Sensity is a customizable suite of tools for deepfake detection, offering cloud-based or on-premise analysis of videos, images, and audio through both UI and API access. Users can analyze files or URLs for multilayer assessment within seconds.

The steps in this guide briefly describes how to integrate Sensity API with your app and backend. To create verification tasks, receive notifications and retrieve the results of the analysis follow the next steps:

1. Create a developer account

To get started with Sensity, please send your request for creating a developer account via the contact form on www.sensity.ai.

Once you have a Sensity developer account, request an API token to the /auth/tokens API endpoint with your credentials using HTTP BasicAuth.

You will receive a JSON object containing an authorization Bearer token that you can use in the SDKs to make subsequent calls to the API.

2. (Optional) Setup the Webhooks

We use webhooks to notify your backend when an event happens. Webhooks are useful to trigger actions for asynchronous events like when a verification task completes successfully or not. This is best described in the Webhooks section.

3. Retrieve data from the API

Use the Sensity API via our SDKs to retrieve specific data from the already completed verification tasks.

You can find more details on the endpoints and the response models in the API section of the documentation.

Sensity's API provides the following solutions that describe standard use cases. Each solution is composed of different services, each with its endpoint. API's users can serve customized use cases by grouping multiple services in their user flow.

Performs an analysis designed to recognize the latest AI-based image, video and audio manipulation and generation.

This solution is composed by:

• Face manipulation
• AI Generated Image detection
• Forensic analysis
• Voice analysis

The above services accept media files, such as images, videos, and audio, as well as URLs as input.

Supported Input Formats

Input File Limitations

Input URL Limitations

The above services also accept URLs pointing to supported media files. For example, https://www.example.com/image.jpg, https://www.example.com/video.mp4 or https://www.example.com/audio.wav.

This section shows how to use the Sensity API endpoints to create and retrieve the results of a verification task using the provided endpoints. The guide assumes you have followed the steps in Initial setup and already have a developer account with an API token ready to use.

To create a specific analysis task you can send a POST request with /tasks/{task_name} and the task parameters.

You will receive a JSON response with the following fields:

{
  "report_id": "123456789",
  "success": true
}

• report_id represents the requested analysis
• success whether the task is successfully created or not

There are two ways to retrieve the results from the requested analysis:

1. Polling
2. Webhooks

Due to the asynchronous nature of the API, we strongly recommend using Webhooks.

Polling

To get the result of a task, make a GET request to /tasks/{task_name}/{report_id}.

You will receive a JSON object with a status field indicating the status of the task.

{
  "status": "message"
  // ... other task related fields
}

The possible message value for status is one of:

• completed: the other fields in the response will be populated with the analysis result.

• in_progress: the analysis is still running. Try to call the endpoint later to get the result.

• failed: something went wrong. Try to re-create the task and ask support if there are subsequent issues.

Webhooks

In this approach, the user associates a custom user-provided endpoint to an API analysis event. Once the event occurs, the API takes care of sending the result of the event to the endpoint.

How to set up a webhook server is discussed in the Webhooks server side SDK section.

Once a verification task is created by a user, Sensity starts processing the data asynchronously. As the processing happens at a later time and not in the response to your code execution. We need a mechanism to notify your system about status changes of the verification task.

This is where webhook comes in place, as it can notify your system once a verification task is processed.

1. Webhook configuration

A Webhook is nothing else than a simple user-provided POST endpoint of a server. The server should support:

• POST requests in application/json type.

• If the server runs behind a firewall you should whitelist our IP address to allow incoming traffic from our servers. Get in contact with Sensity support for more information.

You can use a service like webhook.site to test receiving webhook events. Or a tool such as ngrok to forward the requests to your local development server.

2. Event types

Sensity webbook service supports the following events:

• face_manipulation_analysis_complete
• ai_generated_image_detection_analysis_complete
• voice_analysis_complete
• forensic_analysis_complete

When a verification task is done your server will receive an event with the resulting data of the task. An explanation of each event response is described in the GET endpoints response of the API reference and Models reference.

3. Register a webhook URL

To register a webhook, make a POST request to /webhooks/{event_name} using the name of the event and the URL you want to register.

Once the analysis has finished, a POST request will be sent by the API to the registered URL with the analysis result.

Note: if no webhook URL is registered before the task analysis is requested, Sensity API will retry to send the result message for a total of 10 retries. After the 10th consecutive failure, the result can only be retrieved through polling.

This section shows you how to create a express nodejs server and listen to incoming webhook events. This assumes you have already Register a webhook in the previous section.

1. Install the required dependencies

npm install express cors body-parser

2. Create the server and the webhook endpoint

Add the following code to index.js. This creates a server listening to incoming requests at http://localhost:3030/webhook.

const express = require("express");
const cors = require("cors");
const bodyParser = require("body-parser");

const app = express();
const port = 3000;

app.use(bodyParser.json());
app.use(cors());

app.post("/webhook", (req, res) => {
  // do logic here
  console.log(req.body);

  res.status(200).send("Success");
});

app.listen(port, () => {
  console.log(`Listening at http://localhost:${port}`);
});

3. Run the server & create a task

Run the express server with:

node index.js

And then create a task. You can do this with CURL as its explained in the API Reference examples or using one of the Client-side SDKs.

As soon as the verification task is completed you should see the results in the terminal output.

Listening at http://localhost:3000
{
  id: '808264ae-bd79-457e-89cc-b643e4f55517',
  status: 'completed',
  result: { live: 'Live' }
}

Management for the user's API tokens

This analysis detects potential AI manipulation of faces present in images and videos, as in the case of face swaps, face reenactment and lipsync.

Supported input files: image / video

This analysis detects AI-generated photo-realistic images and video. Images may be created for example by Generative Adversarial Networks (GAN) and Diffusion Models (DM) like Stable Diffusion, MidJourney, Dalle-2 and others. Videos may be made with tools such Runway, Sora, Luma, Pika, Kling and others. An optional explanation of prediction is provided for some classes of model generators and if input is predicted as fake.

Supported input files: image / video

This analysis detects AI-generated voices and voice cloning in media. It also provides transcription and translation to English.

Supported input files: audio / video

This service uses forensic analysis techniques to determine whether the file was digitally created or manipulated. In particular, the test will find traces of suspect software editing, use of screenshot software, file dates mismatch, multiple versions of the files, and more.

Supported input files: image / video / pdf / audio

Webhook management