AI Image Detector

Free
No upload
No watermark

Toolvy / AI Image DetectorRuns on your device · nothing uploaded

Drop a file to checkor click to browse — checked on your device, never uploaded

Quick answer

An AI image detector estimates whether a photo or illustration came from a generator such as Midjourney, GPT Image, FLUX or Gemini. Toolvy first reads what the image file itself declares — Content Credentials, AI labels, prompts, camera data — then runs a visual classifier over five regions of the picture, on your own device. You get a verdict, an AI likelihood and the evidence behind both, and the image is never uploaded.

Step by step

How to ai image detector

01
Drop the image
JPEG, PNG, WebP, GIF, AVIF or HEIC. Use the largest, least-processed copy you can find — an original beats a screenshot.
02
Let the classifier load
The first check downloads the visual model, about 44 MB, and caches it. It then scores the center and four corners of the image.
03
Read evidence before the score
A signer, an AI label or a prompt in File evidence outweighs any score. Then read the AI likelihood and the verdict.
04
Keep a record
Copy report puts the findings on your clipboard for a fact-check log or moderation note.

In depth

Can it spot Midjourney, GPT Image, FLUX and Gemini images?

The two checks work differently. File evidence catches whatever a tool writes: OpenAI’s image models, Adobe Firefly, Microsoft and Google attach Content Credentials to many outputs, and Stable Diffusion tools such as ComfyUI and InvokeAI save the prompt in PNG text. The visual check needs none of that — it looks for traces generated images share, so it can flag a Midjourney or FLUX picture with no data left in it, though not every one. Gemini images may also carry SynthID, an invisible watermark Toolvy can’t read; Google’s SynthID Detector checks for it.

What does each result mean?

Common results on images, and a sensible next step for each.
What you see	What it means	Next step
"Made with AI — confirmed"	The generator, such as OpenAI or Adobe, signed a statement saying so	Treat it as AI-made; note the signer
"Declared as AI-made" with a prompt	The file still holds the text it was generated from	Read the prompt to see what was asked for
"Edited with AI" on a camera photo	A real photo with AI changes, such as a generative fill	Find the unedited original to judge how much changed
"Likely AI-generated", no file evidence	No data survived, and every region looked generated	Find the earliest copy and who posted it
"Possibly AI-generated" or "Inconclusive"	A weak lean toward AI, or none	Don’t share it as real or as fake yet
"Captured by a camera — signed"	A camera such as a Leica, Nikon or Pixel signed the photo at capture	Strong evidence of a real photo; ask what changed since

How do I read the AI likelihood?

It is the score of the least AI-looking of the five regions checked. That is deliberate: one odd patch of sky won’t get a real photo called AI, because the other regions pull the number down — but a genuine photo with an inpainted face can score low too. Read it as signal strength, not exact odds, and "No signs of AI found" as the classifier finding nothing, not proof the photo is real. For scale, its developer reports that on development data it flagged about 3–5% of real photos and caught roughly 66–93% of web-compressed images from recent generators; those are the developer’s figures, not an independent test.

Which images are hardest to judge?

Small images under roughly 300 pixels on a side, which hold too little detail.
Heavy compression, which wears away the fine traces the classifier relies on — common in pictures forwarded through chat apps.
Screenshots and photos of screens, which add scaling, moiré and glare and leave the metadata behind.
Heavily edited real photos: strong beauty filters or stylized grading can make them look generated.
Small AI edits, such as a removed object, where most of the picture is genuine.

What about drawings and digital art?

File evidence works the same for an illustration as for a photo. Be more cautious with the visual check: clean digital art made by hand can share the polished look of generated images. For drawings and paintings, lean on the artist’s process — sketches, layered files, time-lapses — and never accuse an artist of using AI on the strength of a score.

Does my photo leave my device?

No. The metadata and the pixels are both read inside your browser tab, and the Network tab in DevTools confirms the photo is never sent. The model is fetched once, on first use, and then cached.

For clips, use the AI video detector; to check whether an essay, PDF or Word file was written by AI, the AI text detector; and for any of these in one place, the all-format AI detector. Our guide on how to tell if an image is AI-generated covers the manual checks that belong alongside any detector.

Your files never leave your device

Processed locally in your browser — nothing is uploaded. You can verify it yourself: open your browser’s developer tools, watch the Network tab, and use the tool — no upload request is ever made.

No file is ever uploaded to a server
Works offline after the first visit
No account, no watermark, no limits

A little more clarity

Frequently asked questions

Can it detect Midjourney images?

It can, but not every time. If the file still carries an AI label or prompt, File evidence shows it. Once that data has been stripped, as often happens on social media, the verdict rests on the visual check — which, by its developer’s own measure, misses some recent generators’ images.

Can it tell if a picture was made with ChatGPT?

If the image still carries the Content Credentials OpenAI attaches to many of its outputs, yes: File evidence names the signer and the verdict reads "Made with AI — confirmed". If it was screenshotted or re-saved by an app that dropped them, only the visual check remains, and that is an estimate.

Why was my real photo flagged as AI?

False alarms happen: on its developer’s development data, about 3–5% of real photos were flagged. They are likelier with strong filters, small or re-compressed copies, and photos of screens. If the photo is yours, the original from your camera or phone, with its camera data intact, is your best evidence.

What does "Edited with AI" mean on a photo?

The file’s data records that an AI tool changed an existing image — the IPTC source type compositeWithTrainedAlgorithmicMedia, which some editors write when features such as generative fill are used. The photo may be mostly real: the label says AI was involved, not how much.

Does it work on screenshots?

It runs, but treat the answer with caution. A screenshot keeps none of the original file’s metadata, so File evidence is usually empty, and re-rendering on screen adds scaling and color shifts that can confuse the visual check. Find the original file and check that instead.

Can it show which part of the image is fake?

No. The verdict and AI likelihood describe the picture as a whole, with no map of suspicious areas. Because an image only counts as AI when all five regions look generated, one where AI changed a small area, such as a swapped face, can come back clean.

Guides

More tools, same promise.

123 tools for images, video, audio and subtitles — every one of them runs on your device, and a result can go straight into the next.

Browse all tools

Free forever
No sign-up
Nothing uploaded