AI Image Detector: Was This Image Made by AI?
Upload a picture, and Convertavo's AI image detector estimates whether it was generated. It's an estimate, not proof. You get a verdict, how sure the model is, and the evidence behind it. When the honest answer is "can't tell", the report says so.
What the detector checks
An honest estimate, not proof. We tested the model on a labelled set of images, and it got about three in four right. So you'll always see how sure it is and what it found, and when it can't tell, it says so.
Your files stay private
Uploads are deleted from our servers one hour after processing. Nothing is read or reused, and nothing is kept unless you save it to your account.
No account required
20 free credits a day, with no email address and no watermarks on the output.
Runs on any device
Works in any modern browser on Windows, macOS, Linux, Android and iOS.
How to check if an image is AI-generated
Three steps, nothing to install, no account needed.
- STEP 01
Add the image
Drop in a JPG, PNG or WebP up to 8 MB. It can be a photo, a screenshot or a piece of artwork. You can also paste an image link with Paste a URL.
- STEP 02
Click Check image
Two checks run at once: one reads the pixels, the other reads the file's metadata. It usually takes 2 to 4 seconds.
- STEP 03
Read the report
You see a verdict, how confident it is, and every piece of evidence behind it.
Got an iPhone photo in HEIC format? Turn it into a JPG first with our HEIC to JPG converter.
What the AI Image Detector does
Pixels and metadata
A trained model reads noise, edges and texture, while a second check reads what the file says about where it came from.
An honest confidence
You see how sure the model is. A low-confidence result is labelled as one, not dressed up as an answer.
Private by design
Runs on our own server. Your image is never sent to an AI company, and it is deleted an hour after the check.
How this AI image detector works
Most AI image checkers give you one percentage and nothing else. This one runs two separate checks and shows you both. It's much easier to trust a verdict when you can see the reasons.
Check 1: the pixels
An open-source AI model looks at the pixels themselves. Cameras and AI generators leave different fingerprints in an image:
- the grain of sensor noise
- the way edges and fine texture are drawn
- the patterns of colour across the frame
The model learned from thousands of real photos, human-made art and AI images. It guesses which group your picture belongs to.
Because it reads the pixels, it still works on a screenshot. It also works on a re-saved copy with its metadata stripped away.
Check 2: the metadata
Metadata is the hidden information a file carries about itself. A photo straight from a phone usually records the camera model, the exposure settings, and sometimes a GPS location. AI image generators never write those.
Many generators label their images instead. Some write their name into the file. Stable Diffusion tools often save the full prompt and settings. A growing number add Content Credentials (C2PA), a record that says the image was made with AI.
When any of these are there, the report tells you. A label inside the file is harder evidence than any model's opinion.
Can it detect Midjourney, ChatGPT and Gemini images?
Yes. The pixel model answers one question: does this image look generated or real? It does that for images from:
- Midjourney
- DALL·E and ChatGPT's image generator
- Stable Diffusion and Flux
- Adobe Firefly
- Google's Gemini and Imagen
- and other AI image generators
It can't reliably say which one made an image. A tool that claims to name the generator from the pixels alone is promising more than is possible.
The metadata check is different. If a file still carries a generator's label, a Stable Diffusion prompt, or Content Credentials, the report names the source. That's because the file itself is saying so.
How accurate is this AI image detector?
Not accurate enough to accuse anyone. On our own labelled test set, the model was right about three times in four.
It did better on human-made artwork and ordinary photos. It was weaker on the newest generators. That's normal for any AI image detector. It's why this tool shows you the evidence, not just a verdict.
How to read the confidence score
The report shows how the model split its vote between "AI-generated" and "real". It also shows how sure it is.
| Confidence | What it means |
|---|---|
| Above about 90% | The model is usually right |
| 65% to 90% | It leans one way, but it's wrong often enough that you should weigh the other evidence |
| Below 65% | Close to a coin toss. The report says "low confidence" and lets the metadata speak |
How to tell if an image is AI-generated by looking at it
Run the detector first. Then look at the image yourself with these signs in mind. None of them is proof on its own, but several together are a strong hint.
- Hands, teeth and ears. Extra or fused fingers, too many teeth, earrings that don't match.
- Text and signs. Letters that almost form words, logos that melt into shapes.
- Background logic. Railings that go nowhere, windows that don't line up, repeated faces in a crowd.
- Skin and surfaces. An even, poreless smoothness, or a plastic shine on hair and fabric.
- Lighting. Shadows falling in different directions, or reflections that don't match the scene.
- Missing history. No earlier copy anywhere online, and no camera information in the file.
Checking AI art vs photographs
Spotting AI art is harder than spotting AI photos. Digital paintings, 3D renders and heavily styled drawings aren't photos. So they already look "unusual" to a model trained partly on camera images.
We tested it against another popular detector, which called almost every human-made illustration AI. This one got 7 of 8 right. Still, for art, always read the confidence.
If it really matters, ask the artist for sketches, layers or a time-lapse of their work. You can also try our AI Art Detector. It runs the same two checks, with a guide written for drawings, paintings and illustrations, and for artists who need to show their work is their own.
Why does it say my real photo is AI-generated?
Some normal photo edits make a real picture look fake. And some AI images slip past. Here's why:
- Heavy editing. Retouching, beauty filters, upscaling and noise reduction smooth an image the way generators do.
- Phone camera modes. "Night mode" and portrait blur are already computer-made, so they can nudge the score.
- Small or squeezed files. Very small or heavily compressed images leave the model little detail to work with.
- Brand-new generators. They're newer than what any detector learned from, so their images can slip past.
- Edits on top of AI. Editing an AI image in Photoshop, or photographing a screen, blurs the evidence in both checks.
Got a strange result? Check the confidence first. Then look at the metadata in the report.
Content Credentials, watermarks and metadata
Content Credentials follow the C2PA standard. They're a signed record attached to a file that says how it was made and edited.
Adobe, OpenAI, Microsoft and others add them to AI-generated images. Some cameras add them to real photos too.
They're strong evidence when they're there. But they're easily lost, since most social networks remove them on upload.
Invisible watermarks, like Google's SynthID, are a separate system. Only the maker's own checker can read them.
Who uses an AI image checker?
Journalists and fact-checkers
Checking a viral photo before they share it.
Teachers
Looking at artwork students hand in.
Shoppers
Checking if product photos or a rental listing are real.
Dating apps and marketplaces
Checking a profile picture before trusting the person behind it.
Artists and competition organisers
Checking entries.
In every case, the detector is the first step, not the last. A strong result tells you where to look. The source and history of the image finish the job.
Is Convertavo's AI image detector free and private?
Is it free?
Yes. You get 20 free credits a day, and each image check uses 10. That's two free checks a day, with no email address and no watermarks. A free account raises that to 50 credits a day (five checks), and paid plans add monthly credits.
Free checks use the normal queue. Priority processing comes with Standard and above.
What happens to my image?
The check runs on our own server. Your image is never sent to an AI company.
Uploads are deleted one hour after processing. Nothing is stored, read or reused. The privacy policy has the details.
Does it work on my phone?
Yes. It works in any modern browser on Windows, macOS, Linux, Android and iOS. There is nothing to install.
Not sure if a photo is real? Check it now, read the evidence, and decide for yourself.
Check an imageAI Image Detector — frequently asked questions
Short, straight answers to what people ask most.
How accurate is this AI image detector?
Not accurate enough to accuse anyone. On our own labelled test set, it was right about three times in four. It does best on photos and human-made art, and worst on the newest generators. Treat a result as a strong hint, never as proof.
Is the AI image detector free?
Yes. You get 20 free credits a day, and each check uses 10, so that's two free checks a day. You don't need an email address, and there are no watermarks.
Can it tell if an image was made with Midjourney?
It can flag a Midjourney image as likely AI-generated. It can't reliably name Midjourney from the pixels alone. If the file still carries Midjourney's label, the report names it.
Can it detect images made with ChatGPT or DALL·E?
Yes, images from DALL·E and ChatGPT's image generator are among those the pixel model checks. OpenAI adds Content Credentials to its images, so if the file still has them, the report shows that too.
Why does it say my real photo is AI-generated?
Heavy retouching, beauty filters, upscaling and noise reduction can make a real photo look generated. Night mode, portrait blur and heavy compression can nudge the score too. Check the confidence and the metadata evidence in the report.
Does it work on screenshots and images from social media?
Yes. The pixel check works on screenshots and re-saved copies. Social media usually removes metadata labels, so the pixel check does most of the work there. Very small or heavily compressed images are harder to judge.
Can it detect deepfakes or face swaps?
It's built to spot images that were generated from scratch. A real photo with one face swapped in may not be caught, because most of its pixels still come from a real camera. Dedicated deepfake tools look at faces specifically.
What happens to the images I upload?
The check runs on our own server, and your image is never sent to an AI company. It's deleted one hour after processing. Nothing is stored, read or reused.
More free image tools
Not sure if a photo is real? Check it free on Convertavo, read the evidence, and then decide for yourself.