Ollama Models

MoneyPrinter uses local Ollama models for AI-powered script generation and metadata creation.

Why Ollama?

Ollama provides local LLM inference with several advantages:

Privacy

All AI processing happens locally. No data sent to cloud APIs.

Cost

Zero per-token charges. Run unlimited generations.

Speed

Low-latency inference with local models.

Flexibility

Choose from dozens of open-source models.

Installing Ollama

Download and Install

curl -fsSL https://ollama.com/install.sh | sh

Verify installation:

ollama --version

Start Ollama Service

ollama serve

Ollama runs on http://localhost:11434 by default.

On macOS/Windows, Ollama runs as a background service after installation. No need to manually run ollama serve.

Pulling Models

Basic Usage

Pull a model:

ollama pull llama3.1:8b

List installed models:

ollama list

Expected output:

NAME              ID              SIZE      MODIFIED
llama3.1:8b       42182419e950    4.7 GB    2 hours ago
mistral:7b        f974a74358d6    4.1 GB    1 day ago

Remove a model:

ollama rm mistral:7b

Recommended Models

For Script Generation

Model	Size	RAM Required	Speed	Quality	Best For
llama3.1:8b	4.7GB	8GB	Fast	Excellent	General purpose, balanced
mistral:7b	4.1GB	8GB	Very Fast	Good	Quick iterations
llama3.1:70b	40GB	64GB	Slow	Outstanding	Best quality (GPU required)
qwen2.5:7b	4.7GB	8GB	Fast	Good	Multilingual content
phi3:mini	2.3GB	4GB	Very Fast	Fair	Low-resource environments

Selecting a Model

Consider these factors:

Available RAM: Model size + 2GB overhead
GPU: CUDA/Metal acceleration for larger models
Script quality: Larger models produce better scripts
Generation speed: Smaller models are faster

Start with llama3.1:8b for the best balance of quality and performance.

Configuring MoneyPrinter

Default Model

Set the fallback model in .env:

.env

OLLAMA_MODEL="llama3.1:8b"

This model is used when the frontend doesn’t specify one:

Backend/gpt.py

OLLAMA_MODEL = os.getenv("OLLAMA_MODEL", "llama3.1:8b")

def generate_response(prompt: str, ai_model: str) -> str:
    model_name = (ai_model or "").strip() or OLLAMA_MODEL
    # ...

Remote Ollama Server

Run Ollama on a different machine:

.env

OLLAMA_BASE_URL="http://192.168.1.100:11434"

Or use a GPU server:

.env

OLLAMA_BASE_URL="http://gpu-server:11434"

Docker Configuration

When running MoneyPrinter in Docker with Ollama on the host:

.env

OLLAMA_BASE_URL="http://host.docker.internal:11434"

This works on macOS, Windows, and Linux (via extra_hosts in docker-compose.yml).

Model Selection in UI

The frontend fetches available models from the API:

Frontend/app.js

async function loadOllamaModels() {
  const response = await apiRequest('/api/models');
  
  if (response.status === 'success') {
    const modelSelect = document.getElementById('aiModel');
    modelSelect.innerHTML = '';
    
    response.models.forEach(model => {
      const option = document.createElement('option');
      option.value = model;
      option.textContent = model;
      if (model === response.default) {
        option.selected = true;
      }
      modelSelect.appendChild(option);
    });
  }
}

The API endpoint:

Backend/main.py

@app.route("/api/models", methods=["GET"])
def models():
    try:
        available_models, default_model = list_ollama_models()
        return jsonify(
            {
                "status": "success",
                "models": available_models,
                "default": default_model,
            }
        )
    except Exception as err:
        log(f"[-] Error fetching Ollama models: {str(err)}", "error")
        return jsonify(
            {
                "status": "error",
                "message": "Could not fetch Ollama models. Is Ollama running?",
                "models": [os.getenv("OLLAMA_MODEL", "llama3.1:8b")],
                "default": os.getenv("OLLAMA_MODEL", "llama3.1:8b"),
            }
        )

Model Usage in Pipeline

Script Generation

Backend/gpt.py

def generate_script(
    video_subject: str,
    paragraph_number: int,
    ai_model: str,
    voice: str,
    customPrompt: str,
) -> Optional[str]:
    prompt = f"""
    Generate a script for a video about {video_subject}.
    Number of paragraphs: {paragraph_number}
    Language: {voice}
    """
    
    response = generate_response(prompt, ai_model)
    # Clean and return script

Search Terms

Backend/gpt.py

def get_search_terms(
    video_subject: str, amount: int, script: str, ai_model: str
) -> List[str]:
    prompt = f"""
    Generate {amount} search terms for stock videos.
    Subject: {video_subject}
    Return as JSON array: ["term1", "term2", ...]
    """
    
    response = generate_response(prompt, ai_model)
    return json.loads(response)

Metadata Generation

Backend/gpt.py

def generate_metadata(
    video_subject: str, script: str, ai_model: str
) -> Tuple[str, str, List[str]]:
    # Generate title
    title_prompt = f"Generate a catchy YouTube title for: {video_subject}"
    title = generate_response(title_prompt, ai_model)
    
    # Generate description
    desc_prompt = f"Write a YouTube description for: {script}"
    description = generate_response(desc_prompt, ai_model)
    
    # Generate keywords
    keywords = get_search_terms(video_subject, 6, script, ai_model)
    
    return title, description, keywords

Model Performance

Benchmarks

Approximate generation times on Apple M2 Pro:

Model	Script (1 para)	Search Terms	Metadata	Total
llama3.1:8b	5s	3s	8s	~16s
mistral:7b	3s	2s	5s	~10s
llama3.1:70b	25s	15s	35s	~75s
phi3:mini	2s	1s	3s	~6s

Timings exclude video download, TTS, and rendering. GPU acceleration significantly improves performance.

Troubleshooting

No models in frontend dropdown

Verify Ollama is running:

curl http://localhost:11434/api/tags

Expected response:

{
  "models": [
    {"name": "llama3.1:8b", "model": "llama3.1:8b", "size": 4661211648}
  ]
}

If connection refused:

ollama serve

Model not found error

Error message:

RuntimeError: Ollama model 'llama3.1:8b' is not installed.
Available models: mistral:7b
Install it with: ollama pull llama3.1:8b

Solution:

ollama pull llama3.1:8b

The backend checks available models and provides clear instructions:

Backend/gpt.py

except ResponseError as fallback_err:
    if fallback_err.status_code == 404 and "not found" in str(fallback_err).lower():
        available_models, _ = list_ollama_models()
        available = ", ".join(available_models) if available_models else "none"
        raise RuntimeError(
            f"Ollama model '{model_name}' is not installed. "
            f"Available models: {available}. "
            f"Install it with: ollama pull {model_name}"
        ) from fallback_err

Ollama connection refused (Docker)

Check OLLAMA_BASE_URL:

.env

OLLAMA_BASE_URL="http://host.docker.internal:11434"

Test connectivity from container:

docker exec -it backend curl http://host.docker.internal:11434/api/tags

Linux: Verify host.docker.internal resolves:

docker exec -it backend ping host.docker.internal

Out of memory errors

Symptoms:

Ollama crashes during generation
System becomes unresponsive
“Out of memory” errors

Solutions:

Use a smaller model:
```
ollama pull mistral:7b
```
Limit concurrent generations (single worker)

Add swap space (Linux):

sudo fallocate -l 16G /swapfile
sudo chmod 600 /swapfile
sudo mkswap /swapfile
sudo swapon /swapfile

Advanced Configuration

Model Parameters

Customize generation parameters (requires code changes):

Backend/gpt.py

response = client.chat(
    model=model_name,
    messages=[{"role": "user", "content": prompt}],
    stream=False,
    options={
        "temperature": 0.7,  # Creativity (0.0-1.0)
        "top_p": 0.9,        # Nucleus sampling
        "top_k": 40,         # Token selection
        "num_predict": 500,  # Max tokens
    }
)

Timeout Configuration

Adjust timeout for slow models:

.env

OLLAMA_TIMEOUT=300  # 5 minutes

Used by the client:

Backend/gpt.py

OLLAMA_TIMEOUT = float(os.getenv("OLLAMA_TIMEOUT", "180"))

def _ollama_client() -> Client:
    return Client(host=OLLAMA_BASE_URL, timeout=OLLAMA_TIMEOUT)

Next Steps

Generating Videos

Create your first video

Pipeline

Understand the generation process

Configuration

All environment variables

Troubleshooting

Common issues and solutions

Documentation Index

​Ollama Models

​Why Ollama?

Privacy

Cost

Speed

Flexibility

​Installing Ollama

​Download and Install

​Start Ollama Service

​Pulling Models

​Basic Usage

​Recommended Models

​For Script Generation

​Selecting a Model

​Configuring MoneyPrinter

​Default Model

​Remote Ollama Server

​Docker Configuration

​Model Selection in UI

​Model Usage in Pipeline

​Script Generation

​Search Terms

​Metadata Generation

​Model Performance

​Benchmarks

​Troubleshooting

​Advanced Configuration

​Model Parameters

​Timeout Configuration

​Next Steps

Generating Videos

Pipeline

Configuration

Troubleshooting

Ollama Models

Why Ollama?

Installing Ollama

Download and Install

Start Ollama Service

Pulling Models

Basic Usage

Recommended Models

For Script Generation

Selecting a Model

Configuring MoneyPrinter

Default Model

Remote Ollama Server

Docker Configuration

Model Selection in UI

Model Usage in Pipeline

Script Generation

Search Terms

Metadata Generation

Model Performance

Benchmarks

Troubleshooting

Advanced Configuration

Model Parameters

Timeout Configuration

Next Steps