TRIBE v2 Translates Everyday Media Into Virtual Brain Maps
TRIBE v2 is an open-source software framework that translates videos, audio clips, and written text into detailed predictions of human brain activity. The system uses machine learning to map sensory inputs directly onto simulated brain surfaces.
Meta’s research team built the tool after processing more than a thousand hours of medical imaging scans collected from seven hundred twenty participants. Researchers designed the model to run complex cognitive experiments entirely on a computer, removing the need for live clinical trials.
Model Size: 0.7GB & VRAM GPU: requirements vary
Predicting neural responses from everyday media
- Accepts video, audio, or plain text files as direct input triggers.
- Generates high-resolution activation maps across twenty thousand simulated brain points.
- Automatically shifts output timing to account for natural blood flow delays.
- Converts written content into synthetic speech before analyzing linguistic patterns.
- Provides ready-made visualization modules for inspecting cortical activity.
Independent analysts can download the package to run cognitive simulations without accessing restricted medical hardware. The included scripts handle file conversion and data alignment through standard command prompts, allowing users to focus on interpreting results rather than managing infrastructure.
Research notes and TRIBE v2 limitations
The team emphasizes that predictions reflect an averaged neural profile rather than the unique wiring of any single person. Running the code locally requires a standard Python environment, though training custom versions demands substantial compute resources.
"TRIBE v2 is a deep multimodal brain encoding model that predicts fMRI brain responses to naturalistic stimuli,"
noted the developers in a repository description. Early adopters should note that the software currently operates under a non-commercial license, restricting enterprise deployments.
The pretrained weights and source code live at GitHub, while the full checkpoint files are available at Hugging Face.
