Run Fruit AI locally for video generation

Fruit AI is a local-first video generation tool designed to create fruit-themed content. Running it on bare metal gives you full control over inference, no cloud bills, and privacy for your video projects. This guide walks you through a production-ready setup on a Linux machine.

Prerequisites

You’ll need:

• A Linux machine with at least 8GB RAM (16GB+ recommended for faster inference)
• NVIDIA GPU strongly recommended (CUDA 11.8+); CPU inference is possible but slow
• 50GB free disk space for model weights
• Python 3.8 or later
• Git

Step 1: Clone the repository

git clone https://github.com/fruitai/fruit-ai.git
cd fruit-ai

Step 2: Create a Python virtual environment

python3 -m venv venv
source venv/bin/activate

Step 3: Install dependencies

pip install --upgrade pip
pip install -r requirements.txt

If you have an NVIDIA GPU, also install CUDA support:

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

Step 4: Download model weights

Fruit AI downloads models on first run. Expect 10–30GB depending on which models you use. Run a test generation to trigger the download:

python fruit_ai.py --help

Refer to the official docs at https://fruitai.io/docs for the latest model URLs and configuration options.

Step 5: Generate your first video

Create a simple prompt file or use the CLI directly. Check the repo README for exact command syntax, as it may vary by version. A typical workflow looks like:

python fruit_ai.py --prompt "apple falling in slow motion" --output video.mp4

Consult https://fruitai.io/docs for the exact parameters your version supports.

Step 6: Run as a background service (optional)

For continuous operation, create a systemd service file at /etc/systemd/system/fruit-ai.service:

[Unit]
Description=Fruit AI Video Generation Service
After=network.target

[Service]
Type=simple
User=fruitai
WorkingDirectory=/opt/fruit-ai
Environment="PATH=/opt/fruit-ai/venv/bin"
ExecStart=/opt/fruit-ai/venv/bin/python fruit_ai.py --server
Restart=on-failure
RestartSec=10

[Install]
WantedBy=multi-user.target

Enable and start it:

sudo systemctl daemon-reload
sudo systemctl enable fruit-ai
sudo systemctl start fruit-ai

Check logs with sudo journalctl -u fruit-ai -f.

Step 7: Secure your setup

Important: Fruit AI should never be exposed to the public internet. Keep it on your LAN or behind a VPN only. If you run a web interface, bind it to 127.0.0.1 or a private network address, not 0.0.0.0.

If you need remote access, use SSH tunneling or WireGuard instead of opening ports.

Monitoring and maintenance

• Watch GPU memory usage with nvidia-smi during video generation
• Keep the repo updated: git pull periodically
• Monitor disk space; old videos and temp files can fill your drive
• Check the official docs for new model releases and performance tips

Is it worth it?

Yes, if you’re generating fruit videos regularly and want to avoid cloud costs or API rate limits. The upfront hardware cost (especially a decent GPU) is real, but inference on your own iron is fast and free after that. If you only generate a handful of videos per month, a cloud API might be simpler. But for a homelab that doubles as a creative tool, this pays for itself quickly and keeps your work private.

Sources

• https://github.com/fruitai/fruit-ai
• https://fruitai.io/docs