape Meta Ad Library for competitor video ads. Download videos, transcribe with local Whisper. Sorted oldest-first (older = more proven/profitable).
2. Generate — Analyze each competitor ad's angle → rewrite in your brand voice → multiply by ICP segments. All using local Ollama.
3. Run ads — Send scripts to UGC creators ($15-50/video) or use free AI video tools.
4. Import results — Drop your Meta/TikTok CSV export into slop import-results.
5. Optimize — System finds winners, identifies patterns, generates next batch. Self-improving loop.

---

Setup (5 minutes)

1. Install Ollama (the free LLM)

# macOS / Linux
curl -fsSL https://ollama.com/install.sh | sh
ollama pull llama3.1    # 8B model, runs on any modern machine

# Windows: download from https://ollama.com

That's it. No API keys. No accounts. Runs locally forever.

2. Install Slop

git clone <this-repo>
cd Slop
pip install -r requirements.txt
playwright install chromium   # optional: only for video download fallback

3. Configure

cp .env.example .env

Edit .env — only one thing to set:

# Free token from: https://developers.facebook.com/tools/explorer/
META_ACCESS_TOKEN=your_token_here

# Ollama is already the default — no config needed
OLLAMA_MODEL=llama3.1

4. Fill in your brain files

The brain/ folder is pre-filled for Prayer Lock as an example. Replace with your own product:

• brain/voice.md — How your brand talks (with real examples)
• brain/product.md — What your product does, features, social proof, price
• brain/icp.md — Customer segments with their exact words and pain points
• brain/writing-rules.md — Anti-AI-slop filter (already filled, customize as needed)

The more specific these files are, the better your scripts will be.

---

Usage

Step 1 — Research competitors

# Scrape + download + transcribe 50 competitor ads
python main.py research "prayer app" --limit 50

# Fast mode: metadata only, skip video/transcription
python main.py research "prayer app" --limit 50 --no-video

Step 2 — Generate scripts

# Generate scripts with 3 ICP variations each
python main.py generate "prayer app" --variations 3

Output: data/scripts/prayer_app_scripts.json

Step 3 — Run ads, then import results

Export CSV from Meta Ads Manager / TikTok Ads with columns: ad_name, spend, installs, cpa, ctr

python main.py import-results meta_export.csv

Step 4 — Self-improvement loop

# See performance rankings
python main.py analyze --keyword "prayer app"

# Generate next batch based on winners
python main.py optimize "prayer app" --count 30

View scripts

python main.py list-scripts "prayer app"
python main.py list-scripts "prayer app" --output scripts.txt
python main.py list-scripts "prayer app" --type optimized

---

Project Structure

Slop/
├── main.py              # CLI entry point
├── requirements.txt     # All free/open-source deps
├── .env.example
├── brain/               # Your brand's "brain"
│   ├── writing-rules.md # Anti-AI-slop filter (ban list, tone rules)
│   ├── voice.md         # Brand voice + real script examples
│   ├── product.md       # Product facts, social proof, pricing
│   └── icp.md           # Customer segments with their language
├── slop/
│   ├── db.py            # SQLite storage (replaces Singular)
│   ├── research.py      # Meta Ad Library + yt-dlp + local Whisper
│   ├── generator.py     # Ollama script generation
│   └── optimizer.py     # Self-improvement loop
└── data/                # Generated output (gitignored)
    ├── scripts/         # JSON script files
    └── slop.db          # SQLite database

---

Free Stack Details

Meta Ad Library Token (free)

1. Go to developers.facebook.com/tools/explorer
2. Create a free app (type: "Business")
3. Generate a token with ads_read permission
4. Paste into .env

Ollama Models (free, local)

ollama pull llama3.1     # best quality (8B, ~4.7GB)
ollama pull mistral      # faster, smaller
ollama pull llama3.2     # latest, good balance
ollama pull deepseek-r1  # strong reasoning

Runs 100% on your machine. No internet needed after download. No cost ever.

Whisper Models (free, local)

Set in .env with WHISPER_MODEL=:

• tiny — fastest, lowest quality (75MB)
• base — good balance (150MB) ← default
• small — better accuracy (500MB)
• medium — high accuracy (1.5GB)
• large — best accuracy (3GB)

All run locally on CPU. No API calls.