Install

FlixML runs on your own hardware. You need a machine (or machines) that can run ComfyUI, plus a small API and UI that route work to it.

Prerequisites

• Python 3.10+
• Node.js 18+
• PostgreSQL
• A running ComfyUI instance — local, remote, or serverless

Install

git clone https://github.com/ortegarod/flixml.git
cd flixml

# Config: environment variables and your ComfyUI nodes
cp .env.example .env
cp config.example.json config.json
# Edit .env and config.json for your machine

# API (from the repo root). Migrations run automatically.
pip install -r requirements.txt
createdb flixml_studio
set -a; . ./.env; set +a
PYTHONPATH=app python -m flixml

# Studio UI, in a second terminal from the repo root
set -a; . ./.env; set +a
cd studio && npm install && npm run dev

Configure your GPU

A provider is any GPU that runs ComfyUI. FlixML doesn’t care where it lives — a box under your desk, a server across the network, or serverless GPU. The API stays identical; the location is just configuration.

Add your ComfyUI URL to config.json so FlixML can route jobs to it:

{
  "gpu_nodes": [
    { "id": "gpu-1", "roles": ["image"], "comfyui": { "url": "http://<host>:<port>" } },
    { "id": "gpu-2", "roles": ["video"], "comfyui": { "url": "http://<host>:<port>" } }
  ]
}

Role-based routing

Each node declares roles. FlixML routes each job to a node that can serve it, so a video job never lands on an image-only node. Common roles:

• image — image generation
• video — video generation

A single node can carry multiple roles (["image", "video"]) if its VRAM can handle both.

Scaling out

Add more nodes to gpu_nodes and FlixML load-balances across them by role — no code changes. Start with one GPU; add more when you need throughput.

Discovering providers at runtime

Ask the API what’s configured and healthy:

curl <your-api-url>/api/providers

Reference a provider by its id when you generate.

Secure your install

A fresh install requires no key, so anyone who can reach it has full access. Before you expose it beyond your own machine, create an admin key in Studio under Settings → API keys, then set "security": { "require_api_key": true } in config.json. API Keys & Access has the steps, including a command for machines without a browser.

Key environment variables

Variable	Description
FLIXML_API_URL	URL the Studio UI uses to reach the backend
DATABASE_URL	PostgreSQL connection string
FLIXML_OUTPUT_DIR	Directory where generated media is stored
FLIXML_TRAINING_DIR	Required. Local folder for LoRA training datasets and configs
FLIXML_LORA_OUTPUT_DIR	Required. Local folder where trained LoRAs are saved
FLIXML_COMFY_LORA_DIR	Required. ComfyUI models/loras folder trained LoRAs are copied to
ELEVENLABS_API_KEY	Optional. Lists ElevenLabs voices at /api/tts/voices and speaks lines at /api/tts/generate.
ELEVENLABS_VOICE_ID	Optional. Default voice for /api/tts/generate when the request names none.

See .env.example for the full list.

Installed? Getting Started runs your first generation.