Local Image Classifier
Identify what's in a photo using MobileNet, a real image classification model trained on the 1,000-category ImageNet dataset, running entirely on your device. Returns the top 5 most likely labels with their own confidence percentages rather than forcing a single guess. Nothing about your photo is ever uploaded — the model (~17MB) downloads once and every photo after that classifies completely offline.
About this tool
This tool identifies what's in a photo — a dog breed, a type of food, a vehicle, a household object — using MobileNet, a real image classification model trained on the standard 1,000-category ImageNet dataset, running entirely on your device. It returns the top 5 most likely labels, each with its own confidence percentage, rather than forcing a single guess — useful when the correct answer is close between two similar categories. This is whole-image classification ("what is this a picture of"), not object detection ("where are the objects in this picture") — for drawing boxes around multiple objects in one photo, see the Object Detector instead. Nothing about your photo is ever uploaded — the model (~17MB) downloads once and every photo after that classifies completely offline.
What you can do
- Identify what's in a photo — 1,000 everyday object, animal, and scene categories
- Runs a real image classification model on your own device, not a server
- See the top 5 predictions, each with its own confidence percentage
- Your photo is never uploaded — classification runs entirely in your browser
How to use
Step 1
Upload a photo
The model recognizes 1,000 ImageNet categories — animals, objects, food, scenes, and more.
Step 2
The model classifies it
Runs entirely on your device via WebAssembly/WebGL — nothing is sent anywhere.
Step 3
See the top 5 results
Each prediction comes with its own confidence percentage, most likely first.
Example
A photo of a golden retriever:
Top predictions
golden retriever — 87% Labrador retriever — 6% kuvasz — 2% cocker spaniel — 1% tennis ball — 1%
the model is confident about the top result here, with the remaining probability spread across visually similar breeds — typical of a clear, well-lit photo of a common category.
How your data is handled
This tool is labeled 🔵 Local AI: your photo is read directly using your browser's File API and classification runs entirely on your device via TensorFlow.js. It's never uploaded anywhere. The one network request this tool makes is downloading the model files (~17MB total) once, which are cached by your browser afterward — every photo you process after that runs completely offline.
Frequently asked questions
What's the difference between this and the Object Detector?+
This tool answers "what is this whole photo of" with one ranked list of labels. The Object Detector instead finds and draws a box around each individual object in a photo — useful when a photo contains several different things you need located separately, not just labeled as a whole.
What can it recognize?+
1,000 everyday categories from the standard ImageNet dataset — animals (often down to specific breeds/species), household objects, food, vehicles, instruments, clothing, and more. It won't recognize people by name, brand logos, or anything outside its training categories.
Why does it show 5 results instead of just the top one?+
Many real photos are genuinely ambiguous between similar categories (e.g. two dog breeds, or a bowl versus a cup) — showing the top 5 with their actual confidence gives a truer picture than forcing a single, possibly wrong, best guess.
Does it work well on photos with multiple different objects in them?+
Not really — this model produces one classification for the whole image, so a photo containing several unrelated objects will get a less confident, sometimes misleading result. It works best on a photo that's clearly "of" one main subject.
Is my photo uploaded anywhere?+
No — it's read directly in your browser and classified on your device using TensorFlow.js. It never leaves your device, and nothing is retained once you leave the page. The only network request is a one-time model download (~17MB), cached by your browser afterward.