ComfyUI

Node-based interface for Stable Diffusion with ultimate flexibility on CLORE.AI GPUs.

All examples can be run on GPU servers rented through CLORE.AI Marketplace.

Server Requirements

Parameter	Minimum	Recommended
RAM	16GB	32GB+
VRAM	8GB (SDXL)	12GB+
Network	500Mbps	1Gbps+
Startup Time	5-10 minutes	-

Startup Time: First launch downloads dependencies and models (5-10 minutes depending on network speed). HTTP 502 during this time is normal.

Important: ComfyUI with FLUX models requires 16GB+ VRAM. For SDXL with ControlNet, ensure at least 10GB VRAM.

Why ComfyUI?

• Node-based workflow - Visual programming for image generation

• Maximum control - Fine-tune every step of the pipeline

• Efficient - Lower VRAM usage than alternatives

• Extensible - Huge ecosystem of custom nodes

• Workflow sharing - Import/export as JSON

Quick Deploy on CLORE.AI

Docker Image:

yanwk/comfyui-boot:cu126-slim

Ports:

22/tcp
8188/http

Environment:

CLI_ARGS=--listen 0.0.0.0

Verify It's Working

After deployment, find your http_pub URL in My Orders:

# Check if UI is accessible (may take 5-10 min on first run)
curl https://your-http-pub.clorecloud.net/

If you get HTTP 502 for more than 15 minutes, check:

1. Server has 16GB+ RAM

2. Server has 8GB+ VRAM for SDXL, 16GB+ for FLUX

3. Network speed is adequate for downloading models

Accessing Your Service

When deployed on CLORE.AI, access ComfyUI via the http_pub URL:

• Web UI: https://your-http-pub.clorecloud.net/

• API: https://your-http-pub.clorecloud.net/prompt

• WebSocket: wss://your-http-pub.clorecloud.net/ws

All localhost:8188 examples below work when connected via SSH. For external access, replace with your https://your-http-pub.clorecloud.net/ URL.

Installation

Using Docker (Recommended)

docker run -d --gpus all \
  -p 8188:8188 \
  -v comfyui-data:/root \
  -e CLI_ARGS="--listen 0.0.0.0" \
  yanwk/comfyui-boot:cu126-slim

Manual Installation

# Clone repository
git clone https://github.com/comfyanonymous/ComfyUI.git
cd ComfyUI

# Create virtual environment
python -m venv venv
source venv/bin/activate

# Install PyTorch with CUDA
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu124

# Install requirements
pip install -r requirements.txt

# Run
python main.py --listen 0.0.0.0

Directory Structure

ComfyUI/
├── models/
│   ├── checkpoints/     # SD models (.safetensors)
│   ├── loras/           # LoRA models
│   ├── vae/             # VAE models
│   ├── controlnet/      # ControlNet models
│   ├── upscale_models/  # Upscalers
│   └── clip/            # CLIP models
├── input/               # Input images
├── output/              # Generated images
└── custom_nodes/        # Extensions

Download Models

Stable Diffusion Checkpoints

cd ComfyUI/models/checkpoints

# SDXL Base
wget https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0/resolve/main/sd_xl_base_1.0.safetensors

# SDXL Refiner
wget https://huggingface.co/stabilityai/stable-diffusion-xl-refiner-1.0/resolve/main/sd_xl_refiner_1.0.safetensors

# SD 1.5 (smaller, faster)
wget https://huggingface.co/runwayml/stable-diffusion-v1-5/resolve/main/v1-5-pruned.safetensors

# Realistic Vision (photorealistic)
wget https://huggingface.co/SG161222/Realistic_Vision_V6.0_B1_noVAE/resolve/main/Realistic_Vision_V6.0_B1_fp16.safetensors

VAE

cd ComfyUI/models/vae

# SDXL VAE
wget https://huggingface.co/stabilityai/sdxl-vae/resolve/main/sdxl_vae.safetensors

# SD 1.5 VAE (better colors)
wget https://huggingface.co/stabilityai/sd-vae-ft-mse-original/resolve/main/vae-ft-mse-840000-ema-pruned.safetensors

LoRAs

cd ComfyUI/models/loras

# Download from CivitAI or HuggingFace
# Place .safetensors files here

Basic Workflow

Text to Image

1. Add nodes:

   • Load Checkpoint → select model

   • CLIP Text Encode (x2) → positive & negative prompts

   • Empty Latent Image → set dimensions

   • KSampler → connect all

   • VAE Decode → latent to image

   • Save Image → output

2. Connect:

[Checkpoint] → MODEL → [KSampler]
[Checkpoint] → CLIP → [CLIP Text Encode +]
[Checkpoint] → CLIP → [CLIP Text Encode -]
[Checkpoint] → VAE → [VAE Decode]
[Text Encode +] → CONDITIONING → [KSampler]
[Text Encode -] → CONDITIONING → [KSampler]
[Empty Latent] → LATENT → [KSampler]
[KSampler] → LATENT → [VAE Decode]
[VAE Decode] → IMAGE → [Save Image]

Image to Image

Replace Empty Latent Image with:

1. Load Image → your source image

2. VAE Encode → convert to latent

3. Adjust denoise in KSampler (0.5-0.8)

Essential Custom Nodes

ComfyUI Manager (Must Have)

cd ComfyUI/custom_nodes
git clone https://github.com/ltdrdata/ComfyUI-Manager.git

Restart ComfyUI, then use Manager to install other nodes from UI.

Popular Node Packs

Node Pack	Use Case
ComfyUI-Manager	Install/update nodes
ComfyUI-Impact-Pack	Face detection, SAM
ComfyUI-Inspire-Pack	Advanced samplers
ComfyUI-AnimateDiff	Video generation
ComfyUI-VideoHelperSuite	Video I/O
ComfyUI-GGUF	Run GGUF models
ComfyUI-KJNodes	Utility nodes
rgthree-comfy	Workflow helpers

Install via Manager

1. Click Manager button

2. Install Custom Nodes

3. Search and install

4. Restart ComfyUI

Advanced Workflows

ControlNet

# Download ControlNet models
cd ComfyUI/models/controlnet

# Canny
wget https://huggingface.co/lllyasviel/ControlNet-v1-1/resolve/main/control_v11p_sd15_canny.pth

# Depth
wget https://huggingface.co/lllyasviel/ControlNet-v1-1/resolve/main/control_v11f1p_sd15_depth.pth

# OpenPose
wget https://huggingface.co/lllyasviel/ControlNet-v1-1/resolve/main/control_v11p_sd15_openpose.pth

Workflow:

1. Load Image → Canny Edge Detector

2. Apply ControlNet → KSampler

3. Generate with pose/edge guidance

Upscaling

# Download upscaler
cd ComfyUI/models/upscale_models
wget https://github.com/xinntao/Real-ESRGAN/releases/download/v0.1.0/RealESRGAN_x4plus.pth

Workflow:

1. Generate image at lower res (768x768)

2. Upscale Image (Model) node

3. Optional: img2img pass for detail

SDXL + Refiner

1. Generate with SDXL base (steps 1-20)

2. Pass latent to SDXL refiner (steps 21-30)

3. VAE Decode final result

Keyboard Shortcuts

Key	Action
Ctrl+Enter	Queue prompt
Ctrl+Shift+Enter	Queue prompt (front)
Ctrl+Z	Undo
Ctrl+Y	Redo
Ctrl+S	Save workflow
Ctrl+O	Load workflow
Ctrl+A	Select all
Delete	Delete selected
Ctrl+M	Mute node
Ctrl+B	Bypass node

API Usage

Queue Prompt

import json
import urllib.request

# For external access, use your http_pub URL:
SERVER = "your-http-pub.clorecloud.net"
# Or via SSH: SERVER = "localhost:8188"

def queue_prompt(prompt, server=SERVER):
    data = json.dumps({"prompt": prompt}).encode('utf-8')
    req = urllib.request.Request(f"https://{server}/prompt", data=data)
    urllib.request.urlopen(req)

# Load workflow JSON and queue
with open("workflow.json") as f:
    workflow = json.load(f)
queue_prompt(workflow)

WebSocket for Progress

import websocket
import json

# For external access, use wss:// with your http_pub URL
ws = websocket.WebSocket()
ws.connect(f"wss://{SERVER}/ws")

while True:
    msg = json.loads(ws.recv())
    if msg['type'] == 'progress':
        print(f"Step {msg['data']['value']}/{msg['data']['max']}")
    elif msg['type'] == 'executed':
        print("Done!")
        break

Performance Tips

1. Enable --lowvram for <8GB VRAM

2. Use fp16 models when possible

3. Batch size 1 for limited VRAM

4. Tiled VAE for high-res images

5. Disable preview for faster generation

GPU Requirements

Model	Minimum VRAM	Recommended VRAM	Min RAM
SD 1.5	4GB	8GB	16GB
SDXL	8GB	12GB	16GB
SDXL + ControlNet	10GB	16GB	16GB
FLUX	16GB	24GB	32GB

GPU Presets

RTX 3060 12GB (Budget)

# Launch with optimizations
python main.py --lowvram --force-fp16

# Recommended settings:
# - SDXL: 768x768, batch 1
# - SD 1.5: 512x512, batch 4
# - Use VAE tiling
# - 20-30 steps

Best for: SD 1.5, SDXL (with limits)

RTX 3090 24GB (Optimal)

# Standard launch
python main.py --force-fp16

# Recommended settings:
# - SDXL: 1024x1024, batch 2
# - FLUX schnell: 1024x1024, batch 1
# - ControlNet + SDXL works well
# - 30-50 steps

Best for: SDXL, ControlNet workflows, moderate FLUX

RTX 4090 24GB (Performance)

# Full speed launch
python main.py

# Recommended settings:
# - SDXL: 1024x1024, batch 4
# - FLUX dev: 1024x1024, batch 1-2
# - Complex workflows with multiple ControlNets
# - 50+ steps for quality

Best for: FLUX, complex workflows, batch generation

A100 40GB/80GB (Production)

# Maximum performance
python main.py --highvram

# Recommended settings:
# - SDXL: 1024x1024, batch 8+
# - FLUX: 1024x1024, batch 2-4
# - Multiple models loaded simultaneously
# - High-res outputs (2048x2048)

Best for: Production workloads, FLUX, high-res generation

Cost Estimate

Typical CLORE.AI marketplace rates:

GPU	CLORE/day	Approx USD/hr	SDXL Speed
RTX 3060 12GB	~80	~$0.02	~15 sec/img
RTX 3090 24GB	~150	~$0.03	~8 sec/img
RTX 4090 24GB	~200	~$0.04	~4 sec/img
A100 40GB	~400	~$0.08	~3 sec/img

Prices vary by provider. Check CLORE.AI Marketplace for current rates.

Troubleshooting

HTTP 502 for a long time

1. Check RAM: Server must have 16GB+ RAM

2. Check VRAM: 8GB+ for SDXL, 16GB+ for FLUX

3. Dependencies downloading: First run takes 5-10 min

4. Model downloading: Large models take longer

Out of Memory

# Run with low VRAM mode
python main.py --lowvram

# Or force fp16
python main.py --force-fp16

Black Images

• Check VAE is loaded

• Try different VAE

• Reduce image size

Slow Generation

• Enable CUDA

• Use fp16 models

• Reduce steps (20-30 is often enough)

Workflow Examples

Import these JSON workflows in ComfyUI:

• Basic txt2img

• SDXL + Refiner

• ControlNet Canny

• AnimateDiff Video

Next Steps

• ControlNet Guide

• Real-ESRGAN Upscaling

• Kohya Training - Train custom LoRAs

• Fooocus - Simpler alternative