# Skill: Build and run the standard six-node text-to-image workflow in ComfyUI

> Lets an AI agent walk a user through the default ComfyUI text-to-image pipeline node by node, explain wire colors and data types, and produce a working first image -- whenever someone is running ComfyUI for the first time or debugging a broken basic workflow.

Source: sodigi·learn -- comfyui/pervyy-workflow · https://sodigi.io/learn/comfyui/pervyy-workflow

## When to use
- User is running ComfyUI for the first time and needs to understand the default workflow.
- User has a red wire or a type-mismatch error and cannot figure out what connects to what.
- User wants to understand what each node does before modifying the workflow.
- User's model list is empty or Save Image is not producing a file.

## Core rules
- The default workflow is six nodes in order: Load Checkpoint -> CLIP Text Encode (positive) -> CLIP Text Encode (negative) -> Empty Latent Image -> KSampler -> VAE Decode -> Save Image.
- Wire color equals data type; you cannot physically connect mismatched colors -- use this as a self-correcting mechanism.
- Wire color reference: pink/purple = MODEL, yellow = CLIP, orange = CONDITIONING, red = VAE, pinkish-red = LATENT, blue = IMAGE.
- Positive and negative CLIP Text Encode nodes are separate inputs into KSampler -- swapping them inverts intent.
- Empty Latent Image sets resolution; do not go far above the model's native training resolution on the first run.
- Do not load the default workflow by wiring from scratch -- use the built-in template (Image Generation) and only swap text.
- Finished images are saved to the `output/` folder on disk, not displayed only in the UI.

## Procedure
1. Open ComfyUI and load the default text-to-image template from the Workflows > Templates menu.
2. In Load Checkpoint, pick an installed model. If the list is empty, the file is missing from `models/checkpoints/`.
3. In the top CLIP Text Encode node (positive input to KSampler), type what you want in the image.
4. In the bottom CLIP Text Encode node (negative input to KSampler), type what to exclude (blurry, extra fingers, watermark, etc.).
5. In Empty Latent Image, leave the size at the model's default for the first run.
6. Click Queue. Wait for the image to appear in the Save Image node preview; the file is already in the `output/` folder.
7. To explore variations, change only the seed in KSampler and re-run. Lock a good seed before changing anything else.

## Ready-to-use prompt
```
Load the default ComfyUI text-to-image workflow template and generate an image with these settings:

Positive prompt: a ginger cat wearing round glasses, sitting by a window, soft morning light, detailed photo
Negative prompt: blurry, low quality, extra limbs, text, watermark

Use the default canvas size. Leave all KSampler settings at their defaults for this first run.

After generating, tell me:
- What each of the six nodes does.
- What the wire colors mean.
- Where the output file is saved on disk.
```

## Pitfalls
- Swapping positive and negative prompts -- "beautiful" in the negative tells the model to avoid beauty.
- Trying to drag a wire to a socket of the wrong color; if it will not connect, the target type is wrong.
- Setting a very large canvas size on the first run -- causes duplicated subjects and artifacts.
- Changing multiple KSampler settings simultaneously -- makes it impossible to know what changed the result.
- Looking for the output image inside the ComfyUI UI only -- it is always saved to the `output/` folder on disk.
- Starting with an empty Load Checkpoint list -- you need a downloaded model file in `models/checkpoints/`.