Einstein Vision
Einstein Vision is a Salesforce computer vision API that is no longer sold and no longer answers for most orgs.
Definition
Einstein Vision is a Salesforce computer vision API that is no longer sold and no longer answers for most orgs. You posted an image to a REST endpoint and got predictions back as JSON. It shipped inside Einstein Platform Services next to Einstein Language, covering image classification, object detection, and optical character recognition.
The classification and detection endpoints stopped being reachable on 31 July 2023, and Einstein Vision for Planogram Compliance went dark the same day. Subscriptions stopped renewing after 1 May 2024. The pair still sits on Salesforce's active retirements list at order end date, so the real cutoff was whatever your contract said.
Custom vision work now runs through AI Models, the console previously called Einstein Studio, inside Data 360.
In plain English
“Salesforce used to rent out a service that could look at a photo and say what was in it. You fed it labeled examples and it learned to sort new pictures. Salesforce switched it off, so you now train the model elsewhere and plug it in.”
Worked example
A regional auto insurer built claims triage on Einstein Vision in 2021. An adjuster uploaded an accident photo to a Claim__c record, an Apex callout posted it to the classification endpoint with the modelId of their trained damage model, and the response wrote Rear_Bumper_Damage into Damage_Type__c and a confidence of 0.87 into Damage_Confidence__c. On 31 July 2023 the endpoint stopped answering and every callout failed. The rebuild kept both fields and the record-triggered Flow untouched, swapping only the callout target for a vision model the team trained and hosted themselves.
How Einstein Vision worked and why it went dark
Three jobs wearing one badge
The name covered three jobs. Image classification returned a single label from a class set you defined, such as good panel or dented panel. Object detection returned bounding boxes, labels, and confidence scores for several items in one photo. OCR pulled typed and handwritten text out of images and PDFs. The visual product search demos people remember were classification underneath, one class per product, not a separate endpoint.
Training a model, then calling it
You uploaded a labeled dataset as a zip of folders, started training, then polled until the job returned a model ID. Prediction requests carried that ID plus an image URL or base64 bytes. Auth used the OAuth JWT bearer flow: sign an assertion with the einstein_platform.pem key, POST it to api.einstein.ai/v2/oauth2/token, then call api.einstein.ai/v2/vision/ocr with the token. No admin ever configured any of it by clicking.
Where the work lives now
Nothing replaced it with the same shape, which is what catches teams out. You own the training and the hosting now, and the callout points at whatever you stood up. AI Models is where in-platform model work happens, though the documented path there runs to predictive models rather than image ones. Extraction that used OCR usually gets rebuilt against a multimodal model called through a prompt template.
Trust & references
Cross-checked against the following references.
- Einstein Object Detection & Image Classification Retirement (opens in new tab)Salesforce
- Salesforce Active Product & Feature Retirements (opens in new tab)Salesforce
- Salesforce Past Product & Feature Retirements (opens in new tab)Salesforce
- Einstein Vision for Planogram Compliance Retirement (opens in new tab)Salesforce
Straight from the source - Salesforce's reference material on Einstein Vision.
Hands-on resources to go deeper on Einstein Vision.
About the Author
Dipojjal Chakrabarti is a B2C Solution Architect with 29 Salesforce certifications and over 13 years in the Salesforce ecosystem. He writes and edits salesforcedictionary.com, published by KineticBit Inc., to help admins, developers, architects, and cert/interview candidates sharpen their fundamentals. More about Dipojjal.
Test your knowledge
+5 pts / dayQ1. When did the Einstein Vision image classification and object detection endpoints stop being accessible?
Q2. A legacy app still posts images to the Einstein Vision classification endpoint from Apex and every call fails. What is the supported fix?
Q3. Which statement about the current status of Einstein Vision is accurate?
Discussion
Loading discussion…