What it does
This KPI measures an asset’s potential to activate and arouse, using a deep vision model. It quantifies how likely an image or video is to physiologically engage and stimulate viewers, reflecting its ability to capture attention, evoke emotion, and prompt a response.
Why it matters
Arousal in advertising is critical because it influences the viewer's level of alertness and attention. It ranges from low (as in deep sleep) to high (as in intense excitement or fear). High arousal can increase attention to assets, enhance emotional engagement, improve message processing, and stimulate action. Achieving the right balance is key, however, as too much arousal can lead to discomfort or negative reactions.
How it works
The image is analyzed by a deep learning model trained to predict basic human motivations, including those relevant to arousal (e.g., excitement). The final score, ranging from 0 to 100, indicates the overall arousal potential of the image, taking into account these core motivational elements.
How to achieve great results
The illustration shows sample images with model predictions. Top left image has lowest, bottom right image has highest activation potential.
Image activation is driven by dynamic and energetic visuals which trigger excitement. Images with high activation (arousal) tend to involve dynamic activities, intense emotions, or visually striking elements. In contrast, low-activation images often depict static scenes, mundane objects, or calm, neutral expressions that lack emotional intensity or action.
Examples with assets:
AI models used
Arousal/valence models at aimpower are based on the latest deep learning frameworks used in scientific literature for the purpose of emotion modelling, including convolutional deep learning networks and vision transformers. The images used to collect human ground truth data originate from various sources, including social media platforms, large scale webscrapes (Common Crawl) and marketing materials (e.g. ads).
In total we leverage 500 Mio+ images/text pairs for pretraining and 15K+ images for finetuning our models with human ground truth data. Human ground truth is collected via online / web-based tasks. Evaluation of our models follows the scientific best practices. Our model predicts arousal ratings with r = 0.88 (P < 0.0001; RMSE, 0.6743).