ComfyUI

Installation

bash
# install.sh — Clone and set up ComfyUI
git clone https://github.com/comfyanonymous/ComfyUI.git
cd ComfyUI

# Install dependencies (NVIDIA GPU)
pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu121
pip install -r requirements.txt

# Start the server
python main.py --listen 0.0.0.0 --port 8188
# Visit http://localhost:8188

Model Setup

bash
# setup_models.sh — Download and place models in the correct directories
cd ComfyUI

# SDXL base model
wget -P models/checkpoints/ \
    "https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/resolve/main/sd_xl_base_1.0.safetensors"

# VAE
wget -P models/vae/ \
    "https://huggingface.co/stabilityai/sdxl-vae/resolve/main/sdxl_vae.safetensors"

# LoRA adapters go in models/loras/
# ControlNet models go in models/controlnet/
# Upscale models go in models/upscale_models/

API: Queue a Workflow

python
# queue_prompt.py — Submit a workflow to ComfyUI via the API
import json
import requests
import uuid

COMFYUI_URL = "http://localhost:8188"

# Basic txt2img workflow
workflow = {
    "3": {
        "class_type": "KSampler",
        "inputs": {
            "seed": 42,
            "steps": 25,
            "cfg": 7.5,
            "sampler_name": "euler_ancestral",
            "scheduler": "normal",
            "denoise": 1.0,
            "model": ["4", 0],
            "positive": ["6", 0],
            "negative": ["7", 0],
            "latent_image": ["5", 0],
        },
    },
    "4": {
        "class_type": "CheckpointLoaderSimple",
        "inputs": {"ckpt_name": "sd_xl_base_1.0.safetensors"},
    },
    "5": {
        "class_type": "EmptyLatentImage",
        "inputs": {"width": 1024, "height": 1024, "batch_size": 1},
    },
    "6": {
        "class_type": "CLIPTextEncode",
        "inputs": {
            "text": "A majestic mountain landscape at golden hour, photorealistic, 8k",
            "clip": ["4", 1],
        },
    },
    "7": {
        "class_type": "CLIPTextEncode",
        "inputs": {
            "text": "blurry, low quality, distorted",
            "clip": ["4", 1],
        },
    },
    "8": {
        "class_type": "VAEDecode",
        "inputs": {"samples": ["3", 0], "vae": ["4", 2]},
    },
    "9": {
        "class_type": "SaveImage",
        "inputs": {"filename_prefix": "comfyui_output", "images": ["8", 0]},
    },
}

client_id = str(uuid.uuid4())
response = requests.post(
    f"{COMFYUI_URL}/prompt",
    json={"prompt": workflow, "client_id": client_id},
)
print(f"Queued: {response.json()}")

API: Get Results and Download Images

python
# get_results.py — Poll for completion and download generated images
import requests
import time
import urllib.request

COMFYUI_URL = "http://localhost:8188"

def wait_for_completion(prompt_id: str) -> dict:
    while True:
        response = requests.get(f"{COMFYUI_URL}/history/{prompt_id}")
        history = response.json()
        if prompt_id in history:
            return history[prompt_id]
        time.sleep(1)

def download_images(history: dict, output_dir: str = "./outputs"):
    import os
    os.makedirs(output_dir, exist_ok=True)
    for node_id, node_output in history["outputs"].items():
        if "images" in node_output:
            for image in node_output["images"]:
                url = f"{COMFYUI_URL}/view?filename={image['filename']}&subfolder={image.get('subfolder', '')}&type={image['type']}"
                filepath = os.path.join(output_dir, image["filename"])
                urllib.request.urlretrieve(url, filepath)
                print(f"Saved: {filepath}")

# Usage after queuing a prompt
prompt_id = "your-prompt-id"
history = wait_for_completion(prompt_id)
download_images(history)

Custom Nodes (ComfyUI Manager)

bash
# install_manager.sh — Install ComfyUI Manager for easy custom node management
cd ComfyUI/custom_nodes
git clone https://github.com/ltdrdata/ComfyUI-Manager.git

# Restart ComfyUI — Manager button appears in the UI
# Popular custom node packs:
# - ComfyUI-Impact-Pack: Detection, segmentation, inpainting
# - ComfyUI-AnimateDiff: Animation from static images
# - ComfyUI-IPAdapter: Image prompt adapter for style transfer
# - rgthree-comfy: Workflow organization utilities

ControlNet Workflow

python
# controlnet_workflow.py — Generate images guided by ControlNet (edge detection, depth, pose)
controlnet_nodes = {
    "10": {
        "class_type": "ControlNetLoader",
        "inputs": {"control_net_name": "control_v11p_sd15_canny.pth"},
    },
    "11": {
        "class_type": "LoadImage",
        "inputs": {"image": "input_image.png"},
    },
    "12": {
        "class_type": "CannyEdgePreprocessor",
        "inputs": {"image": ["11", 0], "low_threshold": 100, "high_threshold": 200},
    },
    "13": {
        "class_type": "ControlNetApply",
        "inputs": {
            "conditioning": ["6", 0],
            "control_net": ["10", 0],
            "image": ["12", 0],
            "strength": 0.8,
        },
    },
}
# Connect node "13" output to KSampler positive conditioning instead of "6"

Docker Deployment

yaml
# docker-compose.yml — Run ComfyUI in Docker with GPU support
version: "3.8"
services:
  comfyui:
    image: ghcr.io/ai-dock/comfyui:latest
    ports:
      - "8188:8188"
    volumes:
      - ./models:/workspace/ComfyUI/models
      - ./output:/workspace/ComfyUI/output
      - ./custom_nodes:/workspace/ComfyUI/custom_nodes
    deploy:
      resources:
        reservations:
          devices:
            - capabilities: [gpu]

Key Concepts

• Nodes and links: Visual programming — connect output slots to input slots to build pipelines
• Workflows: Saved as JSON files — shareable, version-controllable, API-submittable
• Custom nodes: Extend functionality via Python — community ecosystem via ComfyUI Manager
• Checkpoints: Model files (.safetensors) placed in models/checkpoints/
• LoRA: Lightweight fine-tuned adapters loaded alongside base models
• ControlNet: Guide generation with structural inputs (edges, depth, pose)
• API-first: Full HTTP API for queuing prompts and retrieving results programmatically