Is This Image AI Generated? How to Check (Any Vendor)

Finding out whether an image was generated by AI isn't always obvious to the naked eye. But every major AI image generator leaves behind identifiable fingerprints — metadata signatures, pixel-level artifacts, and embedded watermarks that reveal their origin.

This guide walks through how to detect AI images from every major vendor.

How to Inspect Any Image for AI Fingerprints

Before diving into vendor-specific details, start with the AI Image Inspector. It scans any image for C2PA manifests, SynthID watermarks, Stability AI metadata, EXIF data, and generation parameters — all in your browser with no upload.

ChatGPT / DALL-E 3 (OpenAI)

What it embeds: C2PA Content Credentials with the full DALL-E 3 generation manifest. Includes the prompt, model version, generation timestamp, and a cryptographic signature verifying OpenAI as the origin.

How to detect: Open the image with the AI Image Inspector and check for a C2PA manifest. If present, it will display the full provenance chain including the generating prompt. Even without C2PA support, DALL-E images often contain an "OpenAI" signature in their PNG metadata chunks.

Watermark: SynthID — Google's imperceptible watermark embedded at the pixel level. Survives screenshots and compression.

Midjourney

What it embeds: PNG text chunks (tEXt/iTXt) containing the full generation parameters — prompt, aspect ratio, model version, chaos, stylize, seed, and the Midjourney job ID. No C2PA or SynthID as of mid-2026.

How to detect: Open the image metadata. Midjourney images are the easiest to identify because they store rich, human-readable text metadata directly in the file. Look for the "Description" field or the "Prompt" tag in the image properties. The presence of parameters like "--ar 16:9 --v 6.0" is a dead giveaway.

Watermark: None. Midjourney relies solely on PNG metadata chunks.

Stable Diffusion (AUTOMATIC1111 / ComfyUI)

What it embeds: PNG tEXt chunks with the full generation recipe — positive prompt, negative prompt, seed, sampler, steps, CFG scale, model hash, and dimensions. Images from ComfyUI embed the entire workflow JSON, not just the prompt.

How to detect: Open with any metadata viewer. If the image contains a "parameters" text block starting with the prompt and ending with "Steps:", it's almost certainly from Stable Diffusion. The model hash (a 10-character hex string) identifies the exact checkpoint used.

Invisible watermark: The invisible-watermark library (dwtDct) embeds a frequency-domain pattern that encodes a 32-bit identifier. This survives light editing but not aggressive resizing.

Google Imagen

What it embeds: SynthID watermark at the pixel level. Google was the first to deploy SynthID at scale, and all Imagen outputs contain it.

How to detect: The AI Image Inspector can detect SynthID signals. There's typically no text metadata beyond basic image properties from Imagen.

Watermark: SynthID — the most robust watermark in production, designed to survive screenshots, resizing, and compression.

Adobe Firefly

What it embeds: C2PA Content Credentials with Adobe's provenance chain. Includes the generating prompt, editing history, and cryptographic signature chain. Adobe is the strongest C2PA adopter — Firefly images have the most detailed provenance manifests.

How to detect: Look for C2PA metadata. Adobe's manifest includes an "issued by" field linking back to Adobe's certificate authority. Photoshop's "Content Credentials" panel can also verify these.

Watermark: C2PA only — no pixel-level watermark as of mid-2026.

Microsoft Designer (DALL-E powered)

What it embeds: C2PA Content Credentials via Microsoft's Azure signing infrastructure. Similar to DALL-E 3 but with Microsoft's certificate chain. May also include SynthID depending on the backend pipeline.

How to detect: C2PA manifest inspection. Designer images exported from Copilot or the Designer web app consistently include C2PA metadata.

Flux (Black Forest Labs)

What it embeds: PNG text metadata with prompt and generation parameters. Similar to Stable Diffusion's metadata pattern but with Flux-specific fields like guidance scale. No C2PA support as of mid-2026.

How to detect: Check PNG metadata chunks. If you see fields like "flux" in the software tag or model identifier, it's a Flux image.

Ideogram

What it embeds: PNG text metadata with prompt and generation parameters. Ideogram 2.0 includes a "generated_by" tag that identifies the platform.

How to detect: The "by ideogram" tag in metadata is the clearest signal. Prompt text is typically included in standard PNG comment chunks.

Quick Reference

Vendor	Metadata Type	Pixel Watermark
ChatGPT/DALL-E 3	C2PA + prompt	SynthID
Midjourney	PNG text chunks	None
Stable Diffusion	PNG text + JSON	dwtDct
Google Imagen	Minimal	SynthID
Adobe Firefly	C2PA (detailed)	None
Microsoft Designer	C2PA	SynthID (varies)
Flux	PNG text	None
Ideogram	PNG text	None

To quickly check any image, use the AI Image Inspector. It analyzes metadata, C2PA manifests, SynthID signals, and generation parameters in one pass — all in your browser.