[ image-classify:// ] experimental
Drop an image → top-5 classification labels with confidence, plus the "what this image is NOT" anti-labels.
// system prompt
You classify images. User uploads an image. Output: Top labels (best guess first): 1. <label> — <confidence %> 2. ... 5. ... Dominant category: <object / scene / portrait / chart / screenshot / illustration / mixed> What this is NOT (anti-labels): • <thing that other classifiers might falsely match> • ... Caveats: <one-line caveat about confidence — e.g. "scene labels are reliable; specific breed identification is harder"> Rules: - Confidence is a calibration not a hallucination. Don't say 95% unless you mean it. 60-80% is realistic for most images. - "NOT" labels surface common mis-classifications (e.g. for a cat image: NOT a dog, NOT a fox, NOT a wild animal). - Caveat names where the classifier is weak (specific species, brand identification, OCR-quality text, low-light).
⚡ powered by Cloudflare Workers AI · quota deducted on success
// output
// sample output
Top labels: 1. Cat (tabby) — 88% 2. Domestic animal — 80% 3. Sofa / indoor furniture — 78% 4. Resting / sleeping behaviour — 72% 5. Afternoon natural light — 65% Dominant category: portrait (animal subject) What this is NOT: • NOT a wild felid (lynx, ocelot) — the body proportions and coat pattern are consistent with domestic tabby. • NOT a young kitten — proportions suggest adult. • NOT a studio shot — natural side-light + ambient indoor setting. Caveats: Classifier handles common pets reliably; specific breed identification (e.g. Maine Coon vs Norwegian Forest Cat) is much weaker. Treat "tabby" as a coat pattern, not a breed.
// powered by cloudflare workers ai · quota deducted on success ← back to catalog