Car Recognition AI
Upload any car photo and get its brand, exact model, color, body type and camera angle back as structured JSON — from the same API that reads license plates. Trained on millions of real-world photos from 20+ car marketplaces across North America, Europe and Asia.
181
car brands
2,003
car models
8
body types
4
view angles
Each model handles one step. Together they turn a raw photo into clean, structured data.
Finds every car in the frame and crops it out, so busy backgrounds, showrooms and street scenes never confuse the classifiers.
Classifies the camera angle — front, back, side or three-quarter — with 93.9% accuracy. Perfect for auto-sorting listing galleries.
Predicts the manufacturer out of 181 brands with 92.8% accuracy — from Toyota and BMW to the newest Chinese EV makers.
Goes one level deeper: the exact model out of 2,003 classes — Land Cruiser 300, X5, Sportage — with 85.2% accuracy.
brand_uncertain
Two independent networks vote on the final brand and model verdict. When they disagree and confidence is low, the API deliberately withholds the answer instead of guessing — so you can trust the results you do get.
POST an image to /v1/car and get the full vehicle profile back in a single JSON payload.
Sedan, SUV, coupe, hatchback, wagon, convertible, pickup or minivan — with full probability distribution.
Named color plus exact hex and RGB values of the dominant color cluster.
Front, back, side or semi-side — pick the best frame automatically.
Manufacturer prediction with confidence scores across the top 5 candidates.
Exact model prediction as a brand + model pair, with the top 5 candidates.
The recommended field to build on: both networks fused with calibrated confidence gates.
POST /v1/car
{
"success": true,
"body": {
"label": "coupe",
"confidence": 0.75,
"car_bbox": { "x1": 234, "y1": 616, "x2": 1166, "y2": 1071 },
"color": { "name": "silver", "hex": "#b9c2cb", "rgb": [185, 194, 203] },
"position": { "label": "semi_side", "confidence": 0.99 },
"brand": { "label": "Porsche", "confidence": 0.77 },
"model": { "label": "Porsche|911", "confidence": 0.68 },
"make_model": {
"brand": "Porsche",
"model": "911",
"decision": "agreement"
}
},
"cost": 0.02
}Every component can be switched off with a query parameter to skip its model and speed up the response. Pay only for what you run.
| Query parameter | What it controls |
|---|---|
body_type | Body-type classification (8 classes) |
color | Dominant color analysis |
position | View-angle classification |
brand | Brand classification (181 brands) |
car_model | Model classification (2,003 classes) |
Example request
curl -X POST "https://api.example.com/v1/car" \ -H "Authorization: Bearer YOUR_API_KEY" \ -F "[email protected]"
Calling /v1/detect? The full car profile is already included in the response next to the plate data. Need plates only? Pass plate_only=true and skip every car model for a roughly 2.5x faster response.
From parking barriers to marketplace listings — the same four models power very different products.
Cross-check the plate against the car profile to catch cloned plates: if the plate says Camry but the camera sees a pickup, flag it. Log color and body type with every entry and exit.
Auto-tag every listing photo with make, model and color. Validate what sellers typed against what the photos actually show, and order galleries by view angle.
Verify that the vehicle in claim photos matches the insured make and model before a human ever looks at the file.
Audit yard and gate cameras automatically: confirm the right vehicle type arrived, and build a searchable visual log of every movement.
Catalogue trade-ins and auction lots from photos in seconds instead of typing specs by hand.
Measure the vehicle mix on any road — brands, body types, colors — without storing a single plate number.
Search every brand and model the API recognizes before you write a line of code.
One call to /v1/car returns the car’s body type (8 classes), dominant color with hex value, view angle (front, back, side, semi-side), brand (181 classes) and exact model (2,003 classes), plus the bounding box of the detected car. If you call /v1/detect instead, you get all of that alongside full license plate reading.
Measured on large sets of real-world photos the models never saw during training: 92.8% brand accuracy, 85.2% model accuracy across more than 2,000 possible classes, and 93.9% view-angle accuracy. Every response includes confidence scores so you can set your own thresholds.
The fused make_model verdict uses calibrated confidence gates over two independent networks. On an unresolvable disagreement the API returns null with brand_uncertain set to true instead of guessing, and a decision field tells you exactly which fusion rule produced the verdict.
Yes. Each component — body type, color, view angle, brand, model — has its own query toggle on /v1/car. Disabling components skips their models entirely and returns faster.
Yes. The API accepts ordinary JPEG, PNG or WebP images from any camera or phone. No special hardware, firmware or SDK required — if you can capture a photo, you can analyze it.
Car recognition uses the same per-request pricing as plate recognition: from $0.007 per request on volume tiers, with a free tier of 150 requests per month to build and test against.
Free tier included — 150 requests per month, no credit card required. Your first car profile is one POST request away.