Creating Your First Complete Workflow

In this lesson, we'll build your first functional text-to-image workflow in ComfyUI from the ground up. By the end, you'll have a complete workflow that can generate images from text prompts using the essential nodes that form the backbone of most ComfyUI workflows.

Understanding the Basic Workflow Structure

Every ComfyUI text-to-image workflow follows a similar pattern:

1. Text Input → Text Encoding → Sampling → Decoding → Image Output
2. Model Loading nodes that provide the AI models needed for generation
3. Parameter nodes that control the generation process

Think of it like a factory assembly line where each node performs a specific function, and the connections between nodes determine how data flows through your workflow.

Step 1: Loading Your Model

Start by adding a Load Checkpoint node to your canvas. This node loads the Stable Diffusion model that will generate your images.

Load Checkpoint
├── ckpt_name: [Select your model file]
├── MODEL (output) → connects to sampling
├── CLIP (output) → connects to text encoding
└── VAE (output) → connects to decoding

The Load Checkpoint node outputs three essential components:

• MODEL: The neural network that generates images
• CLIP: The text encoder that understands prompts
• VAE: The decoder that converts latent space to actual images

Step 2: Setting Up Text Prompts

Add two CLIP Text Encode (Prompt) nodes - one for positive prompts and one for negative prompts:

Positive Prompt Node:

• Connect the CLIP output from Load Checkpoint to the "clip" input
• Enter your desired image description in the text field
• Example: "a beautiful landscape with mountains and a lake, sunset lighting, highly detailed"

Negative Prompt Node:

• Connect the same CLIP output to this node's "clip" input
• Enter what you don't want in the image
• Example: "blurry, low quality, distorted, ugly"

Step 3: Configuring the Sampler

Add a KSampler node, which is the heart of the generation process:

KSampler Connections:
├── model ← Load Checkpoint (MODEL)
├── positive ← Positive CLIP Text Encode (CONDITIONING)
├── negative ← Negative CLIP Text Encode (CONDITIONING)
├── latent_image ← Empty Latent Image (LATENT)
└── LATENT (output) → VAE Decode

Key Parameters:

• seed: Controls randomness (use -1 for random)
• steps: Number of denoising steps (20-30 is typical)
• cfg: How closely to follow your prompt (7-12 works well)
• sampler_name: Algorithm used (euler, dpmpp_2m are popular)
• scheduler: How steps are scheduled (normal, karras)

Step 4: Creating the Canvas

Add an Empty Latent Image node to define your image dimensions:

• width: Image width in pixels (512, 768, 1024)
• height: Image height in pixels (512, 768, 1024)
• batch_size: How many images to generate at once (start with 1)

Connect this node's LATENT output to the KSampler's "latent_image" input.

Step 5: Decoding to Final Image

Add a VAE Decode node:

• Connect KSampler's LATENT output to "samples" input
• Connect Load Checkpoint's VAE output to "vae" input
• This converts the latent representation back to a viewable image

Step 6: Saving Your Result

Finally, add a Save Image node:

• Connect VAE Decode's IMAGE output to "images" input
• Set your preferred filename prefix
• Images will be saved to ComfyUI's output folder

Testing Your Workflow

1. Queue Prompt: Click the "Queue Prompt" button to start generation
2. Monitor Progress: Watch the progress bar and any console output
3. View Results: Check the Save Image node or output folder for your generated image
4. Iterate: Adjust prompts, parameters, or seed values and run again

Troubleshooting Common Issues

• Red nodes: Usually indicate missing connections or invalid parameters
• Slow generation: Try reducing image size or step count
• Poor results: Experiment with different samplers, CFG values, or more descriptive prompts
• Out of memory: Reduce batch size or image dimensions

Next Steps

Once you have this basic workflow running, you can:

• Experiment with different models and samplers
• Add LoRA nodes for style modifications
• Include upscaling nodes for higher resolution
• Save your workflow as a template for future use

This fundamental workflow structure serves as the foundation for more complex ComfyUI creations. Master these connections and parameters before moving on to advanced techniques.