Sign In

[ComfyUI] Simple Prompt & LORA Tester with Jobs

3

Jul 22, 2025

(Updated: 10 days ago)

workflows
[ComfyUI] Simple Prompt & LORA Tester with Jobs

This is beginner friendly automation workflow for ComfyUI.
If you wished for a simple way to find your favorite LORA strength, here it is.


This workflow allows to create many renders in one Run with multiple different parameters:

  • different LORA

  • different model and clip values for LORA

  • different prompts

  • any value you want to change

    We will start from basics.

image.png

This is basic graph for generation with "Save Image w/Metadata" in output.
Based on incoming TOS, all renders must have meta included, so here it is.


We will use some custom nodes (use ComfyUI Manager, or clone them from GitHub):

  1. comfyui-job-iterator - https://github.com/ali1234/comfyui-job-iterator
    Core Custom Node required for this workflow to work

  2. Save Image with Generation Metadata - https://github.com/giriss/comfy-image-saver
    Node for Image with Metadata. It's useful if you want to reload the whole workflow from one single image.

  3. Basic data handling - https://github.com/StableLlama/ComfyUI-basic_data_handling
    Nodes used for converting types of values and simple math.

  4. WAS Node Suite (Revised) - https://github.com/ltdrdata/was-node-suite-comfyui
    Nodes used for text handling in this workflow.

Now, we will implement Job Nodes into the graph.

image.png


Only five nodes for now,

  • Literal - Input a string of values, which will change each generation.

  • MakeJob - Naming the Attribute

  • JobToList - Turning Attributes into Jobs (each job is one generation)

  • GetAttributeFloat - Getting Attribute in specific type (in this case Float)

  • Divide - It's simpler to input whole numbers in "Literal" and later "Divide" them by 100, to get proper values for LORA strength.

image.png


With just this simple setup, we get three renders, one for each value in Literal.
As You probably noticed, values are input into "Strenght_value" for LORA (Velvet's Mythic Fantasy Style)

The first image has [0], second [40] and third [80].
Very simple and fast way to generate multiple renders and compare, which value is your favorite.

image.png

To make it clear and obvious which one is which, let's do something with the naming of renders.

We can use our "A" value. It's a Float, so we need to convert it into String and after that we can add "IMG_M" or whatever you want, using "Concatenate" node.

image.pngimage.png

Much better

Those were basics, now we can get to the real meat and potato.
How about testing "strength_model" and "strength_clip"?


MULTIPLE ATTRIBUTES

We can create multiple Attributes and Combine them in two possible ways:

  • product

  • zip

image.png

In this example we have to literals groups:
A = 0, 40, 80
B = 0, 35, 70

Product - generate for us every possible combination of jobs
(0,0), (0,35), (0,70), (40,0), (40, 35), (40, 70), (80, 0), (80, 35), (80, 70).

Zip - generate for us combination of first with first, second with second, etc.
(0,0), (40, 35), (80, 70)

Product is good, when you want to test many combinations. Zip, when you want to be more precise.

Here is example, how to use it in Clip and Model strength and naming:

image.png


And final renders with naming using Zip method:

image.png

PROMPT COMBINATIONS

We can use "Literal" to also include simple strings.
After combining those strings into jobs, we can "concatenate" them into "Clip Text" of our prompt.

image.png

Renders using "product" method, gave us all possible combinations.

image.png

SWITCHING PROMPTS

Using Switch (Any) and "Literal" as index numbers, you can change the prompt each generation. Using combined methods and multiple attributes, it's a simple way to mix in many different ways.

Workflow is simple, requiring understanding of a few nodes, but gives a wide range of options to generate many renders, while you are away from the computer.

image.png

And final renders:

image.png

Here is an example how to swap checkpoints, if you would like to test how your prompt will look many styles:

image.png

And renders in two different checkpoints:

image.png

Thanks for reading :)
Have fun with your generations!


Checkpoints used in article:
Nova Anime XL
WAI-NSFW-Illustrious-SDXL

LORA used in article:
Velvet's Mythic Fantasy Style

3

Comments