Neuroimaging experiment design
Pre-screen video, audio, image, or language stimuli before committing scanner time and participant resources.
Independent tool overview
TRIBE v2 at a glance
TRIBE v2 is Meta FAIR's research model for predicting high-resolution fMRI brain responses to video, audio, and language. Meta released the code and pretrained weights for noncommercial use, but this is a neuroscience toolkit for researchers—not a consumer app, diagnostic product, or mind-reading system.
Visit the official TRIBE v2 site ↗
- Developer
- Meta FAIR
- Inputs
- Video, audio, and text
- Output
- Predicted fMRI activity on a cortical mesh
- Research data
- 1,117.7 fMRI hours across 720 subjects
- Released checkpoint
- Average-subject cortical prediction
- License
- CC BY-NC 4.0
Overview
What TRIBE v2 is
TRIBE v2 maps representations from video, audio, and text onto predicted activity across the brain. Researchers can supply a media file, generate a time-aligned event representation, and receive predicted responses on a standard cortical mesh.
The research used more than 1,000 hours of fMRI across 720 subjects, combining deep datasets with extensive observations per participant and wider datasets covering larger groups. The released checkpoint predicts an average subject and can generalize to unseen subjects and tasks without collecting individual calibration data.
The model is built for in-silico neuroscience: piloting stimuli, reproducing known response patterns, studying multisensory integration, and testing research hypotheses before expensive scanning. Its outputs remain model predictions and must not be presented as measured responses from a specific person.
Use cases
Who TRIBE v2 is best for
The strongest fit depends on the job you need the product to complete, not the size of its feature list.
In-silico neuroscience
Test whether predicted cortical patterns reproduce established visual, language, and multisensory effects.
Brain-encoding research
Compare a deep multimodal encoder with linear baselines on held-out fMRI datasets and new experimental conditions.
Model interpretation
Analyze learned components and spatial response maps to study how modalities contribute across cortical regions.
Capabilities
Core TRIBE v2 features
Tri-modal encoding
Combines video, audio, and language representations in one Transformer-based brain-response model.
High-resolution cortical output
The public quick start produces predictions on the fsaverage5 cortical mesh with roughly 20,000 vertices.
Zero-shot subject generalization
Predicts group-average responses for unseen participants without requiring a personalized scan during inference.
Research fine-tuning
The paper describes adapting the model with limited participant data to improve subject-specific prediction.
Open code and weights
Meta provides the training and evaluation code on GitHub and a pretrained checkpoint through Hugging Face.
Visualization utilities
Optional dependencies support cortical plotting and region-of-interest analysis with standard neuroimaging tools.
Process
How the TRIBE v2 workflow works
Step 1
Confirm the research and license fit
Use TRIBE v2 only for permitted noncommercial work and complete the appropriate ethics and data-governance review for any human-subject research.
Step 2
Prepare a Python environment
Clone the repository, use Python 3.11 or later, and install the inference or optional plotting dependencies.
Step 3
Load the pretrained checkpoint
Download facebook/tribev2 through the repository's TribeModel helper and select a local cache directory.
Step 4
Create aligned events
Pass a video, audio, or text input through the provided event-dataframe utility so modalities have the timing information needed by the model.
Step 5
Generate brain-response predictions
Run inference to produce the subject-average cortical time series, accounting for the documented five-second hemodynamic offset.
Step 6
Validate scientifically
Compare predictions with measured data, preregister hypotheses where appropriate, report uncertainty, and avoid individual or clinical claims unsupported by the model.
Cost
TRIBE v2 pricing and free plan
Meta publishes the TRIBE v2 repository and pretrained weights without a subscription fee under CC BY-NC 4.0. Users pay their own compute, storage, engineering, and research costs. The license does not grant commercial use, so organizations considering a paid product need separate permission from Meta.
Research code and weights
Free
Self-hosted model release for permitted noncommercial use.
- GitHub training and inference code
- Pretrained weights on Hugging Face
- CC BY-NC 4.0 license
- User supplies compute and storage
Commercial use
Not included
The public license does not authorize commercial deployment.
- No public commercial price
- Separate permission may be required
- Do not assume cost recovery or a paid service is permitted
Hosted API
Not offered
Meta does not list a production TRIBE v2 API plan.
- Run inference in your own environment
- Infrastructure costs vary by workload
- Interactive demo is for evaluation, not a service-level deployment
Pricing checked . Check current pricing at the source ↗
Assessment
TRIBE v2 strengths and limitations
Where it stands out
- Unifies video, audio, and language in one brain-encoding architecture
- Trained and evaluated across more than 1,000 hours of fMRI and 720 subjects
- Supports zero-shot group-average predictions for unseen subjects and tasks
- Code, pretrained weights, paper, and demo are publicly available
- Includes an accessible inference helper and optional brain visualization tools
- Useful for piloting expensive neuroimaging experiments
What to consider
- It predicts brain activity; it does not measure a person's current neural response or read thoughts
- The public checkpoint represents an average subject rather than a specific individual
- The released license is noncommercial and does not cover paid products
- There is no managed production API or service-level guarantee
- fMRI limits temporal resolution and cannot capture millisecond neural firing dynamics
- The public checkpoint and dependency stack may require research engineering to install and reproduce
- It is not validated as a diagnostic, treatment, audience-engagement, or individual decision-making tool
Compare
TRIBE v2 alternatives
The right alternative depends on the specific output, workflow, controls and budget your project requires.
Miscellaneous
SAM Audio
Choose SAM Audio for an open Meta research model focused on separating sounds rather than predicting neural responses.
Explore SAM Audio →Miscellaneous
Sam 3D
Choose SAM 3D for Meta research workflows centered on reconstructing objects or people from images.
Explore Sam 3D →Coding
Meta Llama Models
Choose Meta Llama when the task is general language generation or reasoning rather than fMRI encoding.
Explore Meta Llama Models →Questions
TRIBE v2 FAQs
What does TRIBE v2 do?
TRIBE v2 predicts high-resolution fMRI brain-response patterns from video, audio, and language inputs. It is intended for computational neuroscience and in-silico experiments.
Can TRIBE v2 read minds?
No. It predicts an average brain response to supplied stimuli; it does not decode a person's private thoughts or directly measure an individual's current brain activity.
Is TRIBE v2 open source?
Meta provides the code and pretrained weights publicly under CC BY-NC 4.0. That permits many noncommercial research uses with attribution but does not grant commercial use.
How much does TRIBE v2 cost?
There is no subscription fee for the public research release. Users provide their own compute and storage, and the noncommercial license does not include a paid production deployment.
What data was TRIBE v2 trained and evaluated on?
The paper reports a combined 1,117.7 hours of fMRI across 720 subjects, using video, audio, text, and controlled experimental conditions across multiple datasets.
Bottom line
Our TRIBE v2 verdict
TRIBE v2 is a significant open research release for neuroscientists who need multimodal, high-resolution brain-response predictions and can operate a Python-based fMRI workflow. It should be evaluated as a scientific model with noncommercial licensing and substantial validation requirements—not repackaged as a consumer mind-reading, clinical, or engagement-scoring product.
Visit TRIBE v2 website ↗Similar Science Tools
View all Science AI tools →
Gemini 3.1 Deep Think
Gemini 3 Deep Think - Google's upgraded AI reasoning mode now solving PhD-level math and science research problems
Claude Science
Claude Science - Anthropic’s app for AI-powered scientific research
Nomos 1
Nomos 1 - Nous Research’s powerful open-source AI math reasoning system
Leanstral 1.5
Leanstral 1.5 - Mistral's open model for writing verified math proofs
AI training for the future of work.
Get access to all our AI courses, hundreds of real-world AI use cases, live expert-led workshops, an exclusive network of AI early adopters, and more.
AI Courses
Get unlimited access to all of our current & upcoming industry-specific AI courses for the duration of your subscription.
Daily Guides
To keep up with the rapid pace of AI, our team publishes AI implementation guides daily. Our library contains 300+ practical use cases to automate real-world work.
Workshops
Join weekly, live, interactive sessions with industry leaders who are at the forefront of AI for hands-on implementation guidance and exclusive insights.
Community
Network with an exclusive community of AI-first professionals who are working smarter with AI. Learn how early adopters are using AI in their work and businesses.