Is This Image AI Generated? A 12-Point Checklist and a Free Check
You have an image open and you want a yes or no. Two routes: upload it to a detector and read the evidence, or run the 12 checks below and score the flags. Both take under two minutes.
To find out if an image is AI generated, upload the original file to an AI image detector for a verdict, a 0 to 100 score and any generator name found in its metadata. If you cannot upload it, check hands, text, lighting, reflections and textures, then reverse-search the image and inspect the poster's account. Score the flags.
- How do you check if an image is AI generated?
- What is the 12-point checklist?
- How do you score the checklist?
- What does a detector score mean?
- FAQ
- Conclusion
How do you check if an image is AI generated?
A purpose-built detector is faster and more reliable than your eyes. Here is the whole process.
Upload the original file
Open the AI image detector. JPG, PNG, WEBP, BMP, TIFF, HEIC and HEIF up to 12 MB. Use the original rather than a screenshot where you can, because a screenshot deletes the metadata that could name the generator.
Let the two layers run
A pixel model scores the statistical fingerprints that Midjourney, DALL-E, Stable Diffusion and Flux leave behind. A metadata layer, the source check, reads EXIF, XMP, PNG text and C2PA records and matches them against 29 known generators.
Read the result
You get a whole-image verdict, a 0 to 100 score, and a source check result showing the generator's name with an evidence level of strong, moderate, light or none.
Free accounts include 50 detections per month, no card required.
What is the 12-point checklist?
Cannot upload the image, or want to confirm the detector's call? Work through these 12 checks. Score one flag for every check the image fails.
- Warped hands or fingers, rings sinking into skin
- Mismatched earrings, glasses frames or pupils
- Waxy, poreless skin at 200% zoom
- Teeth that merge or eyes with different catchlights
- Garbled text, logos or signage anywhere in the frame
- Shadows that point in two directions or float free of their objects
- Reflections that show a different scene
- Backgrounds that melt or repeat the same element
- No EXIF, no Content Credentials, no camera make or model
- Reverse image search finds no earlier copy
- The poster's account is new, thin or unverifiable
- The image exists on only one site, with no higher-resolution original
Face and body checks (1 to 4)
1. Hands and fingers. Count the fingers and follow each one to a knuckle. Current generators draw simple poses well. They still fail on interlaced fingers and hands gripping objects.
2. Paired features. Compare left and right. Mismatched earrings, two different frame shapes and pupils of different sizes are classic GAN-era tells and still common in fake profile pictures.
3. Skin texture. Zoom in on cheeks and foreheads. AI skin tends toward an airbrushed, waxy sheen. Beauty filters cause the same look on real photos, so weigh this with the other checks.
4. Teeth and eyes. Look for merged teeth and for catchlights that differ between the two eyes. Real photos reflect one light source in both eyes.
Scene checks (5 to 8)
5. Text and logos. Read every sign, label and shirt. Generators paint the shape of text rather than words, so lettering goes alien once it gets small.
6. Lighting and shadows. Find the brightest light and trace every shadow away from it. Two sun directions means the scene was assembled, not photographed.
7. Reflections. Mirrors, windows, sunglasses and water should show the same scene. A reflection of a different room is decisive.
8. Background melt and repetition. Fences, railings and windows that warp, or the same face, leaf or window appearing twice.
Source checks (9 to 12)
9. Metadata and Content Credentials. Check the file at the Content Credentials verify tool. DALL-E 3 and Adobe Firefly attach signed C2PA records. Stable Diffusion tools write the prompt into PNG text. A blank file proves nothing, because every social upload strips this data, but a present record is strong evidence.
10. Reverse image search. Run it through Google Lens or TinEye. An image that only exists on one account posted yesterday deserves suspicion.
11. The poster's account. New account, stolen bio, a feed of suspiciously perfect photos. Context flags matter as much as pixels.
12. Does an original exist? Real photos have a source: a wire service, a photographer, an event. AI images appear from nowhere.
How do you score the checklist?
Add up your flags and use this rubric. The detector is the tiebreaker.
| Flags | Read | Next step |
|---|---|---|
| 0 to 1 | Likely real | Still verify the source if money, reputation or safety is on the line |
| 2 to 3 | Suspicious | Run it through the AI image detector and read the source check result |
| 4 or more | Probably AI | Treat as synthetic until an original surfaces |
Human eyes are bad at this. In a controlled study, participants identified AI-synthesised faces with 48.2% accuracy, below chance, and rated the fakes as more trustworthy than real faces (Nightingale and Farid, PNAS, 2022). The checklist narrows the field. The detector decides.
What does a detector score mean?
A detector score is a probability, not a certainty. Ours is the model's estimate that the image was generated or AI-edited. We publish no single accuracy percentage, because an accuracy number without its test set is marketing.
Read the result in three parts. A high score either way is strong evidence, especially when it agrees with your checklist. A mid-range score means look closer: compression, cropping and filters weaken the pixel signal. A source check result is a separate line of evidence. A named generator at strong evidence settles the question. A result of none means the metadata was stripped, which is normal for anything saved from a feed and is not evidence the image is real.
If the detector and the checklist disagree, weight the source checks. Provenance beats pixels. For the mechanics, read how AI image detectors work.
FAQ
How can I check if an image is AI generated for free?
Upload it to the AI image detector. Free accounts include 50 detections per month and return a verdict, a 0 to 100 score and a source check result naming the generator where the file's metadata allows. Pair that with the 12-point manual checklist above for a confident answer.
Is there an app to tell if an image is AI?
The detector runs in any browser, on a phone or a desktop, with nothing to install. Save or share the image to your device, open the detector page and upload it. Upload the original where possible, since a screenshot strips the metadata a detector could use to name the generator.
Can ChatGPT tell if an image is AI generated?
Not reliably. A vision chatbot describes what an image looks like and guesses. A forensic detector measures statistical artifacts the generation process leaves in the pixels and reads the C2PA and EXIF record in the file. Use a chatbot to describe an image and a detector to test it.
What if the detector score is around 50?
Treat it as unresolved. Find a higher-resolution original, run the source checks from the checklist, and look at the source check result. Heavy compression and small files are the most common causes of a mid-range score, because they smear the pixel signal the model depends on.
Does a missing watermark mean the image is real?
No. Most AI images carry no visible watermark, and invisible watermarks like Google's SynthID only mark images from the tools that apply them. Cropping and re-saving also remove metadata. Absence of a watermark or of Content Credentials tells you nothing either way. Presence is strong evidence.
Conclusion: get your answer, then trust it
So, is this image AI generated? Upload it and read both layers of evidence, or run the 12 checks, score the flags and let the detector break the tie. For the full visual training see how to tell if an image is AI generated. If you are vetting a profile photo, start with AI generated profile pictures.
Check the image now. Free accounts include 50 detections per month.