Visual components determine the credibility of AI-generated videos

Alba Junquera

To analyze the degree of visual fidelity and credibility of videos generated with Artificial Intelligence (AI), the group cyberimaginary The URJC conducted a quasi-experimental design, organizing participants into two predefined groups: a control group and an experimental group. In the control group, participants watched two real videos obtained from the YouTube platform.

After viewing the video, participants completed an evaluation form assessing key visual components: lighting, saturation, color, scene composition, and perceived credibility. Using these results, researchers conducted a quantitative and descriptive analysis, based on frequencies and percentages, to evaluate variables such as visual appeal, composition, and realism, while also considering the participants' sociodemographic profile. An analysis of the recognition of the locations shown was also performed.

Analyzing the videos generated with Sora, researchers observed that the lighting and saturation in the Santorini footage were exaggerated, leading participants to perceive the video as artificial. “These findings indicate that the perceived realism of AI-generated videos depends on a balanced combination of visual components, where aesthetics are key for the content to be perceived as authentic and credible,” explains researcher Alberto Sánchez.

El study It was developed by two research groups: the Ciberimaginario group at URJC and VARLab (Virtual and Augmented Reality Lab) at the University of Bologna (Italy). The project was carried out within the framework of MICS (Made in Italy–Circular and Sustainable) and received funding from the European Union's Next-GenerationEU program.

The results obtained offer developers of generative artificial intelligence tools the opportunity to optimize their models and produce more natural and believable content. They also have applications in the fields of communication, marketing, and audiovisual media, since the improved image quality allows for their integration into advertising campaigns or audiovisual narratives.

Furthermore, the project provides invaluable insights into media literacy, as it helps us understand how people perceive and distinguish between real and AI-generated content. This is crucial for combating disinformation and the deepfake (ultra-fake content) of everyday life.

The study concludes that, although AI has advanced significantly, further improvements are needed to achieve complete realism in images and videos, especially regarding lighting and color. Artificial intelligence tools must continue to be refined to achieve greater authenticity.