Google's Open-Source Wildlife AI Is Quietly Reshaping How Scientists Track Thousands Of Species

Key details: SpeciesNet wildlife identification model

SpeciesNet is an AI model from Google that classifies animals in motion-triggered camera trap images. It can currently recognize 2,498 animal categories across mammals, birds, and reptiles, using labels from about 65 million images supplied by conservation partners.

• SpeciesNet and the MegaDetector model work together, first detecting animals in images then assigning species labels and confidence scores.
• The system can process about 30,000 images per day on a standard laptop, according to Google.
• On a low-end gaming GPU, processing can reach 250,000 or more images per day.
• In tests across multiple camera trap projects, SpeciesNet detected animals in 99.4% of images that contained wildlife.
• It produced a species classification in 83% of cases, and 94.5% of those labels matched human verification.

Over the past year, researchers used SpeciesNet on camera trap images from Colombia, Idaho, Australia, and Tanzania's Serengeti, reporting detections of pumas, ocelots, elk, black bears, cassowaries, and lions. Todd Michael Anderson at Wake Forest University applied SpeciesNet locally to about 11 million images from the Snapshot Serengeti project.

Background on SpeciesNet and wildlife monitoring

Most contemporary wildlife monitoring projects rely on motion-triggered camera traps mounted on trees or other fixed locations. These devices capture short bursts of images when heat or motion triggers their sensors, creating large datasets that are difficult to classify manually.

SpeciesNet is part of Google Earth AI, which groups geospatial tools, datasets, and models for environmental analysis. Google first integrated SpeciesNet with Wildlife Insights in 2019, then released the model as open-source code in 2025. The GitHub repository includes code, documentation, and resources for adapting the model.

Several organizations have adapted SpeciesNet to regional species or integrated it into existing workflows. The Wildlife Observatory of Australia (WildObs) trained a version covering local animals such as musky rat-kangaroos and orange-footed scrubfowl. The Idaho Department of Fish and Game uses SpeciesNet as an initial classification step across hundreds of monitoring cameras, feeding into its broader species monitoring existing workflow.

Web-based and desktop tools now include SpeciesNet, such as The Nature Conservancy's Animl, AddaxAI, and Okala. Colombia's Humboldt Institute uses SpeciesNet within the national Red Otus camera network, which reached 446 cameras and more than 100,000 images in 2025. For organizations that prefer a managed platform, Wildlife Insights offers hosted SpeciesNet-based workflows.

Source citations and official references

Google Research described SpeciesNet in a March 6, 2026 blog post by Tanya Birch and Dan Morris. Technical performance details appear in a 2024 IET Computer Vision publication.

• SpeciesNet GitHub repository - official code, documentation, and model resources.
• Wildlife Insights and Wildlife Insights AI - Google Cloud platform for biodiversity monitoring and AI tools.
• MegaDetector GitHub repository - companion model for detecting animals in images.
• SpeciesNet performance paper - IET Computer Vision, 2024.
• Google Earth AI overview - geospatial AI tools collection including SpeciesNet.
• Wildlife Observatory of Australia - organization that trained a regional SpeciesNet model for Australian species.
• Animl documentation - The Nature Conservancy's camera trap data platform including SpeciesNet.
• AddaxAI and Okala - third-party applications that integrate SpeciesNet for biodiversity monitoring.
• Red Otus launch information - Humboldt Institute announcement on Colombia's national camera trap network.