VISORA REDACTION DOCS

Integrate image redaction into your app

Redaction projects process images and return a new redacted image. Use this for face blur, OCR text blur, custom word redaction, ID-document sensitive fields, and license plates.

Current API base URL: https://6p0ws7vu2f.execute-api.us-east-1.amazonaws.com/dev

Redaction API keys must be used from server code only. Redaction is available on paid plans and requires an API key attached to a redaction project.

Faces

Detect and redact faces in user photos, profiles, screenshots, and media uploads.

Text

Redact selected text categories, custom words, and sensitive ID-document values.

License plates

Detect plate-like OCR values and redact them before image delivery.

SDK usage

Install @visoracloud/client and call redactImage() when your backend receives an upload. The SDK sends multipart data to POST /redact.

Install

npm install @visoracloud/client

Upload and redact

import { readFile } from "node:fs/promises";
import { Visora } from "@visoracloud/client";
 
const visora = new Visora({
  apiKey: process.env.VISORA_REDACTION_API_KEY!,
});
 
const image = await readFile("./profile.jpg");
 
const result = await visora.redactImage({
  file: image,
  filename: "profile.jpg",
  contentType: "image/jpeg",
});
 
console.log(result.redactionId);
console.log(result.redactedImageUrl);

If you already uploaded an object through Visora scoped upload paths, call redactImageKey(). The key must belong to the same project as the API key.

Redact an existing image key

const result = await visora.redactImageKey({
  imageKey: "accounts/acc_123/projects/proj_123/uploads/image.jpg",
});

Next.js server route

import { Visora } from "@visoracloud/client";
 
const visora = new Visora({ apiKey: process.env.VISORA_REDACTION_API_KEY! });
 
export async function POST(request: Request) {
  const form = await request.formData();
  const image = form.get("image") as File | null;
 
  if (!image) {
    return Response.json({ error: "Missing image" }, { status: 400 });
  }
 
  const result = await visora.redactImage({
    file: Buffer.from(await image.arrayBuffer()),
    filename: image.name,
    contentType: image.type || "image/jpeg",
  });
 
  return Response.json({
    redactionId: result.redactionId,
    redactedImageUrl: result.redactedImageUrl,
    counts: {
      faces: result.facesBlurred,
      text: result.textBlurred,
      licensePlates: result.licensePlatesBlurred,
    },
  });
}

REST API

POST /redact accepts either multipart file upload or JSON with an existing scoped image key.

Multipart

200 OK

curl -X POST "$VISORA_API_URL/redact" \
  -H "x-api-key: $VISORA_REDACTION_API_KEY" \
  -F "image=@./profile.jpg;type=image/jpeg"

Existing imageKey

200 OK

curl -X POST "$VISORA_API_URL/redact" \
  -H "content-type: application/json" \
  -H "x-api-key: $VISORA_REDACTION_API_KEY" \
  -d '{"imageKey":"accounts/acc_123/projects/proj_123/uploads/image.jpg"}'

Project settings

Settings are configured per redaction project in the dashboard. They are not sent on every API request, which keeps production requests simple and consistent.

Setting	Type	Behavior
faceBlur	boolean	Blur or black-box detected faces.
textBlur	boolean	Master text redaction switch. With no categories it blurs all detected text; with categories it redacts only those data types.
licensePlateBlur	boolean	Redact license plate-like text detected in the image.
redactionStyle	blur \| black_box	Choose visual treatment for redacted regions.
textCategories	id_document \| pii \| dates \| financial \| credentials \| medical \| sexual \| profanity	Selective detection categories (require textBlur). When set, only these data types are redacted instead of all text. id_document and medical blur sensitive values near their labels; pii/dates/financial/credentials match value patterns (emails, phones, dates in any format, cards, IBAN/CLABE, API keys, JWTs).
customWords	string[]	Exact words or phrases your project wants to redact.
ignoredWords	string[]	Words that should not be redacted, useful for ID labels like name or address.
minConfidence	number	Minimum detection confidence from 0 to 100. Default is 80.

Exported SDK types

type RedactionStyle = "blur" | "black_box";
type RedactionTextCategory =
  | "id_document"   // passport, license, national ID — sensitive fields only
  | "pii"           // emails, phones, addresses, IDs, tax numbers
  | "dates"         // dates in any format — DOB, issue, expiry
  | "financial"     // card numbers, bank accounts, IBAN, CLABE
  | "credentials"   // passwords, API keys, tokens, secrets
  | "medical"       // patient IDs, record numbers, case numbers
  | "sexual"
  | "profanity";
 
interface RedactionSettings {
  faceBlur: boolean;
  textBlur: boolean;
  licensePlateBlur: boolean;
  redactionStyle: RedactionStyle;
  textCategories: RedactionTextCategory[];
  customWords: string[];
  ignoredWords: string[];
  minConfidence: number;
}

Detection categories

textCategories requires textBlur. With text blur on and no categories, all detected text is blurred. Add categories to redact only specific data types instead. id_document and medical blur the sensitive value next to a label; the others match value patterns directly.

Category	Detects
id_document	Identity-document fields (passport, license, national ID) — blurs sensitive values near their labels, not the whole document.
pii	Emails, phone numbers, addresses, IDs, and tax numbers (RFC, CURP, SSN).
dates	Dates in any format — DOB, issue, and expiry (01/02/2024, 2024-01-02, Jan 5 2024).
financial	Card numbers (Luhn-checked), bank account numbers, IBAN, and CLABE.
credentials	Passwords, API keys (sk_, ghp_, AWS), tokens, JWTs, and secrets.
medical	Patient IDs, record numbers, and case numbers near their labels.
sexual / profanity	Explicit or offensive wording.

Pattern-based categories are deterministic but not perfect. Validate on real images for your use case — addresses and generic IDs are the hardest to detect.

Response model

The response includes the original object key, the redacted object key, a temporary 5 minute redactedImageUrl, aggregate counts, and normalized bounding boxes for the redacted regions.

200 OK

application/json

{
  "redactionId": "red_01KVMZ6K4RBZ1PAMWJH7EK4XQW",
  "imageKey": "accounts/acc_123/projects/proj_123/uploads/profile.jpg",
  "redactedImageKey": "accounts/acc_123/projects/proj_123/redacted/01KVMZ6K.jpg",
  "redactedImageUrl": "https://...",
  "facesBlurred": 1,
  "textBlurred": 2,
  "licensePlatesBlurred": 1,
  "faces": [
    {
      "confidence": 99.7,
      "boundingBox": { "left": 0.34, "top": 0.18, "width": 0.16, "height": 0.22 }
    }
  ],
  "regions": [
    {
      "type": "license_plate",
      "text": "GR-3004D",
      "confidence": 93.1,
      "boundingBox": { "left": 0.41, "top": 0.78, "width": 0.14, "height": 0.05 }
    }
  ]
}

Webhooks

Subscribe a redaction project endpoint to the redaction.completed event in the dashboard. Every successful POST /redact emits one. Deliveries are asynchronous, HMAC-signed, retried, and moved to a dead-letter queue after repeated failures, so they never block or slow the redaction response.

redaction.completed event

POST to your endpoint

{
  "id": "evt_01KVMZ7P2YQ9XJ8R5C3F0N2T6B",
  "type": "redaction.completed",
  "createdAt": "2026-06-21T18:24:11.482Z",
  "accountId": "acc_123",
  "projectId": "proj_123",
  "data": {
    "redactionId": "red_01KVMZ6K4RBZ1PAMWJH7EK4XQW",
    "imageKey": "accounts/acc_123/projects/proj_123/uploads/profile.jpg",
    "redactedImageKey": "accounts/acc_123/projects/proj_123/redacted/01KVMZ6K.jpg",
    "style": "blur",
    "facesBlurred": 1,
    "textBlurred": 2,
    "licensePlatesBlurred": 1,
    "regions": [
      {
        "type": "license_plate",
        "text": "GR-3004D",
        "confidence": 93.1,
        "boundingBox": { "left": 0.41, "top": 0.78, "width": 0.14, "height": 0.05 }
      }
    ],
    "createdAt": "2026-06-21T18:24:11.300Z"
  }
}

Each delivery carries visora-timestamp and visora-signature headers. The SDK verifies the signature and narrows event.data by event type, so no manual casting is required.

Next.js webhook route

import { createNextWebhookHandler } from "@visoracloud/client";
 
// Visora signs every delivery; the handler verifies the signature for you.
export const POST = createNextWebhookHandler({
  secret: process.env.VISORA_WEBHOOK_SECRET!,
  onEvent: async (event) => {
    if (event.type === "redaction.completed") {
      // event.data is narrowed to VisoraRedactionCompletedData
      const { redactionId, redactedImageKey, regions } = event.data;
      console.log("Redacted", redactionId, "->", redactedImageKey);
      console.log(regions.length, "regions blurred");
    }
  },
});

Redaction endpoints only receive redaction.completed. Moderation and review events are scoped to moderation projects.

Errors and constraints

Status	Meaning
400	Missing image, invalid JSON, unsupported image type, empty file, or file larger than 8 MB.
401	Missing or invalid x-api-key.
403	Free plan, non-redaction project API key, revoked key, or imageKey outside the project scope.
429	Monthly usage limit exceeded.
500	Unexpected image download, image processing, storage, or detection failure.

Supported upload types are JPEG and PNG. Multipart images must be 8 MB or smaller. Usage is counted against the same account-level monthly usage system as moderation.

← PreviousModeration docs Next →Dashboard