By the end of this guide you will have checked a suspicious image in three ways: by looking at it and its context, by running it through an AI detection tool, and by weighing the result so you can decide how far to trust it. The tools that do the detection step are AI image detectors, which return a verdict or a probability rather than a certain answer.
If you want to know how to tell whether an image is AI-generated, the useful starting point is that no single check settles it. This guide pairs simple manual checks with detector services from our Image generation category and compares three tools, from a free web checker to enterprise deepfake platforms.
Key takeaways
- Combine a manual look at the image and its source with at least one detector, and treat the output as evidence, not proof.
- The main decision is scale: a free web checker such as AI or Not suits a one-off check, while Resemble AI and Deep Media are built for teams that screen media through an API.
- Detectors give confidence scores. AI or Not's listing states that results are confidence scores, not proof of authorship.
- For anything with legal, newsroom or safety consequences, a human expert must review the finding.
What you need before you start
- The highest-quality copy of the image you can find. The original file is better than a screenshot or a heavily compressed repost, because re-saving can strip clues and add noise.
- Where it came from. The account, page or message that shared it, and when.
- A web checker or an API key. AI or Not has a Free plan for occasional checks. Deep Media starts with a demo request, and Resemble AI's enterprise options go through sales.
- Time. A single check takes minutes. A careful investigation of an important image can take under an hour.
- Difficulty. Easy for the web checkers, moderate for an API.
- Privacy. Do not upload images of private people or confidential material to a service you have not read the terms of.
Which kind of AI tool fits this job
Pick by how many images you check and how much is at stake.
A web checker for one-off questions. AI or Not estimates whether an image, text, audio or video was made by a generative model and returns a verdict with confidence scores rather than a bare yes or no. Its Free plan includes 20 AI image checks, and it also has a Chrome extension called AI Blocker that blurs AI-generated images as you browse.
A deepfake and media authenticity platform for teams. Resemble AI offers Resemble Detect for audio, video and images, and Resemble Intelligence adds human-readable explanations of why a file was flagged. Deep Media targets trust and safety teams: its DeepID returns predictions, generator attribution, annotated heatmaps and written reasoning for each verdict. Both are delivered by API, and Deep Media does not publish pricing.
Step 1: Gather the image and its context
Good evidence starts with the original source.
- Save the highest-quality version of the image you can get.
- Record where you found it, who posted it and when.
- Look for earlier or higher-quality copies, since the first appearance of an image often tells you a lot.
- Note any claim attached to it, such as "photographed yesterday in this city".
Check it worked by writing down the source and the claim in one line each.
Step 2: Do a manual check
A careful look can raise or lower your suspicion before any tool is involved.
- Zoom in and look at hands, teeth, eyes, jewelry, hair edges and the background. Generated images often contain odd details.
- Read any text in the picture, such as signs, labels and clothing. Garbled lettering is a common warning sign.
- Check lighting and shadows for consistency across the scene.
- Ask whether the story fits: does the place, the date or the people match what you know?
Check it worked by listing what looked off, if anything. Treat that list as a lead, not a verdict, since real photos can have strange details and good fakes may have none.
Step 3: Run the image through a detector
Detectors give you a second opinion based on patterns people cannot see.
- Open a checker such as AI or Not and upload the file. For a team workflow, call its REST API, which uses a bearer-token key.
- Choose the image detector, not the text checker.
- Read the full output: the verdict, the confidence score and any model-specific details, not only the headline.
- Note that AI or Not's docs say uploaded media is processed and deleted right after inference. For other services, read the data terms yourself.
Check it worked by seeing a result that contains both a verdict and a confidence figure.
Step 4: Use a second detector or a forensic tool for important images
One tool can be wrong, so cross-check anything that matters.
- Run the same file through a second service. Resemble Detect checks audio, video and images, and Resemble Intelligence explains why a file was flagged.
- For newsroom, platform or compliance work, consider a forensic option such as Deep Media's DeepID, which provides predictions, generator attribution and annotated heatmaps through an API and SDK.
- If the image is a face swap or a possible deepfake of a real person, keep a record of every result.
- Compare the outputs. Agreement raises confidence, while disagreement means you should lean on the manual and source evidence.
Check it worked by having two independent results next to your manual notes.
Step 5: Weigh the evidence
Combine everything into a judgment you can explain.
- Put manual signs, detector results and source checks side by side.
- Give more weight to a clear source trail than to any single score, because detectors only estimate.
- Be cautious with heavily edited, cropped or compressed images, which are harder to judge.
- Decide on a plain-language conclusion: likely AI-generated, likely real, or unclear.
Check it worked by being able to state your conclusion and the main reasons in two sentences.
Step 6: Record the result and act carefully
What you do next matters more than the label.
- Save the screenshots, scores and notes with the date.
- If you publish, say what you checked and how sure you are. Do not call an image fake on the basis of one tool.
- If the image may be an illegal or abusive deepfake, report it to the platform instead of passing it on.
Check it worked by having a record another person could follow and reach the same view.
Tools for this job compared
| Tool | Good fit for | Pricing | Stated limits |
|---|---|---|---|
| AI or Not | Web, API and Chrome extension checks for images, text, audio and video | Freemium, paid from $5/mo (checked 24 Oct 2026) | Free plan has 20 AI image checks; API rate limits of 1 to 5 requests per second; results are confidence scores, not proof |
| Resemble AI | Deepfake detection for audio, video and images through an API | Freemium, paid from $280/mo (checked 24 Oct 2026) | Flex is pay as you go; large uploads, batch uploads and SSO need paid or enterprise plans |
| Deep Media | Trust and safety teams needing heatmaps and generator attribution | Contact sales | No published pricing; access starts with a demo request |
Prices and plan limits change, so confirm the current terms on each vendor's site before you subscribe.
Common mistakes to avoid
- Treating a detector score as proof. Detection is probabilistic, and AI or Not's listing says its results are confidence scores. Fix it by using the score as one input next to source checks.
- Checking a low-quality copy. Screenshots and reposts lose detail, so signals get weaker. Fix it by tracing the image to its earliest, highest-quality version first.
- Relying on a single tool. Every detector can miss or mislabel. Fix it by running a second service on anything important and noting disagreements.
- Accusing someone on thin evidence. Calling a photographer's work fake can do real harm. Fix it by stating your confidence honestly and getting expert review before making public claims.
- Uploading sensitive images carelessly. Private photos of real people deserve care. Fix it by reading each service's data terms and avoiding uploads you have no right to share.
How we put this together
This is an evidence-based guide built from each tool's listing in our directory, which is compiled from the vendor's own site and pricing page. It is not a review based on using the tools. We don't rank tools, and nobody paid to be included. Some outbound links may be affiliate links; see our disclosure.
Frequently asked questions
Is there a free way to tell if an image is AI-generated?
Yes, in a limited way. Your own manual checks cost nothing, and AI or Not's Free plan includes 20 AI image checks. Resemble AI's Flex plan has no subscription fee and bills for usage. AI or Not's free allowance is small, so use it for occasional checks.
Can AI detect AI-generated images accurately?
Not reliably enough to be the only evidence. Detectors estimate probability, and AI or Not states that its results are confidence scores, not proof. Edited, compressed or cropped images are harder to judge, so combine detector output with source checks.
Can detectors find deepfakes as well?
Several do. AI or Not lists deepfake detection, Resemble Detect covers audio, video and images, and Deep Media focuses on deepfake analysis. For a deepfake involving a real person, keep records and involve an expert before acting.
Which tool is good for checking images in a browser?
AI or Not has a Chrome extension called AI Blocker, which scans pages and blurs AI-generated images as you browse. It is a convenience feature, so confirm anything important with a proper check.
Can I keep the image private?
AI or Not's docs say uploaded media is processed and deleted right after inference. Deep Media is delivered by API or MCP, and Resemble AI lists cloud, on-prem and air-gapped deployment. Read each vendor's data terms before uploading sensitive images.
Compare more options in our Image generation category. If you are making images yourself, see how to turn a sketch into finished art with AI.
