ComfyUI Text-to-Image Tutorial: Generate Your First AI Image (2026)

Q: What is the difference between positive and negative prompts in ComfyUI?

Positive prompts describe what you want to see in the image. Negative prompts describe what you want to exclude — such as blurry, watermark, or bad anatomy. Both are entered in separate CLIP Text Encoder nodes.

Introduction

What you'll learn: How to load a workflow in ComfyUI, what each node does, how to write positive and negative prompts, and how to run your first image generation.

Curious about creating AI images with ComfyUI? In this beginner-friendly guide, you will learn how to set up ComfyUI, load an image generation workflow, understand what each node does, write your first prompts, and generate your first AI image. Follow these step-by-step instructions to start creating AI art quickly.

Step 1: Install ComfyUI and Basic Setup

Before starting, make sure ComfyUI is installed on your computer. If it isn't yet, follow the full installation guide first — then come back here.

How to Install ComfyUI Manager Setup Guide

💡 Tip: After installing ComfyUI, also install ComfyUI Manager and ComfyUI-Crystools for a much smoother experience — they make managing nodes and monitoring your GPU effortless.

Step 2: Load an Image Generation Workflow

ComfyUI uses workflows — visual graphs of connected nodes — to generate images. The easiest way to get started is by loading the built-in template.

Click Templates in the sidebar

Open the ComfyUI sidebar and click on the "Templates" option.

Open the Image Generation template

Select the Image Generations template. This loads a basic workflow with all the nodes you need to start generating.

ComfyUI sidebar showing the Templates panel with Image Generation workflow selected — Select the Image Generation template from the ComfyUI sidebar

If a message appears saying models are missing, click the option to download models. ComfyUI will handle the download automatically.

ComfyUI prompt to download missing model files when loading a workflow — Click to download missing models when prompted

If the automatic download does not work, download the model manually from Hugging Face and place it in this folder:

ComfyUI_windows_portable\ComfyUI\models\checkpoints

After placing the file, refresh your browser. Then in the Load Checkpoint node, select v1-5-pruned-emaonly-fp16 from the dropdown.

Step 3: Understanding Key Workflow Nodes

ComfyUI workflows use different nodes to handle the image creation process. Here is what each one does, explained simply for beginners:

ComfyUI text-to-image workflow showing all nodes connected: Load Checkpoint, CLIP Text Encoder, Empty Latent Image, KSampler, VAE Decode, and Save Image — The default ComfyUI image generation workflow — all nodes connected and ready

Load Checkpoint Node

Loads the AI model used for image generation (such as Stable Diffusion).
The selected checkpoint determines the overall style and capability of your outputs.

CLIP Text Encoder (Prompt Nodes)

Positive prompt — describe what you want to see in the image.
Negative prompt — describe what you want to exclude (e.g. blurry, watermark, bad anatomy).
There are two separate CLIP Text Encoder nodes — one for each.

ComfyUI CLIP Text Encoder nodes showing positive and negative prompt input fields — Two CLIP Text Encoder nodes — one for positive, one for negative prompts

Empty Latent Image Node

Sets the output image dimensions (width and height in pixels).
Also sets the batch size — how many images to generate at once.

KSampler Node

The KSampler is the core engine — it takes your prompts and model and generates an image from random noise. These are the key settings to know:

Setting	What it does	Beginner value
`Seed`	Controls image randomness. Same seed = same image.	Any number
`Steps`	More steps = more detail, but slower generation.	30
`CFG Scale`	How closely the image follows your prompt. Higher = more literal.	7
`Sampler Name`	The algorithm used to reduce noise step by step.	euler
`Scheduler`	Controls how step size changes during sampling.	normal
`Denoise`	How much the image changes from its initial noise. 1.0 = full generation from scratch.	1.0

ComfyUI KSampler node showing seed, steps, CFG scale, sampler name, scheduler, and denoise controls — KSampler settings — start with Steps: 30 and CFG Scale: 7 for good beginner results

VAE Decode Node

Converts the model's internal latent representation into a viewable image.
Without this node, you would only have raw latent data — not a real picture.

Save Image Node

Saves your final generated image to disk.
Default save location: ComfyUI_windows_portable/ComfyUI/output
Edit the filename_prefix to change the subfolder. For example, setting it to images/ComfyUI saves to the output/images/ subfolder.

Step 4: Write Your First Prompts

Now it is time to generate your first AI image. Fill in the positive and negative prompt fields in the CLIP Text Encoder nodes. Here are five ready-to-use prompt examples across different styles — click Copy to paste them directly into ComfyUI.

💡 Tip: For all the prompts below, set Steps to 30 and CFG Scale to 7 in the KSampler node for best beginner results.

Realistic Portrait Photography

Positive Prompt

portrait of a confident older woman with silver hair, sitting near a window, soft daylight on her face, shallow depth of field, realistic skin texture, detailed eyes, elegant style, 85mm lens, high quality photo

Negative Prompt

blurry, low quality, bad anatomy, deformed face, extra fingers, out of focus, watermark, text, logo, jpeg artifacts

AI-generated realistic portrait of a woman with silver hair near a window, created with ComfyUI — Example output — Realistic Portrait Photography

Cinematic Action Scene

Positive Prompt

a lone samurai standing in the rain at night, cinematic lighting, dramatic shadows, neon reflections on wet streets, detailed armor, ultra-realistic, 8k masterpiece, moody atmosphere, movie still

Negative Prompt

lowres, cartoon, flat lighting, unrealistic proportions, bad hands, duplicate body, blurry, watermark, text

AI-generated cinematic image of a samurai in rain with neon reflections, created with ComfyUI — Example output — Cinematic Action Scene

Fantasy Illustration

Positive Prompt

a majestic dragon flying over a glowing forest, fantasy art, highly detailed, vivid colors, volumetric light rays, concept art style, trending on ArtStation, digital painting

Negative Prompt

blurry, dull colors, low detail, cropped, out of frame, watermark, bad anatomy, extra limbs, distorted wings

AI-generated fantasy illustration of a dragon flying over a glowing forest, made with ComfyUI — Example output — Fantasy Illustration

Anime Style Character

Positive Prompt

anime girl with long silver hair and blue eyes, standing in a cherry blossom field, soft lighting, vibrant colors, highly detailed, beautiful background, anime key visual, studio-quality

Negative Prompt

realistic photo, blurry, lowres, deformed, bad anatomy, extra fingers, watermark, text, low detail

AI-generated anime-style character with silver hair in a cherry blossom field, created with ComfyUI — Example output — Anime Style Character

Artistic Oil Painting

Positive Prompt

oil painting of an older woman wearing a vintage hat, warm tones, impressionist style, soft brush strokes, beautiful lighting, fine art masterpiece, hanging in an art gallery

Negative Prompt

blurry, digital art, pixelated, oversaturated, text, watermark, bad proportions, sketch, unfinished, messy

AI-generated oil painting of a woman in a vintage hat with impressionist style, created with ComfyUI — Example output — Artistic Oil Painting

Step 5: Run the Workflow

After entering your prompts and checking your settings, you are ready to generate:

Click "Queue Prompt" to start the workflow

Find the Queue Prompt button at the top of the ComfyUI interface and click it. The workflow will begin processing.

Watch the progress bar

ComfyUI will show a progress bar as it processes each step. Generation time depends on your GPU and the Steps setting.

Find your saved image

Once complete, your image is saved to the output folder. You can also see it previewed directly in the Save Image node.

ComfyUI_windows_portable\ComfyUI\output

💡 Tip: If your image looks off — too blurry, distorted, or not matching the prompt — try adjusting the Seed (for a different result), increasing Steps (for more detail), or tweaking the CFG Scale (higher = more literal, lower = more creative).

Frequently Asked Questions

Load the Image Generation template from the Templates sidebar, select or download a checkpoint model in the Load Checkpoint node, type your prompt in the CLIP Text Encoder, then click Queue Prompt. Your image will appear and be saved to the output folder.

A workflow in ComfyUI is a visual graph of connected nodes that defines how an image is generated. Each node performs a specific task — loading a model, encoding a prompt, sampling noise, decoding the image, and saving it. You can build custom workflows or load pre-made ones.

For beginners, set Steps to 30 and CFG Scale to 7. These give a good balance between image quality and generation speed. Once you are comfortable, experiment with higher step counts for more detail or different CFG values for creative variation.

By default, generated images are saved in ComfyUI_windows_portable/ComfyUI/output. You can change the subfolder by editing the filename_prefix in the Save Image node — for example, setting it to images/ComfyUI saves outputs to the output/images/ subfolder.

Positive prompts describe what you want to see — subjects, styles, lighting, quality keywords. Negative prompts describe what you want to exclude — such as blurry, watermark, bad anatomy, or extra fingers. Both are entered in separate CLIP Text Encoder nodes and work together to guide the image generation.

Conclusion

Starting with ComfyUI is straightforward once you understand the basics. You now know how to load a workflow, what each node does, how to write positive and negative prompts, and how to run your first generation. Use these foundations to experiment with different prompts, checkpoints, and settings.

Once you are comfortable with text-to-image, the next step is exploring LoRA models to control your image style and character consistency — or trying image-to-image workflows to transform existing images.

Next step: Learn how to use LoRA models in ComfyUI to take your image quality and consistency to the next level.