CoursesCreative Prompting BootcampShort Video and Repeatable PipelinesRepeatable Pipelines: Seeds, Settings, and ComfyUI
lesson
40 min

Repeatable Pipelines: Seeds, Settings, and ComfyUI

Learn what makes an AI image repeatable, and how node-based tools like ComfyUI save every step.

What you will learn

By the end of this lesson, you can record the settings that make an image repeatable and explain how a ComfyUI workflow works.

The problem: prompt roulette

You make a great image on Tuesday. On Friday the client asks for the same image in a vertical size. You run the same prompt and get something different. You cannot get it back.

This is prompt roulette. It happens because each run starts from different random static. It also happens when the tool, the model, or the settings changed and nobody wrote it down.

What makes a result repeatable

SettingPlain meaning
Tool and modelWhich program and which version made it, like "Whisk" or "Stable Diffusion XL"
PromptThe exact words, copied, not retyped from memory
Negative promptWhat you told it to leave out (some tools have a separate box)
SeedA number that sets the starting static. Same seed, same start.
SizeWidth and height, or aspect ratio like 1:1 or 9:16
Steps and guidanceHow many cleanup steps, and how closely it follows the prompt (in tools that show these)
InputsAny reference images you used
DateTools change. The date helps explain why results changed.

Simple tools like Whisk do not show every setting. Record what you can see. Always save the prompt, the inputs, and the date.

How ComfyUI fits

ComfyUI is a free, open-source tool for Stable Diffusion models. It runs on a computer with a strong graphics card, or on a cloud service. You build images by connecting boxes called nodes.

Each node does one job:

  1. Load a model.
  2. Read your prompt.
  3. Set the size and seed.
  4. Clean up the static (the sampler).
  5. Turn the result into a picture.
  6. Save the file.

You connect them with lines, like a flowchart. The whole flowchart is called a workflow. You can save a workflow as a file and share it. Anyone who opens it, with the same model, can make the same image.

That is the real lesson of ComfyUI, even if you never install it. A pipeline is only repeatable if every step is written down.

ComfyUI is optional in this course. If you have access through a partner lab, your instructor will show you. If not, you can do everything in the capstone with Whisk, Flow, and a good prompt library.

Judgment for open models

Open tools give you more control. They also give you more responsibility.

  • Some downloadable models were trained on images taken without permission, or were built to copy specific people. Use models from known sources with clear licenses.
  • Open tools may have fewer safety filters. Your rules from Module 1 still apply in full.
  • Keep client and personal photos off shared computers.

Practice: build a record

The editor below has a small script that checks a prompt library entry. Fill in the missing fields. The console tells you what is still missing. Keyboard users: press Esc, then Tab, to leave the editor.

Prompt record checker

Fill in every empty field in the entry with your own image's details. Run it until the console says the record is complete.

Loading the code editor...

Press Esc, then Tab, to move focus out of the editor.

Try this: Take an image you made in Module 4. Write a full record for it. Then ask a classmate to make it again from your record alone. How close did they get?

Go deeper: ComfyUI, Uncomplicated walks through node workflows step by step.