Example of Our "On Brand" Models for Budweiser

How BrandGuard evaluates look and feel

One of the most difficult things about using generative AI for images is that you never quite know what to expect. The stochastic way these models work can surprise you in really great ways but, when you need something specific, those surprises are less desirable.

Below is an example where we used a text prompt to create an image for a beer advertisement. When compared against a model we build for “on brandedness” for Budweiser, this image scored 36% on brand.

Using a slightly modified prompt, we were able to get an image out, seen below, that looks subtly different (or vastly different if you are a designer), and scores very highly as “on brand” for Budweiser.

What do you think? Did the models get it right? Notice there are no logos or indication of what brand this might be - this is our core model that just considers the look and feel of the ad.

We’ve seen companies use this functionality when they generate many ads, sometimes thousands, using generative AI, and then use BrandGuard to rank them by how well they score. It allows marketing and design teams to harness the power of generative AI but saves time and money by not wasting time evaluating thousands of outputs. The human designers can focus on the best outputs of these generative models as scored by BrandGuard.

If you would like to try it out for your brand, please reach out to sales@brandguard.ai to schedule a demo.