A practical guide

How to tell whether an image is AI-generated

Text detection asks whether a person wrote a sentence. Image detection asks something harder: whether a camera was ever involved at all. This guide covers what can and cannot be established about an image, how to read a confidence score without over-reading it, and how to run a verification you could defend.

Midjourney · DALL·E · Stable Diffusion · Flux · Imagen · Firefly · Sora · Updated September 2026

Check an image free Read the guide
3 Outcomes distinguished
Heatmap Region-level evidence
10MB JPG, PNG and WebP
Report Downloadable record
Free No account required
The three outcomes

Real, generated, or quietly altered

Most people think of image detection as a yes-or-no question. It is not. There are three meaningfully different answers, and the one in the middle is the one that matters most in practice.

01

A real photograph

A camera captured a scene that physically existed. This is the baseline case, and it is also where false positives hurt most — wrongly flagging a genuine photograph can cost someone a grade, a job, a byline or a reputation. A detector's value is judged as much by how rarely it misfires here as by how often it catches a fake.

Baseline case
02

Fully AI-generated

No camera was ever involved. The image was synthesised from a text prompt by a generative model. This is the most tractable case, because the entire frame comes from the same non-photographic process. It is also the case most people mean when they ask whether an image is "fake" — and increasingly the least interesting one, because outright synthetic images are often obvious from context alone.

Most tractable
03

A real photo with an AI-altered region

The hard case, and the consequential one. A genuine photograph has had something added, removed or replaced — a person edited out, a background swapped, a face changed, an object inserted. Most of the frame is authentic, so any single overall verdict tends to understate the problem. This is precisely why region-level analysis exists: the useful finding is not "this image scores 64%" but "this specific area does not behave like the rest of the frame."

Hardest and most consequential

The distinction matters because the three cases carry different stakes. A fully generated image is usually a question of disclosure. An altered photograph is usually a question of intent — someone had a real image and chose to change what it showed. In journalism, insurance, HR investigations and academic integrity work, the second is almost always the more serious finding.

Generator families

Not all AI images are equally detectable

"AI-generated" covers several distinct technologies with very different characteristics. Knowing which one you are likely dealing with tells you how much weight to put on any result.

DIF
Midjourney · DALL·E · Stable Diffusion · Flux · Imagen · Firefly

Diffusion models: the dominant case

The technology behind essentially all current image generation. Output is now routinely photorealistic, and the gap between generations is measured in months rather than years. This is the family behind the overwhelming majority of AI images in circulation today, across every domain from stock photography to disinformation.

Photorealistic at high resolution, including faces and hands
Billions of images generated and shared to date
Release cadence measured in months
Freely available through consumer apps and open weights
Detection outlook: reliable on unaltered originals
EDT
Generative fill · Object removal · Face swap · Background replace

AI-edited photographs: the hardest case

A real photograph with an AI-altered region. Now built directly into mainstream photo editors and phone galleries, which means it is no longer a specialist capability — removing a person from a photo takes one tap. Because most of the frame is authentic, a single overall verdict is the wrong tool here.

Built into consumer phones and mainstream editors
The majority of the frame is genuine camera capture
Overall scores are diluted by the authentic majority
Region heatmaps matter more than the headline number
Detection outlook: the active frontier for every vendor
GAN
StyleGAN · ThisPersonDoesNotExist · Older face generators

GAN-generated faces: the legacy case

The older technology behind the synthetic profile pictures that filled social platforms and fake-account networks. Largely superseded for creative work, but still heavily used in fraud, catfishing and bot networks because it is free, instant and endlessly repeatable.

Faces dominate the frame with a characteristic crop
Backgrounds often dissolve into incoherent texture
Accessories such as glasses and earrings render asymmetrically
Framing is highly consistent across a generated set
Detection outlook: the most tractable family
VID
Sora · Veo · Runway · Kling · Extracted frames

AI video frames: the emerging case

A still pulled from an AI-generated video is a different problem from a still generated directly. Video is compressed aggressively before anyone ever sees it, so an extracted frame arrives already degraded. As generated video becomes ordinary, this category is growing fast, and it is currently the weakest ground for every image detector on the market.

Heavy video compression applied before extraction
Frame grabs discard the original delivery file entirely
Dimensions and aspect ratios typical of video, not cameras
Motion blur can resemble photographic characteristics
Detection outlook: treat conclusions with most caution
Interpreting a result

What the score does and does not tell you

A number on its own is close to useless, and is the single most misread part of any detection tool. GPTOne returns a confidence score together with the evidence behind it, because the evidence is what you actually act on.

🔢

The score is a confidence estimate

It expresses how strongly the available evidence points one way. It is not a probability that a court would accept, and it is not a percentage of the image that is fake. Two images with identical scores can warrant completely different responses depending on what produced them.

🗺

The heatmap localises the finding

A region-level view shows which areas of the frame drove the result. This is the difference between "something about this image is unusual" and "this specific area does not match the rest of the photograph" — the second is actionable, the first is not.

📋

The breakdown shows what drove it

Each result comes with the supporting evidence laid out, so you can see whether a verdict rests on one strong indicator or several weak ones. A high score built from a single signal deserves more scepticism than a moderate score built from agreement across many.

📄

The report gives you a record

A downloadable report captures the result, the evidence and the file it applied to. For editorial, academic and HR workflows this matters: a decision that gets challenged later needs a record of what was assessed and when.

Honest limitations

What no image detector can reliably do

These constraints apply to every image detector available today, GPTOne included. A tool that does not tell you about them is not being straight with you.

📷

Screenshots are a poor substitute

A screenshot is a new, lower-quality copy of an image rather than the image itself. Detail is lost the moment it is taken, and any assessment made from it is correspondingly weaker. Wherever you can, work from the file as it was originally saved or received.

📤

Forwarded images have been altered

Messaging apps and social platforms routinely re-compress and resize what you upload. An image that has passed through several hands is not the image that was created, regardless of its origin. Read results on a forwarded copy as materially less certain.

🎟

Heavy processing can mimic generation

Aggressive noise reduction, beauty filters, modern computational photography and very clean studio lighting all smooth away the irregularities of an ordinary photograph. These are the most common sources of false positives, and every one of them is normal photographic practice rather than deception.

🔄

Generators improve continuously

Each generation of image models closes gaps the previous one left open. Detection is a moving target with no finish line, which is why any accuracy figure is a snapshot against the models that existed when it was measured rather than a permanent property of a tool.

👤

Small and low-quality images say less

Thumbnails, avatars and heavily compressed images simply carry less to assess. Results on them are weaker across every tool on the market, and a confident-looking verdict on a tiny image should be treated with suspicion.

A score is evidence, not proof

No image detector is an appropriate standalone basis for disciplinary action, an employment decision, a publication retraction or a legal claim. A confidence score should open a verification process, not conclude one. Where the outcome carries legal or financial weight, consult a qualified forensic examiner.

Best practice

How to verify an image responsibly

The workflow below is what separates a defensible verification from an accusation built on a number someone did not understand.

01

Ask for the original

Request the file as it was originally saved, not a screenshot or a download from a social post. It is the step most often skipped and the one everything else depends on.

02

Read the evidence, not the number

Open the breakdown and the heatmap. A result driven by one localised region means something very different from one spread evenly across the frame.

03

Corroborate outside the tool

Check the source, the capture context and the chain of custody. Reverse image search. Ask the person who supplied it. Technical evidence is one input among several.

04

Escalate proportionately

Match your response to the stakes. For anything with legal, academic or financial consequences, a qualified forensic examiner is the appropriate authority, not a web tool.

Common questions

Questions about AI image detection

No. A detector produces a confidence estimate, not proof. GPTOne reports results as evidence and confidence signals that help you make an informed decision. In academic, hiring, legal, journalistic or other high-stakes contexts, an image score should open a verification process, not close one. Corroborate with the original file, the source, the photographer, and where the outcome carries legal or financial weight, a qualified forensic examiner.
An AI-generated image was created entirely by a generative model from a prompt, with no camera involved. An AI-edited image began as a real photograph that was then altered using AI tools such as generative fill, object removal, background replacement or face swapping. The second case is harder to assess because most of the frame is genuine, which is why a region heatmap matters more there than the overall score. It is also usually the more serious finding, because someone had a real image and chose to change what it showed.
As a strength-of-evidence estimate, not a probability of guilt and not the percentage of the image that is fake. Always read it alongside the evidence breakdown and the region heatmap. A high score resting on a single indicator warrants more scepticism than a moderate score where several independent indicators agree. Two images with the same number can call for completely different responses.
Because they are not the original image. Screenshotting creates a new, lower-quality copy, and messaging apps and social platforms re-compress and resize whatever passes through them. Detail is lost at every hop, and less detail means a less reliable assessment. Always work from the file as it was originally saved or received where that is possible.
GPTOne covers the major generative families in circulation, including Midjourney, OpenAI DALL·E and GPT-4o image generation, Stable Diffusion and its many fine-tunes, Black Forest Labs Flux, Google Imagen and Nano Banana, Adobe Firefly, Sora video frames, and older GAN-based generators such as StyleGAN. Coverage is maintained as new models are released. Detection strength varies by family and by how much the file has been processed since it was created.
False positives cluster around a few recognisable cases: heavy smartphone computational photography, aggressive noise reduction or beauty filtering, studio images with very smooth lighting and clean backgrounds, and heavily compressed or repeatedly re-saved files. In each case ordinary processing has smoothed away the irregularities that mark a frame as camera-original. This is exactly why GPTOne presents a confidence score alongside its supporting evidence rather than a bare verdict.
Yes, and it is one of the most common reasons people check an image — verifying a dating profile, a job applicant's photo, or an account that may be part of a bot network. Synthetic faces from older GAN generators are among the more tractable cases. Bear in mind that profile pictures are usually small, cropped and re-compressed by the platform hosting them, all of which weakens any assessment.
Yes. The AI Image Detector is free to use, with no paywall on the analysis itself. It accepts JPG, PNG and WebP files up to 10MB and returns a confidence score, a signal breakdown, a region heatmap and a downloadable report. Paid plans exist for higher-volume workflows, advanced reporting and API access, aimed at teams, publishers and businesses running image verification at scale.
Treat the score as the first step in a review, never the last. Set and publish a policy before you start scanning, so people know the standard they are being held to. Always request and assess the original file rather than a forwarded copy. Read the evidence breakdown and heatmap rather than the headline number alone. Then corroborate with non-technical evidence: the source of the image, its capture context, and a direct conversation with the person who supplied it.

Related detection guides

Verify text as well as images

Most real verification work involves both. These guides cover the text side in the same depth.

Free tool
AI Image Detector

Check a JPG, PNG or WebP up to 10MB. Confidence score, evidence breakdown, region heatmap and a downloadable report.

Check an image →
Guide
How AI Text Detection Works

The companion guide for written content — how detection performs across model families, and why the model matters more than the score.

View guide →
Guide
ChatGPT, Claude and Gemini Detector

Model-by-model text detection breakdowns across every GPT, Claude and Gemini release, with accuracy benchmarks.

View guide →

Check an image right now

Drop in a JPG, PNG or WebP up to 10MB and get a confidence score, an evidence breakdown and a region heatmap in seconds.

Try the free AI Image Detector

Free · No sign-up · Midjourney · DALL·E · Stable Diffusion · Flux · Imagen · Firefly · Sora