Page Comparison

Introduction

To expose AI services to our system, we need an API to be exposed for the MediaHaven installation. The contract of this API is as follows (anything can be added before the path):When creating new Module Plugins having of the following https://mediahaven.atlassian.net/wiki/spaces/CS/pages/edit-v2/4438884395#Categories

• Embedding

• GenerativeMetadata

then MediaHaven will connect to this plugin using REST API calls, which must implement the following methods depending on the category.

Methods

GET /config-rules

Status
colour Blue title ALL CATEGORIES

It gives the possible configuration parameters for this connector

...

{
  "Rules": [
    {
      "Name": "TextModel",
      "Type": "String|Integer|Boolean",
      "Required": false,
      "Default" : "ada2"
    }
  ]
}

POST /text-embedding

Status
colour Blue title EMBEDDING

Expected parameters

...

Text as JSON body in Textproperty and Config based on config rules

...

Code Block
language json
{ "Text": "Given me pictures of lamas", "Dimensions": 1536, "Config": { "TextModel" : "ada2" } }

Expected headers

...

Code Block
x-api-key: <optional api key>

Output

Array of float (vector) with size equal to dimensions. The vector is normalized

POST /embedding

...

MUST BE normalized. When the provided dimensions are 0, the embedding should be in the natural dimensions used by the embedding. If the embeddings cannot be provided in the requested dimensions, then a 400 HTTP error code should be returned.

Ex.

{
  "Embedding": [ 
    -0.006929283495992422,
    -0.005336422007530928,
    ... (omitted for spacing)
    -4.547132266452536e-05,
    -0.024047505110502243]
}

POST /embedding

Status
colour Blue title EMBEDDING

Expected parameters

Record as JSON body in Metadata property and Config based on config rules

...

{
  "Config": {
    "TextModel" : "ada2"
  },
  "Metadata": {
    "RightsManagement": {
        "Permissions": {...},
        "RecordPhase": "Concept",
        "ExpiryDate": null,
        "ExpiryStatus": null,
        "Zone": {}
    },
    "Dimensions": 1536
  }
}

Expected headers

...

Code Block
x-api-key: <optional api key>

Output

Array of float (vector) with size equal to dimensions. The vector is normalizedMUST BE normalized. When the provided dimensions are 0, the embedding should be in the natural dimensions used by the embedding. If the embeddings cannot be provided in the requested dimensions, then a 400 HTTP error code should be returned.

Ex.

{
  "Embedding": [ 
    -0.006929283495992422,
    -0.005336422007530928,
    ... (omitted for spacing)
    -4.547132266452536e-05,
    -0.024047505110502243]
}

POST /describe
POST /generative-metadata/description

Status
colour Blue title GENERATIVE

Expected parameters

...

Record as JSON body in Metadata property and Config based on config rules

...

Code Block
language json
{ "Config": { "TextModel" : "ada2" }, "Metadata": { "RightsManagement": { "Permissions": {...}, "RecordPhase": "Concept", "ExpiryDate": null, "ExpiryStatus": null, "Zone": {} } } }

Expected headers

...

Code Block
x-api-key: <optional api key>

Output

Textual string

Ex.

{
  "Description": "This is an image of a cat"
}

POST /generative-metadata

Status
colour Blue title GENERATIVE

Record as JSON body in Metadata property, Fields with as items dotted keys and AI hint as value, Locale as the requested target locale for the output and finally Config based on config rules

...

{
  "Fields": { 
    "Descriptive.Description": {
      "Hint": "Give a description for the image"
     },
     "Descriptive.Keywords.Keyword": {
        "Hint": "Give a list of keywords. Generate around 10.",
        "Type": "array"
     }
  },
  "Locale": "en_US",
  "Config": {
    "TextModel" : "ada2"
  },
  "Metadata": {
    "RightsManagement": {
        "Permissions": {...},
        "RecordPhase": "Concept",
        "ExpiryDate": null,
        "ExpiryStatus": null,
        "Zone": {}
    }
  }
}

Expected headers

...

Code Block
x-api-key: <optional api key>

Output

Follow the input fields, the same structure should be returned A JSON object of generated fields with as keys dotted keys and as value the AI's generated value.

Ex:

{ 
  "MetadataGeneratedFields": {
    "Descriptive.Description": "This is an image of a cat",
    "Descriptive.Keywords.Keyword": ["Animal", "Cat", ...]  
  }
}