What they built.
Jay at Hound Hill Homestead experimented with using camera images to determine whether birds were inside a coop. His Home Assistant setup combined a Wyze camera through Frigate, Google's Generative AI integration and ESPHome hardware. The published account includes screenshots and YAML configuration, making the implementation inspectable.
Gemini 2.0 Flash is explicitly named for the initial image tests, but the automation configuration does not independently establish the deployed model version. The author also reports missed birds, including one outside the camera view and another heavily obscured. Those failures are central to understanding the example.
The author's photos and full YAML document a real experiment, including failures involving an off-camera bird and a heavily obscured bird.
Read the builder's experiment and caveatsHow the tools work together
The system takes a snapshot, asks the model about bird presence or count and stores the response in Home Assistant for observations and notifications. The original build also connected an answer to a door relay. That last step turns an uncertain visual interpretation into a physical action, which is a substantially different responsibility from showing a dashboard reading.
Workflow at a glance
Take a periodic snapshot
A Wyze camera connected through Frigate observes the coop.
Request an observation
Gemini receives the image and answers whether birds are present or provides a count.
Show the observation
Home Assistant stores the result for a dashboard and phone notifications. Treat the response as observation assistance, not permission to move a door.
How you could adapt the idea
Treat this as an observation-assistance experiment, not a safe standalone animal-door controller. A reader exploring the idea should keep the result informational, compare it with manual observations and account for what the camera cannot see. Any physical door action needs independent safeguards and verification; the published experiment does not establish the reliability needed to remove them.
Effort, cost and limitations
- Effort
- Advanced camera, Home Assistant and hardware integration. Not a turnkey starter project.
- Cost
- No complete project cost is established by this case study.
This is not a safe standalone animal-door controller. Any physical door action needs manual verification and independent physical safeguards; the vision model can miss animals.
Go to the original work.
Builder's photographs, caveats and YAMLSource-backed editorial account; PrisAIm did not run or independently benchmark this implementation.