Setting Up ComfyUI for Image Generation

Lorefire can generate character portraits and scene art using ComfyUI — a local Stable Diffusion backend that runs entirely on your machine. No API key or internet connection required.

Requirements

A GPU with at least 6 GB VRAM (NVIDIA recommended; AMD supported on Windows via DirectML)
10–20 GB of free disk space for models
Python 3.10 or higher (macOS/Linux) — Windows users can use the portable build

Install ComfyUI

macOS

git clone https://github.com/comfyanonymous/ComfyUI
cd ComfyUI
pip3 install -r requirements.txt

If you are on Apple Silicon, install the MPS-accelerated torch build first:

pip3 install torch torchvision torchaudio

Windows

Download the portable standalone build from the ComfyUI releases page. Extract the zip, then run:

run_nvidia_gpu.bat — for NVIDIA GPUs
run_cpu.bat — if you have no discrete GPU (slower)

No Python installation needed for the portable build.

Download a Model

ComfyUI needs at least one Stable Diffusion checkpoint to generate images. Place .safetensors files in:

macOS/Linux: ComfyUI/models/checkpoints/
Windows (portable): ComfyUI_windows_portable\ComfyUI\models\checkpoints\

Good starting points available on CivitAI and Hugging Face:

Model	Style	Notes
DreamShaper XL	Versatile / painterly	Great all-rounder for portraits
Juggernaut XL	Photorealistic	High detail, needs more VRAM
Realistic Vision v5	Photorealistic (SD 1.5)	Lighter, works on 6 GB VRAM

SDXL models produce higher quality but require more VRAM (8 GB+). SD 1.5 models are faster and run on 6 GB cards.

Start the ComfyUI Server

macOS / Linux

cd ComfyUI
python3 main.py --listen 127.0.0.1 --port 8188

Windows (portable)

Double-click run_nvidia_gpu.bat (or run_cpu.bat). The server starts automatically on port 8188.

The ComfyUI web interface will be available at http://127.0.0.1:8188 — you can leave this tab open but don't need to interact with it.

Connect Lorefire to ComfyUI

Open Lorefire and go to Settings
Under Art Generation, select ComfyUI (Local)
Set the server URL to http://127.0.0.1:8188
Click Test Connection — you should see a green confirmation

You can now generate portraits from the NPC Roster and scene art from session pages.

Troubleshooting

"Connection refused" or timeout

Make sure ComfyUI is running before clicking Test Connection in Lorefire
Confirm the port matches (default is 8188)
On Windows, check Windows Firewall is not blocking the port

Out of memory / crashes during generation

Add low-VRAM flags to the launch command:

python3 main.py --listen 127.0.0.1 --port 8188 --lowvram

Or for CPU-only (very slow but no VRAM needed):

python3 main.py --listen 127.0.0.1 --port 8188 --cpu

Model not appearing in ComfyUI

Verify the .safetensors file is in the checkpoints/ folder (not a subfolder)
Restart ComfyUI after adding new models — it only scans on startup

Slow generation on Apple Silicon

Ensure you installed the MPS-enabled torch build. You can verify acceleration is active by checking the ComfyUI terminal output for mps device usage.