Skip to main content
This guide will help you create your first AI-generated image using ComfyUI.

​
Prerequisites

Before you begin, make sure you have:
  • Installed ComfyUI
  • At least one checkpoint model (e.g., SD 1.5 or SDXL)
  • Basic understanding of text-to-image generation

​
Starting ComfyUI

Launch ComfyUI using your preferred method:
Double-click the ComfyUI application icon.
ComfyUI will start and open in your web browser at http://127.0.0.1:8188

​
Your First Workflow

ComfyUI loads with a default text-to-image workflow. Here’s how to use it:
1

Load a checkpoint model

  1. Click on the CheckpointLoaderSimple node
  2. Select a model from the dropdown (e.g., v1-5-pruned-emaonly.safetensors)
  3. Place your checkpoint files in models/checkpoints/ directory
2

Enter your prompts

Positive Prompt (CLIPTextEncode):
masterpiece, best quality, beautiful landscape, mountains, sunset
Negative Prompt (CLIPTextEncode):
blurry, low quality, distorted
3

Configure generation settings

In the KSampler node:
  • Steps: 20-30 (more steps = higher quality, slower)
  • CFG Scale: 7-8 (how closely to follow your prompt)
  • Sampler: euler or dpmpp_2m (sampling algorithm)
  • Scheduler: normal or karras
  • Seed: Random number (use same seed for consistent results)
4

Set image dimensions

In the EmptyLatentImage node:
  • Width: 512 or 768
  • Height: 512 or 768
  • Keep dimensions as multiples of 64
5

Generate your image

Click Queue Prompt or press Ctrl+EnterWatch the progress bar and wait for generation to complete.
6

View and save your result

  • Your image appears in the SaveImage node
  • Images are automatically saved to output/ directory
  • Right-click the image to download or save

​
Understanding the Basic Workflow

The default workflow connects these nodes: Node Roles:
  • CheckpointLoaderSimple: Loads the AI model
  • CLIPTextEncode: Converts text prompts to embeddings
  • EmptyLatentImage: Creates blank latent space
  • KSampler: Generates image in latent space
  • VAEDecode: Converts latent to viewable image
  • SaveImage: Saves the final result

​
Quick Tips

  • Increase steps to 30-40
  • Use samplers like dpmpp_2m or dpmpp_sde
  • Add more descriptive keywords to your prompt
  • Use negative prompts to avoid unwanted elements
  • Reduce steps to 15-20
  • Use smaller image dimensions (512x512)
  • Enable --highvram if you have enough VRAM
  • Use faster samplers like euler_a or lcm
  • Use the same seed value
  • Lock your prompt and settings
  • Save your workflow: File → Save
  • Change the seed number
  • Adjust CFG scale (7-15 range)
  • Try different samplers and schedulers
  • Modify prompt keywords

​
Keyboard Shortcuts

Essential shortcuts to speed up your workflow:
ShortcutAction
Ctrl+EnterQueue prompt for generation
Ctrl+SSave workflow
Ctrl+OLoad workflow
SpacePan canvas
Double-ClickOpen node search
Ctrl+ZUndo
DeleteDelete selected nodes

​
Next Steps

Learn Core Concepts

Understand how workflows and nodes work together

Explore Models

Discover supported image generation models

Try Tutorials

Follow step-by-step guides for common tasks

Use the API

Integrate ComfyUI into your applications

​
Troubleshooting

If you encounter issues, check the Common Issues guide or visit the Discord community for help.
Common problems:
  • No models available: Place checkpoint files in models/checkpoints/
  • Out of memory: Reduce image size or use --lowvram flag
  • Black images: Try --force-fp32 flag (GTX 16 series cards)
  • Slow generation: Enable --fast for optimizations

Build docs developers (and LLMs) love

Get started for freeTalk to us