Custom Vision Prediction Client
Custom Vision Prediction Client
Custom Vision Prediction API runs inference using your trained Custom Vision model — upload an image, get classification labels and confidence scores.
Pairs with Custom Vision Training API — seamless training-to-inference
Only invokes your own trained models — not a general-purpose vision API
Train and publish a model in the Custom Vision portal to get endpoint + prediction key. POST /customvision/v3.0/Prediction/{projectId}/classify/iterations/{iterationName}/image with the image upload.
Uptime · 30-day window
About this API
Custom Vision is Azure's "no-ML-expertise-needed" tool for training custom image classification/object detection models: upload labeled images, click train in the UI, get a model in minutes — zero code. Prediction API is the inference endpoint — call the published endpoint with the prediction key, get confidence scores per class. Typical uses: QA inspection (train "pass/fail" classifier), UGC auto-tagging (train business-specific tags), branded-SKU detection. Complements the general-purpose Computer Vision API (out-of-the-box, no training).
What you can build
- 1Production inference on a trained image-classification model
- 2Object detection (e.g. defect detection on a production line)
- 3Auto-tag product images
Strengths & limitations
Strengths
- Pairs with Custom Vision Training API — seamless training-to-inference
- Low-latency single-image inference (~500ms typical)
- Supports model export to ONNX/TensorFlow Lite for offline inference
Limitations
- Only invokes your own trained models — not a general-purpose vision API
- Accuracy depends entirely on training-data quality
Getting started
Train and publish a model in the Custom Vision portal to get endpoint + prediction key. POST /customvision/v3.0/Prediction/{projectId}/classify/iterations/{iterationName}/image with the image upload.
FAQ
Are Prediction and Training the same API?+
No. Training runs on *.cognitiveservices.azure.com; Prediction has a different endpoint (per-project prediction key).
Can models be exported for offline use?+
Yes — the Custom Vision portal supports export to ONNX / TensorFlow Lite / CoreML for edge inference.
Technical details
- Auth type
- unknown
- Pricing
- unknown
- Protocols
- REST
- SDKs
- python, javascript, csharp, java, go
- Response time
- 63 ms
- Last health check
- 6/26/2026, 6:23:46 AM
More from Microsoft
Bing Autosuggest API returns search suggestions from a partial query — for search-box type-ahead completion.
Bing Computer Vision API analyzes images — OCR, object detection, face detection, scene captioning, content moderation.
Bing Custom Image Search restricts image search to a custom set of domains — the image counterpart of Bing Custom Search.
Bing Custom Search lets you define a custom domain subset and run Bing search against it — for vertical or enterprise search.
Custom Vision Training API programmatically trains image-classification / object-detection models — create projects, upload labeled images, trigger training, publish iterations.
Bing Entity Search recognizes queries as named entities (people, places, organizations) and returns structured card info (description, image, related links).
Bing Image Search runs image search over Bing's web-wide image index — by keyword, similar image, aspect ratio, or size filter.
Bing Local Business Search finds local businesses (restaurants, hotels, hospitals) by geographic query — returns business info, address, ratings.