ai-image-detector

v1.0.1

Published

a month ago

Detect and verify AI-generated images — check C2PA, SynthID, OpenAI DALL-E, Adobe Firefly, Google Gemini, and Stability AI provenance through a single unified API

ai-image-detector

Detect & Verify AI-Generated Images — C2PA, SynthID, DALL-E, Firefly, Gemini, Stable Diffusion

A single npm package that detects and verifies AI-generated images across all major AI image generators — C2PA, SynthID, OpenAI DALL-E, Adobe Firefly, Google Gemini/Imagen, and Stable Diffusion — through one unified API.

⚠️ ai-image-detector does NOT attempt to visually detect AI-generated images.
It only verifies official provenance evidence — metadata, Content Credentials, watermarks, and cryptographic signatures embedded by AI providers themselves.

Why ai-image-detector?

Before this package, you needed multiple provider-specific integrations:

// ❌ Before
import providerA from "some-c2pa-library";
import providerB from "some-openai-metadata-tool";
import providerC from "custom-stable-diffusion-parser";

With ai-image-detector:

// ✅ After
import { verifyImage, verifyC2PA, verifyOpenAI } from "ai-image-detector";

Installation

npm install ai-image-detector

yarn add ai-image-detector

pnpm add ai-image-detector

Requirements: Node.js ≥ 20, ESM or CJS project.

Quick Start

import { verifyImage } from "ai-image-detector";

const result = await verifyImage("./photo.png");

console.log(result.verified);    // true / false
console.log(result.confidence);  // 0 – 1
console.log(result.providers);   // ["Stability AI", "Metadata"]

Example output

{
  "verified": true,
  "confidence": 0.9625,
  "providers": ["Stability AI", "Metadata"],
  "checks": {
    "metadata": { "passed": true, "message": "Found AI-related PNG metadata chunks: parameters", "confidence": 0.85 },
    "stabilityAI": { "passed": true, "message": "Stable Diffusion PNG parameter chunk found: key=\"parameters\"", "confidence": 0.97 }
  },
  "providerResults": [ ... ],
  "verifiedAt": "2025-01-15T10:30:00.000Z"
}

CLI

npx ai-image-detector photo.png
npx ai-image-detector photo.png --verbose
npx ai-image-detector photo.png --json
npx ai-image-detector https://example.com/image.jpg

CLI output

  ╔══════════════════════════════════════╗
  ║   ai-image-detector  •  v1.0.1       ║
  ╚══════════════════════════════════════╝

  ✓ PROVENANCE VERIFIED  via Stability AI, Metadata

  Confidence   ████████████████░░░░ 94%
  Providers    Stability AI · Metadata
  Verified At  1/15/2025, 10:30:00 AM

  Checks
  ────────────────────────────────────────────
  ✓ metadata (85%)
     Found AI-related PNG metadata chunks: parameters
  ✓ stabilityAI (97%)
     Stable Diffusion PNG parameter chunk found

CLI options

| Option | Description | |--------|-------------| | --json | Machine-readable JSON output | | --verbose | Full per-provider check breakdown | | --providers <names> | Comma-separated provider list | | --timeout <ms> | Provider timeout (default: 10 000) | | --metadata | Run only the Metadata provider | | --c2pa | Run only the C2PA provider | | --synthid | Run only the SynthID provider | | --openai | Run only the OpenAI provider | | --adobe | Run only the Adobe provider | | --gemini | Run only the Gemini provider | | --stability | Run only the Stability AI provider | | --openai-key <key> | OpenAI API key | | --adobe-client <id> | Adobe I/O client ID | | --adobe-token <token> | Adobe I/O access token | | --gemini-key <key> | Google Cloud API key | | --synthid-key <key> | SynthID API key | | --help | Show usage | | --version | Show version |

API Reference

`verifyImage(image, options?)`

Runs all registered providers and returns an aggregated result.

import { verifyImage } from "ai-image-detector";

const result = await verifyImage("./image.png", {
  providers: ["OpenAI", "Adobe"],  // optional: run subset
  timeout: 15_000,                  // optional: ms per provider
  parallel: true,                   // optional: default true
  includeMetadata: true,            // optional: default true
});

Input: string (path or URL), Buffer, Uint8Array, or URL

Individual Verification Methods

Run a single provider and get its result directly:

import {
  verifyMetadata,
  verifyC2PA,
  verifySynthID,
  verifyOpenAI,
  verifyAdobe,
  verifyGemini,
  verifyStabilityAI,
} from "ai-image-detector";

// Metadata (EXIF / XMP / PNG chunks)
const metadata = await verifyMetadata("./image.png");

// C2PA Content Credentials
const c2pa = await verifyC2PA("./image.jpg");

// Google SynthID
const synthid = await verifySynthID("./image.jpg", {
  apiKey: "your-google-cloud-api-key",
  projectId: "your-project-id",
});

// OpenAI DALL-E
const openai = await verifyOpenAI("./image.png", {
  apiKey: "sk-...",
});

// Adobe Firefly
const adobe = await verifyAdobe("./image.jpg", {
  clientId: "your-adobe-client-id",
  accessToken: "your-adobe-token",
});

// Google Gemini / Imagen
const gemini = await verifyGemini("./image.png", {
  apiKey: "your-google-api-key",
  projectId: "your-project-id",
});

// Stable Diffusion / Stability AI
const stability = await verifyStabilityAI("./image.png");

Provider Coverage

| Provider | Metadata | C2PA | Watermark | API Verification | |----------|----------|------|-----------|-----------------| | Metadata | ✅ EXIF/XMP/PNG/IPTC | — | — | — | | C2PA | — | ✅ JUMBF + Claims | — | ✅ optional c2pa-node | | SynthID | ✅ XMP/EXIF | — | 🔜 when API released | 🔜 Google API | | OpenAI | ✅ EXIF/XMP/PNG | ✅ DALL-E 3 C2PA | — | 🔜 when released | | Adobe | ✅ EXIF/XMP | ✅ Firefly C2PA | — | ✅ Adobe I/O API | | Gemini | ✅ EXIF/XMP/IPTC | ✅ C2PA actions | 🔜 SynthID co-proof | 🔜 Vertex AI API | | Stability AI | ✅ EXIF | — | — | ✅ PNG param chunks |

TypeScript Interfaces

interface VerificationResult {
  verified: boolean;
  confidence: number;          // 0 – 1
  providers: string[];

  checks: {
    metadata?: CheckResult;
    c2pa?: CheckResult;
    synthid?: CheckResult;
    openai?: CheckResult;
    adobe?: CheckResult;
    gemini?: CheckResult;
    stabilityAI?: CheckResult;
    [key: string]: CheckResult | undefined;
  };

  providerResults: ProviderResult[];
  metadata?: Record<string, unknown>;
  verifiedAt: string;          // ISO-8601
}

interface ProviderResult {
  provider: string;
  verified: boolean;
  confidence: number;
  checks: Record<string, CheckResult>;
  metadata?: Record<string, unknown>;
  verifiedAt: string;
  warnings?: string[];
  error?: string;
}

interface CheckResult {
  passed: boolean;
  message: string;
  confidence?: number;
  severity?: "info" | "warning" | "error";
  details?: Record<string, unknown>;
}

Plugin System

Extend ai-image-detector with your own provider — no core modification needed.

import { registerProvider, verifyImage } from "ai-image-detector";
import type { ProvenanceProvider, ProviderResult, ImageInput } from "ai-image-detector";

class MidjourneyProvider implements ProvenanceProvider {
  readonly name = "Midjourney";

  async verify(image: ImageInput): Promise<ProviderResult> {
    // Your verification logic here
    const buffer = /* load image */;
    const hasMjMeta = buffer.includes(Buffer.from("Midjourney"));

    return {
      provider: this.name,
      verified: hasMjMeta,
      confidence: hasMjMeta ? 0.9 : 0,
      checks: {
        midjourney_meta: {
          passed: hasMjMeta,
          message: hasMjMeta ? "Midjourney metadata found" : "No Midjourney metadata",
        },
      },
      verifiedAt: new Date().toISOString(),
    };
  }
}

// Register once at startup
registerProvider(new MidjourneyProvider());

// Now included in verifyImage automatically
const result = await verifyImage("./image.png");

What Each Provider Checks

Metadata Provider

Scans EXIF, XMP, IPTC, and PNG tEXt/iTXt chunks for AI generation evidence:

EXIF Software field (e.g. "Stable Diffusion", "DALL-E", "Adobe Firefly")
XMP xmp:CreatorTool namespace
EXIF UserComment / ImageDescription with prompt-like content
PNG parameters / workflow / prompt text chunks

C2PA Provider

Verifies C2PA Content Credentials (the Coalition for Content Provenance and Authenticity standard):

JUMBF box detection in JPEG APP11 / PNG caBX / WebP
C2PA label validation ("c2pa" inside JUMBF superbox)
Claim signature block presence (ES256, ES384, etc.)
AI generation action assertions (c2pa.ai.generatedContent, com.adobe.firefly.generated)
XMP-based C2PA provenance reference
Optional c2pa-node native full validation (install separately)

SynthID Provider

Checks for Google SynthID evidence:

SynthID XMP field detection
Google AI EXIF software hints
IPTC source field scanning
Official Google Cloud API (when credentials provided)

OpenAI Provider

Checks for DALL-E provenance:

EXIF Software = "DALL-E" / "OpenAI DALL-E 3"
OpenAI XMP namespace fields
C2PA action com.openai.generated (DALL-E 3 always includes C2PA)
PNG metadata chunks from the OpenAI API response
Optional OpenAI provenance API

Adobe Provider

Checks for Adobe Firefly and Photoshop Generative Fill provenance:

EXIF Software = "Adobe Firefly"
Adobe XMP namespaces (photoshop:, xmpMM:, Iptc4xmpExt:)
C2PA action com.adobe.firefly.generated
contentcredentials.org origin marker
Optional Adobe I/O Content Authenticity API

Gemini Provider

Checks for Google Gemini / Imagen provenance:

EXIF / binary Google AI software hints
Google XMP namespace fields
C2PA Google action assertions
IPTC source field
SynthID co-presence (expected for Gemini images)
Optional Vertex AI API

Stability AI Provider

Checks for Stable Diffusion provenance (the richest metadata of any AI generator):

PNG parameters tEXt chunk (AUTOMATIC1111/InvokeAI/Fooocus format)
PNG workflow tEXt chunk (ComfyUI JSON format)
EXIF Software field (many SD frontends write this)
Parsed AUTOMATIC1111 parameter extraction (prompt, seed, model, steps, etc.)
Stability AI XMP fields

Advanced Usage

Parallel vs Sequential

// Parallel (default) — fastest
const result = await verifyImage(image, { parallel: true });

// Sequential — useful for debugging
const result = await verifyImage(image, { parallel: false });

Stream / Buffer

import { readFile } from "fs/promises";

const buffer = await readFile("./image.png");
const result = await verifyImage(buffer);

From URL

const result = await verifyImage("https://example.com/photo.jpg");

Custom Timeout

const result = await verifyImage(image, { timeout: 5_000 });

Per-provider Options

const result = await verifyImage(image, {
  providerOptions: {
    SynthID: { apiKey: "gcloud-key", projectId: "my-project" },
    Adobe:   { clientId: "adobe-id", accessToken: "adobe-token" },
    OpenAI:  { apiKey: "sk-..." },
  },
});

Exit Code Integration (CI)

npx ai-image-detector image.png --providers openai,adobe || echo "No provenance found"
# Exit code 0 = verified, 1 = not verified / error

Confidence Scoring

| Score | Meaning | |-------|---------| | 0.90 – 1.00 | Strong evidence — cryptographic signature or authoritative API confirmation | | 0.75 – 0.89 | High confidence — multiple independent signals agree | | 0.50 – 0.74 | Moderate — one clear signal but no cryptographic proof | | 0.00 – 0.49 | Low — weak hints only or conflicting signals | | 0.00 | No provenance evidence found |

The final confidence in VerificationResult is the weighted mean of all successful providers' confidence scores.

Published

Vulnerabilities

Links

Maintainers

Keywords

Readme

ai-image-detector

Why ai-image-detector?

Installation

Quick Start

Example output

CLI

CLI output

CLI options

API Reference

verifyImage(image, options?)

Individual Verification Methods

Provider Coverage

TypeScript Interfaces

Plugin System

What Each Provider Checks

Metadata Provider

C2PA Provider

SynthID Provider

OpenAI Provider

Adobe Provider

Gemini Provider

Stability AI Provider

Advanced Usage

Parallel vs Sequential

Stream / Buffer

From URL

Custom Timeout

Per-provider Options

Exit Code Integration (CI)

Confidence Scoring

License

`verifyImage(image, options?)`