The discovery studio.
Explore a photo or a sound, get tentative clues and learn what evidence to check next.
Your media is analysed on this device. Model files are downloaded from Hugging Face; the analysis runtime is hosted here. No photograph or sound is sent to those services or saved by this studio.
What did you find?
Choose a clear photograph with one animal filling the frame. A second angle helps you check the result.
First use downloads approximately 90 MB of model weights plus runtime files. A desktop browser and Wi-Fi work best. Downloaded models may be cached by your browser.
Keep an open mind
EXPERIMENTALThe evidence starts with you.
Automated suggestions can be wrong. Choose a file, run a comparison and use the field key to test each lead.
How to make the clue stronger
Record the place, time, weather and habitat. For photographs, include scale and useful body details. For sounds, record a quieter stretch and note whether you saw the caller. A recording can contain several animals.
No suggestion determines whether an animal is safe to handle or how to care for it. Keep unknown identifications broad and your pets’ feeding decisions in the care hub.
Models & methods
Photo groups use a general ImageNet-trained Vision Transformer, with narrow class labels generalised to broad groups. Audio uses an AudioSet-trained Audio Spectrogram Transformer. Neither model is calibrated here for Victorian entomology.