Cerebrum: A Neural Engagement Video Analysis Platform Based on Meta TRIBE v2

Cerebrum is an advanced web platform that uses Meta's TRIBE v2 multimodal brain encoding foundation model to directly predict local fMRI-like brain activity from standard MP4 videos. This platform can analyze the impact of video content on human cognitive and emotional responses, providing content creators with neuroscientific insights.

The core technical support of Cerebrum comes from Meta's TRIBE v2 (Temporal Representations for Inference of Brain Encoding) model. It is a multimodal brain encoding foundation model that can map visual and auditory inputs to predicted brain activity patterns. Traditionally, understanding the impact of videos on the brain required expensive fMRI experiments, but TRIBE v2 uses deep learning technology to achieve non-invasive inference of brain activity predictions from video content.

The breakthrough of TRIBE v2 lies in its ability to predict the response intensity of different brain regions to video content without actually scanning the audience's brain. This provides content creators with an unprecedented tool—they can predict the neural-level engagement of their content before it is released.

The Cerebrum platform provides a complete set of video neural analysis functions:

The platform analyzes the activation of six key brain regions, which are associated with different types of cognitive and emotional processing. Through radar charts and heatmap timelines, users can intuitively see how the video activates different brain regions at different time periods. This fine-grained analysis helps creators understand which parts of their content are most cognitively engaging.

Cerebrum supports side-by-side comparison analysis of multiple videos. Users can upload and compare different versions of video content simultaneously, observing differences in neural engagement metrics. This is particularly valuable for A/B testing and content optimization—creators can scientifically verify which version of content better stimulates the audience's brain response.

Analysis results can be exported in CSV and PDF formats, facilitating further data analysis and report creation. This open data format allows the platform to integrate into larger content analysis and decision-making workflows.

The platform integrates Google OAuth login to ensure the security of user data. Users can upload and manage multiple video files, and the system saves a complete analysis history for each video.

The developers of the Cerebrum project conducted a thought-provoking study: they analyzed 20 videos (10 viral YouTube Shorts with over 100,000 views; 10 non-viral videos) to find the correlation between neural engagement metrics and social media viral spread.

The results were unexpected: no correlation was found between the two.

Specific data shows that the average brain activation of viral videos is between 0.38-0.40, while that of non-viral videos is between 0.37-0.40, and the difference is statistically insignificant.

This finding has important practical implications: it indicates that viral spread is mainly driven by algorithmic factors, not just neural engagement. High-quality, highly engaging content may not get algorithmic recommendations, and viral content may not be more cognitively attractive.

For content creators, this means they need to distinguish between two different goals: creating content that deeply engages the audience cognitively, and creating content that can be recommended by algorithms. The two may require different strategies.