Your First Image

This guide walks you through generating your first image in ComfyUI. You should already have ComfyUI installed and a checkpoint model downloaded.

Load the Default Workflow

When you first open ComfyUI, you may see a blank canvas. To load the default text-to-image workflow:

1. Click the menu icon (☰) in the top-left corner
2. Select Load Default Workflow

You should see a set of connected nodes appear on the canvas:

• Load Checkpoint — loads your AI model
• CLIP Text Encode (Prompt) — your positive prompt (what you want)
• CLIP Text Encode (Negative) — your negative prompt (what to avoid)
• KSampler — the generation engine
• Empty Latent Image — sets the output resolution
• VAE Decode — converts the result to a viewable image
• Save Image — saves the output

Select Your Model

1. Click the Load Checkpoint node
2. In the ckpt_name dropdown, select your checkpoint file
3. If no models appear, you haven't placed a model in the models/checkpoints/ folder — see Download Models

Write a Prompt

Click the positive prompt node (the green CLIP Text Encode) and type what you want to generate. For example:

a beautiful sunset over mountains, golden hour, photorealistic, 8k

In the negative prompt node (the red CLIP Text Encode), describe what to avoid:

blurry, low quality, watermark, text, deformed

Set the Resolution

Click the Empty Latent Image node to set the output size:

Model	Recommended Size
SD 1.5	512 × 512
SDXL	1024 × 1024
Flux	1024 × 1024

Important

Using the wrong resolution for your model will produce poor results. SD 1.5 was trained on 512×512 — using 1024×1024 with it will look broken.

Generate

Click the Queue Prompt button in the bottom-right corner (or press Ctrl + Enter).

The progress bar will appear on the KSampler node. Your first generation may take longer as the model loads into GPU memory.

Once complete, the generated image appears in the Save Image node. Images are saved to the output/ folder.

Tips for Better Results

• Be specific — "a red sports car on a mountain road at sunset" works better than "car"
• Add quality tags — words like high quality, detailed, 8k, masterpiece can improve results
• Adjust the seed — change the seed value in KSampler to get different variations
• Increase steps — more steps (20-30) generally means higher quality, but takes longer
• CFG Scale — higher values (7-12) follow your prompt more closely, lower values (3-5) give more creative freedom

What's Next?

• Get More Models — Download different checkpoints, LoRAs, and specialized models.

• Install Custom Nodes — Add new capabilities like face restoration, upscaling, and video generation.

Stuck?

If you run into errors, check the Common Issues page. If the problem is beyond your comfort level, you can book a remote fix session — our experts can diagnose and resolve issues in real time.