In the dynamic world of advertising, creativity is a key differentiator for best-performing ads. At Quilt.AI, we used a fine tuned AI model in order to decode creativity.

We first defined creativity by identifying three fundamental aspects:

Upon defining each creative element, we developed a creative training subset using the initial industry database of high performing ads. We further added to this list by identifying ads that have won the Cannes Lions awards or shown during the Superbowl. These events were used as a proxy to identify unusually creative ads.

Within this creative training subset, we had to be even more selective in order to truly reflect creativity. Considering each frame of the video, we removed transition scenes, single-color frames, and scenes that only contained logos and branding.

Following that, we embarked on a multi-dimensional exploration. We transformed the creative training dataset into vectors across multiple dimensions, by embedding each ad frame into a datapoint in a vector space. This process encapsulated more than mere visual elements like colors and shapes, reaching into the intricacies of photographic compositions and nuanced aesthetic principles.

Once we embedded the ad frames into vectors, we can do sophisticated semantic similarity scores, effectively converting “creativity” into “maths”.With the Creativity AI model, we are not merely classifying or categorizing; we are pioneering an analytical framework that connects the concept of creativity to tangible, actionable insights.

As we continually expand the creative training subset, the Creativity AI model will only grow more robust, further enhancing our ability to decode and apply creativity’s intricate dimensions. This journey under the hood reveals a complex but elegant dance of data and design, mathematics and art.