Technology
How Leafy identifies and diagnoses your plants
Leafy is built around a multimodal vision-language AI model — the same category of system used in modern visual search and image-understanding products. Instead of matching your photo against a fixed, hand-built plant database, Leafy sends the image directly to the model along with a structured prompt, and the model reasons over leaf shape, texture, vein pattern, growth habit, and color to produce a species identification, a Latin name, and a confidence level.
Image preprocessing
Every photo is resized and re-encoded server-side before it reaches the model, keeping uploads fast on mobile connections without losing the detail identification depends on.
Structured output
The model is prompted to return a strict JSON object — species name, Latin name, confidence, and a short care note — so results render consistently in your collection every time.
Retry with backoff
If the AI provider is briefly unavailable, Leafy automatically retries with exponential backoff so a network hiccup doesn't turn into a failed identification.
Confidence, not certainty
Every identification carries a high, medium, or low confidence rating so you can judge how much to trust a given result before you act on it.
Health diagnosis, from multiple photos at once
Diagnosis works the same way, but with more inputs: when you upload close-ups of leaves, stems, soil, and roots, Leafy sends all of them to the vision model in a single pass, labeled by plant part. The model cross-references what it sees across the photos — yellowing on a leaf plus dry, pulling-away soil, for example — to distinguish overwatering from underwatering, pest damage, or a nutrient deficiency. The response comes back as a severity rating, a plain-language summary, and a short list of verb-led guidance steps, so the output reads like advice rather than a raw model dump.
Because the identification and diagnosis are generated by an AI model rather than looked up in a fixed reference table, Leafy can respond to plants and symptom combinations that a static database would miss — but it also means results are estimates. Treat a low-confidence ID or a diagnosis as a strong starting point, and consult a local nursery or extension service for anything high-stakes (a valuable specimen, a suspected toxic plant, or a large-scale garden problem).
Your plant photos, identifications, and diagnoses are stored against your account in a managed database so your collection persists across sessions and devices. See the privacy policy for details on data retention and the third-party processors involved.