ComfyUI LoRA (Model vs. Clip)

Problem:

When using LoRAs, I have no idea what strength_model and strength_clip had on my output image.

Experiment:

Therefore, I selected my most commonly used LoRAs and did a simple test to see how strength_model impacts image results. I didn't use strength_clip as I discovered that it had little impact on achieving the desired results of the LoRA.

Attached is a simple workflow of my experiment.

I controlled for all variables except the LoRA strength_model, which varied from

• 0.00

• 0.25

• 0.50

• 0.75

• 1.00

Results / Findings:

1. strength_model refers to the weight added from the LoRA (training images)

2. strength_clip refers to the weight added from the clip (positive and negative prompts)

In general, most people will want to adjust the strength_model to obtain their desired results when using LoRAs.

You only want to use strength_clip when there is something specific in your prompt (keyword, trigger word) that you are looking for.