Skip to content

Part 7 – ComfyUI

Use at your own risk. All guides and scripts are provided for educational purposes only. Always review and understand any code before running it — especially with administrative privileges. Test in a safe environment before using in production. Your system, your responsibility.

ComfyUI is a node-based interface for running AI creative models — image generation with Stable Diffusion and FLUX, video generation, music with ACE-Step, and voice synthesis with Kokoro TTS. Everything runs locally on the RTX 4080 Super.

This is the creative studio layer of the WCP — used for generating images, videos, music, and voice for projects including the WeichertLabs YouTube pipeline.

Prerequisites

  • ✅ Docker with NVIDIA Container Toolkit installed
  • ✅ NVIDIA drivers running
  • ✅ Caddy configured (optional — ComfyUI can also run on a direct port)
  • ✅ /mnt/ai/ directory created

→ Follow Part 2 for Docker and GPU setup.

Step 1 – Create the folder 

mkdir -p /opt/docker/comfyui
mkdir -p /mnt/ai/comfyui/models
mkdir -p /mnt/ai/comfyui/output
mkdir -p /mnt/ai/comfyui/input
cd /opt/docker/comfyui

Step 2 – Create the compose.yml 

nano compose.yml
services:
  comfyui:
    image: ghcr.io/ai-dock/comfyui:latest-cuda
    container_name: comfyui
    runtime: nvidia
    environment:
      - NVIDIA_VISIBLE_DEVICES=all
      - NVIDIA_DRIVER_CAPABILITIES=all
    ports:
      - "8188:8188"
    volumes:
      - /mnt/ai/comfyui/models:/opt/ComfyUI/models
      - /mnt/ai/comfyui/output:/opt/ComfyUI/output
      - /mnt/ai/comfyui/input:/opt/ComfyUI/input
    networks:
      - wcp-network
    restart: unless-stopped

networks:
  wcp-network:
    external: true

Note: ComfyUI is accessed directly on port 8188 rather than through Caddy — it works better without a reverse proxy for the websocket connections it uses.

Step 3 – Start ComfyUI 

docker compose up -d
docker compose logs -f

The first start downloads the ComfyUI image — this is a large download. Wait until you see the server is ready. Press CTRL+C to exit logs.

Step 4 – Access ComfyUI

Via Tailscale:

http://100.x.x.x:8188

Or if you’re on the same local network:

http://192.168.1.50:8188

You’ll see the ComfyUI node editor interface.

Step 5 – Download models

Models are stored in /mnt/ai/comfyui/models/ — organised by type.

Image generation — FLUX (recommended)

FLUX is the current state-of-the-art for image generation. Download from Hugging Face:

# Place in:
/mnt/ai/comfyui/models/unet/

Image generation — Stable Diffusion

SDXL and SD 1.5 models are widely available on Civitai and Hugging Face:

# Place in:
/mnt/ai/comfyui/models/checkpoints/

Music generation — ACE-Step

ACE-Step generates music from text prompts and is licensed under Apache 2.0 — safe for commercial use:

# Download from Hugging Face and place in:
/mnt/ai/comfyui/models/ace-step/

Voice/TTS — Kokoro

Kokoro TTS generates natural-sounding speech from text:

# Place in:
/mnt/ai/comfyui/models/kokoro/

Step 6 – Install custom nodes

ComfyUI’s power comes from its custom node ecosystem. Install via the ComfyUI Manager:

  1. In the ComfyUI interface, look for the Manager button
  2. If not present, install ComfyUI Manager manually:
docker exec -it comfyui bash
cd /opt/ComfyUI/custom_nodes
git clone https://github.com/ltdrdata/ComfyUI-Manager.git
exit
docker compose restart

From the Manager you can browse and install nodes for video generation, ControlNet, upscaling, and more.

GPU sharing with Ollama

Both Ollama and ComfyUI use the RTX 4080 Super. They handle GPU sharing gracefully — when ComfyUI is generating, Ollama’s responses may be slightly slower, and vice versa. In practice on 16 GB VRAM this rarely causes issues unless both are under heavy load simultaneously.

Monitor GPU usage:

watch -n 1 nvidia-smi

WeichertLabs YouTube pipeline

ComfyUI is central to the WeichertLabs content pipeline:

Tool Purpose
ComfyUI + ACE-Step Background music generation (Apache 2.0 — commercial OK)
ComfyUI + Kokoro TTS Voice narration for videos
Whisper Auto-generate subtitles/SRT from the narration
DaVinci Resolve Final video editing

The entire pipeline runs locally — no subscriptions, no data sent externally.

Useful commands 

# Restart ComfyUI
docker compose restart

# Follow logs
docker compose logs -f

# Check GPU usage while generating
nvidia-smi

What’s next

Part 8 sets up Sunshine and Moonlight for cloud gaming — streaming games from the RTX 4080 Super to any device.

Up next: Part 8 – Sunshine and Moonlight (coming soon)

Related guides