Google open-sources SpeciesNet and launches nature AI initiatives

SpeciesNet is an open-source computer vision model from Google designed to identify wildlife species from camera trap photos.

Anyone conducting field conservation knows the drill: motion-activated cameras sit in forests for months, capturing hundreds of thousands of images. Manually sorting out blank triggers and classifying individual animal species can take researchers entire seasons.

What it does

SpeciesNet automates the tedious pipeline of wildlife photo classification. It filters out empty frames caused by wind or vegetation movement and predicts specific animal species directly from raw camera trap footage.

This is not an untested model. Since 2019, thousands of wildlife biologists have used SpeciesNet via Wildlife Insights, a Google Cloud-powered conservation platform. Open-sourcing the standalone model lets tool developers, academics, and conservation startups deploy the weights locally or weave them into custom software without platform lock-in.

Accelerator and $3M grant support

SpeciesNet is the technical centerpiece of three nature initiatives announced by Google on March 3. The company is backing the open release with funding and mentorship.

First is the Google for Startups Accelerator: AI for Nature, Google's first cohort dedicated to nature and biodiversity startups. Open to ventures across the Americas, the 10-week virtual program pairs founders with Google engineers for technical mentorship. Applications run from March 3 to March 31, 2025, with programming kicking off in May 2025.

Second, Google.org committed $3 million to the Instituto Clima e Sociedade (iCS) to fund AI-enabled conservation projects across Brazil. The grants will back local nonprofits and research centers tackling three focus areas: reversing biodiversity loss to support the global 30x30 targets, developing sustainable bioeconomy value chains, and scaling regenerative agriculture for smallholder farmers.

Why it matters

Field ecology has always suffered from a data bottleneck: collecting sensor data is relatively cheap, but cataloging millions of images by hand is painfully slow and expensive.

SpeciesNet puts a field-tested, production-grade vision model straight into the hands of independent researchers and developers. If you build conservation hardware or manage protected reserves, you no longer have to reinvent animal detection algorithms from scratch.

Google's approach makes sense here. Handing out compute credits or cash grants is fine, but open-sourcing the underlying model architecture removes the technical tollbooth for researchers who need to run inference offline on local edge devices.