AI Generated Logos

Like many of you, we've been exploring a wide range of AI image-generation capabilities over the last year. We recently stumbled upon a use case we hadn't explored yet: redesigning a logo. This isn't something that image models could do very well until recently, and the results are pretty mind-blowing.

How it works

We’ll take a pass at redesigning our Addition logo to demonstrate how this works. We’re going to use Midjourney v4, which currently has the edge on Stable Diffusion and DALLE-2 in terms of image quality and faithfulness to the prompt (although this changes from month to month).

To begin we provide the model an init image of our current logo to use as a starting point.

From there, we can begin prompting the AI and exploring various redesign options.

For example, when I provide it with the prompt: “the letter A :: futuristic logo :: outerspace :: illustration --v 4 --q 2”, I get back the following options:

Logo Redesigns

Ok, enough setup. Let’s look at some of our new AI-generated logo options.

“logo :: plants :: organic materials :: intricate details :: unreal engine”

“The letter A :: modern logo simplistic :: vector design :: symmetrical”

“The letter A made from light beams :: the plus symbol :: logo”

The letter A :: the plus symbol :: logo simplistic

Implications

Given how fast image generation models are advancing, these image-to-image techniques will likely get 10x better by this time next year. This represents an exciting new approach to branding and logo design. But moreover, these capabilities point to a broad set of use cases that have yet to be fully understood in a wide range of fields like product design, interior design, and architecture.