A Look at HOW You Say It

Understanding AI Interpretation and Output

Every AI has a unique way of interpreting and processing requests. I started the "Mid-24 AI Face-Off" series to compare where various AI platforms are at this point in mid-2024. This series aims to help people see the differences, understand the nuances of interpretation, and better equip themselves for prompting in their preferred platform.

This post, while slightly overlapping the series, is more of a standalone article inspired by a LinkedIn post I saw comparing Midjourney and Adobe Firefly using the same prompt. I wanted to explain the importance of how you phrase your prompts and note the variety in results using the same prompt across different platforms.

Midjourney

Adobe Firefly

What is the point?

Every AI has a slightly different way of interpreting and processing requests for output. As exhibited in the series I started, "Mid-24 AI Face-Off" where I began doing a detailed prompt run to see how each platform interprets a request - making adjustments to the original prompt as needed to get as close to the desired result as possible. Midjourney, Leonardo.ai, Musavir... and the list will continue.

I wanted to share the prompts and results. Sure, Midjourney does a more pleasing almost 'finished' result - and is more capable of understanding more nuanced language, sometimes you will need to use more or less details to obtain the result you are trying to achieve.

If you use the exact same prompt (minus the parameters) from Midjourney in Adobe Firefly - you will get some skewed results. Here’s an example prompt:

Fashion photography, fair-skinned Caucasian female model wearing blue-toned red lipstick to highlight skin tone contrast. Avoid orange undertone.

In Adobe Firefly, this prompt produced the images seen below - with two clear female figures, one ambiguous male, and one clearly male figure, all with bold makeup and glistening skin. Additionally, two out of four figures were not Caucasian, the lipstick tone was incorrect, and all images had an orange undertone color cast. In other words - this prompt failed to accomplish the intended result in Adobe Firefly.

This demonstrates that a prompt working well on one platform doesn’t guarantee practical and useful applications elsewhere. To use any AI tool effectively, it’s essential to learn its nuances and understand how it interprets prompts. Simply because it works in one platform, doesn’t create a blanket pass for practical and useful application elsewhere. Learn the tool if you want to use it, understand its nuances of interpretation, then use it well.

Happy prompting!

Brian Sykes

> Look at the prompts and results from Midjourney and Adobe Firefly above. While Midjourney still excels in the photo-quality, what about the structuring causes Firefly to be ‘more’ accurate, than when using the same prompt from Midjourney? Observe and Learn.