Branding in the Age of AI: Starbucks Case Study

AI generations — On brand, Starbucks

Is it possible to train an AI model based on brand guidelines?

The rise of AI-generated imagery is revolutionizing the world of art and design. Suddenly, everyone has the tools to become an artist, without spending the hours needed to hone the craft. However, when it comes to branding, a critical challenge emerges: ensuring consistent brand outputs. AI, inherently statistical and somewhat unpredictable, presents hurdles for designers aiming for consistency. So, I embarked on a journey to discover whether we can train an AI model to adhere to specific brand guidelines.

Crafting Custom Models

Data Set

Every AI journey starts with the data set. Think of it as the foundational learning material for the AI. When customising a visual model, this means amassing a collection of high-resolution images representative of the desired output.

Deep Learning

Once equipped with a dataset, it is fed to the AI who performs deep learning. Here, it identifies and catalogues patterns within the data. This vast array of patterns, stored in the AI’s latent space, remains ready for deployment, waiting to be transformed into pixels upon request.

Custom Model Implementation

Post the deep learning phase, tools like Automatic1111 enable the application of the trained model to generate desired images.

With this in mind (and code), I decided to experiment with one of the most recognizable brands globally: Starbucks.

Starbucks: A Case Study

Illustrations

Coming from a gaming background, I started with illustrations, an area I’m most familiar with. Starbucks illustrations are uniquely recognizable, marked by their analog texture and slightly asymmetric feel. Their guidelines are listed here.

The dataset used of Starbuck’s illustration model

Utilising this as my data set, the resulting images from my custom model showcased the AI’s ability to understand and replicate distinct patterns. For instance, the AI adeptly captured the unique hair pattern from the dataset, even when illustrating a different subject. This reaffirms the potential of AI; it can grasp nuances in design which might be challenging to articulate in words.

Art generated from the fine-tuned model

Notice the line of her hair

Photography

For the photography segment, I used eight distinct images, capturing Starbucks’ product photography guidelines. This data set was chosen to help the AI discern the brand’s photographic style, especially the distinct angles and the intricacies of the Starbucks logo.

The data set used for the photography model

The outputs? Generated images of Starbucks products that stay true to the brand’s photographic style. Additionally, with a custom model for a brand ambassador, say Gal Gadot, AI-generated images eliminate the need for studio shoots.

AI generations of the fine-tuned model

Not real Gal Gadot is not really holding a fake Starbucks coffee

Colors

Starbucks’ color palette is iconic. By integrating eight of these brand-specific colors into my model, I achieved a high accuracy rate. Inputting specific color preferences, the model produced images with the desired color about 80% of the time.

The colors used for the data set

And it was all yellow

Let it be teal!

Harnessing the Model’s Potential

With a brand-aligned AI model in hand, the possibilities are limitless. It enables the creation of myriad branded content, be it redesigning coffee packages, conceptualizing web designs, crafting wedding invitations, or generating numerous postcards to select the best.

Rebranding Starbucks coffee packages with my fine-tuned model & some inpainting magic

Mockups for a website and an invitation

Creating myriad of postcards

Choosing the one we love best, because how can’t we he’s adorable!

Brands in the AI Era

While AI tools like Midjourney and Firefly offer impressive visual capabilities, being closed models, they often fall short in maintaining brand consistency. Hence, the future for branding in the AI realm lies in custom models. And the implications are profound:

Democratising Brand Creation

Whether it’s the official brand artist, a company employee, or a fan, everyone can create on-brand visuals. The next groundbreaking idea might just come from an unexpected source.

Redefining Workflows

Instead of focusing on a singular design, brands can generate a plethora of options and cherry-pick the best.

Strategic Vision

Transition from viewing AI as just a tactical tool to understanding its broader potential. In the AI-driven future, brand visuals could be part of an intricate system, tailoring visuals based on individual user preferences, informed by behavioural patterns.

In the evolving landscape of AI and branding, those who harness custom models’ potential will truly set themselves apart. So, everyone, the time is ripe to own your models.

Dori Adar is the CEO of Hands on Games, a consulting group that help art teams around the world make the leap into AI creation. This is our website.