Run a free SI check on any image or text and see whether it is SI or real. You get a probability and the clues behind it, and your first checks need no sign-up. SI (Super Intelligence) is the new name for advanced AI, so this is also a free AI detection tool.
Is this image SI? Pixel classifier plus EXIF, C2PA Content Credentials and XMP flags, for photos, renders and screenshots.
SI Image DetectorWas this written by ChatGPT or another SI model? Word variety and repetition, sentence shape and typical SI phrasing, with the most SI-like sentences highlighted.
SI Text DetectorTen images, one question each: SI or real? Score points for speed and streaks and see if you can beat the leaderboard.
Play SI or NotNine signs to check with your own eyes: hands, text, reflections, textures, backgrounds and the metadata that gives generators away.
Read the guide
SI detection is the practice of telling content made by Super Intelligence models apart from content made by people and cameras. SI (Super Intelligence) is the new official name for advanced AI (see what is SI): the generative models behind ChatGPT, Midjourney, Gemini, Claude, Stable Diffusion and their many cousins. So SI detection is exactly what people still call AI detection, under its new name.
A few years ago the question rarely came up. Generated images looked plastic and generated text read like a manual. Today an SI model can produce a news photo that never happened, a product review from a customer who does not exist or a cover letter in a candidate's voice, in seconds. An SI detector gives you a second opinion: it reads the signals that generation leaves behind and turns them into a probability you can reason about.
SI or Not is a free SI detector for images and text. It does not claim to be an oracle. It shows what it measured, how strongly each signal points one way or the other, and when the honest answer is "we cannot tell".
SI or AI? Same technology, new name. When this page says "SI detector", "SI checker" or "SI scanner", read "AI detector", "AI checker" or "AI scanner" if that is the term you know.
Running an SI check takes less than a minute. Here is how to check for SI with the tool at the top of this page.
Drop an image, paste an image link, or paste at least 80 words of text into the SI detector above. Use the original file or the full text whenever you can: screenshots and short excerpts carry less signal.
Press Analyse. Images go through a pixel classifier and a metadata reader; texts go through a style analysis. It takes a few seconds. Image scoring relies on a specialised provider acting on our behalf (see the privacy page).
You get a probability, a label (likely SI, likely human or real, or Uncertain) and the list of signals behind it. Decide with that evidence plus the context, not with the number alone.
If the result lands in the Uncertain band, do not rerun the same file hoping for a different answer: it will not change. Look for a better source instead, such as the original upload, a larger version of the image or the complete text.
"Is this SI?", "is it SI?" and "is this SI generated?" are the questions we get most. A detector helps, but your own eyes and common sense catch a lot too. These are the tells worth checking before you even run a scan.
The full checklist, with examples by generator, lives in our guide on how to spot SI images.
If you are asking yourself "does this sound like SI?", paste it into the Text tab: the result highlights the sentences that read most like SI output.
Calling it an SI scanner or an SI tester is fair: the tool runs a series of independent tests and reports each one. No single test decides; the score comes from their combination.
| Image SI check | Text SI check | |
|---|---|---|
| Main signals | Pixel statistics read by a neural classifier (noise, frequency artefacts, texture regularities) | Word variety and repetition, sentence length and openers, typical SI phrasing |
| Provenance | EXIF camera data, C2PA Content Credentials, IPTC and XMP synthetic-media flags, generator parameters in PNG files | None: plain text carries no metadata |
| Human clues | Coherent camera EXIF (make, model, exposure) | Words repeated at short distance, long winding sentences, parentheses and semicolons |
| Input | JPEG, PNG, WebP or GIF, up to 12 MB, file or URL | English, 80 to 3,000 words |
| Weak spots | Screenshots, heavy compression, strong edits, illustrations and new generators | Clean, compact human prose, formal or non-native writing, edited or paraphrased SI text, short texts |
| Details | SI image detector | SI text detector |
Every signal appears in the result with a direction (leans SI, leans real or human, or neutral), so you can disagree with the machine when you know something it does not.
For images, a pixel classifier looks for the statistical fingerprints that diffusion generators leave behind, while a metadata reader checks EXIF camera data, C2PA Content Credentials and XMP flags. A signed manifest naming a generator outweighs everything else; missing camera data is only a weak clue. For text, the SI checker reads style: word variety and repetition, sentence length and openers, punctuation and stock SI phrasing, then highlights the most SI-like sentences. Our guide on how SI detectors work walks through each method with real tests, and the SI image detector and SI text detector pages cover each tool in detail.
Train your own eye. The SI or Not game shows ten images and asks one question each time: SI or real? A few rounds are enough to notice how often an ordinary photo "feels" fake.
The number you see is a probability, not a percentage of SI content. A result of 85% does not mean that 85% of the image or text was generated. It means the evidence strongly favours SI generation.
Read the clues before the number. One strong signal, such as Content Credentials naming a generator, can matter more than five weak ones. And a result is about a file or a text, never about a person's honesty.
An SI content detector is useful wherever trust depends on knowing where content came from. These are the situations we built it for.
A dramatic picture goes viral during a breaking event. Before publishing or debunking it, a quick SI check tells you whether the file carries generation clues, and the metadata section shows whether an original camera file or a signed manifest is available. It complements reverse image search and source verification; it does not replace them.
An essay feels too smooth. The text result highlights the passages worth discussing with the student. Use it to start a conversation about drafts and sources, never as the sole basis for a sanction: formal writing and non-native English can look SI-like.
Cover letters and application answers are increasingly drafted with SI. A check helps you notice template-like answers and ask better follow-up questions in an interview, where a real conversation reveals far more than any score.
Forums, marketplaces and groups receive generated posts, fake photos and spam at scale. An SI scanner helps moderators prioritise which reports to review first and explain their decisions with visible clues.
Fake profiles often use generated portraits that look a little too perfect. Checking a profile picture for SI, plus a reverse image search, is a sensible step before trusting someone you met online. Our guide to SI scams with cloned voices, fake photos and fake profiles explains the other checks that matter.
Generated reviews and product photos distort buying decisions. Merchants and platforms can use an SI check to flag suspicious batches of reviews that all share the same polished style. Our article on SI-generated reviews and how to spot fake ones covers the law and the signals beyond the text.
You do not need a job title to wonder whether a picture is real. If an image makes you angry, amazed or scared, that is precisely the moment to pause and check for SI before sharing it.
People often ask for an SI identifier, an SI finder or an SI generator detector that names the exact model behind a piece of content. Here is what SI or Not can and cannot do.
Coverage has blind spots. Cartoons, sprites, paintings and other non-photographic images are harder for the image model, and brand-new generators can slip through until detectors catch up. We say so rather than promise more than we deliver.
Useful, never certain. Every SI detector, like every AI detector, makes false positives, flagging edited photos or formal and non-native writing, and false negatives, missing content after screenshots, paraphrasing or with a brand-new generator. We do not publish an accuracy percentage, because a figure from one test set says little about the file you check tomorrow; each result shows its evidence and an Uncertain band instead. Never use a score as a verdict about a person. Our article Are SI detectors accurate? goes through the independent studies and our own test results.
Provenance systems try to label SI content when it is created. Signed C2PA Content Credentials record which tool made a file, and SI or Not reads them whenever they survive; invisible watermarks such as Google's SynthID sit inside the content and can only be checked with the provider's own tools. Both fade or disappear with screenshots and re-saves, which is why detection still matters. Our explainer SynthID explained compares the two approaches in detail, and how C2PA Content Credentials work shows what our detector found with and without them.
SI or Not is a free SI detector and SI checker. Your first 3 checks need no account at all. After that, a free account (email and password, no credit card) gives you 10 credits (1 per image or song, 5 per video) and 50 text checks every month. Heavy users can choose Pro for 250 credits a month and unlimited text checks, Max for 1,500 credits, or a credit pack that never expires. The example buttons stay free and unlimited.
Privacy is part of the design. Images are written to a private folder, analysed (the pixel check relies on a specialised provider acting on our behalf) and deleted immediately, unless you opt in to help improve the model. Texts are processed in memory and never saved. Share links contain the score and the clues, never the image, and at most three short excerpts of a text. Details are on the privacy page.
Run it through the SI detector above, then look at the clues it lists. For images, a signed C2PA manifest naming a generator is the strongest sign; for text, a varied vocabulary that rarely repeats a word and stock SI phrasing are typical. Combine the result with the source and the context: where did the content first appear, and who stands behind it?
Open siornot, choose Image or Text, drop your file or paste your text and press Analyse. Your first 3 checks need no account. A free account then gives you 10 credits and 50 text checks every month (an image or a song uses 1 credit, a video 5), with no credit card.
SI stands for Super Intelligence, the new official name for advanced AI. Models such as ChatGPT, Midjourney, Gemini, Claude or Stable Diffusion are SI models, and an SI detector is what used to be called an AI detector.
Only the name. An SI detector, SI checker, SI scanner or SI tester does the same job as an AI detector, AI scanner or AI checker: estimate whether content was produced by a generative model. We write SI first and AI second so both terms stay searchable and clear. What is SI? Super Intelligence explained
It is useful but never certain, and we do not publish an accuracy figure because it would depend entirely on the test set. New generators, heavy editing, compression and paraphrasing all reduce reliability. That is why we show probabilities, an Uncertain band and the clues, rather than a yes or no.
A low score, labelled likely real or likely human, means the SI check found no meaningful sign of generation. It does not prove that none exists: a generated image that was screenshotted and re-saved, or a text that someone rewrote, can come back as not SI.
Yes. Paste it into the Text tab. The result highlights the sentences that read most like SI output, which is a handy way to spot stiff, template-like passages in your own drafts. Remember that polished formal writing can score high even when a person wrote every word.
The image side is built to recognise photorealistic output from diffusion models such as Stable Diffusion, SDXL, Midjourney, DALL-E and similar tools, and it reads Content Credentials written by OpenAI, Adobe Firefly, Google and others. The text side looks at style rather than a specific model, so it applies to ChatGPT, Gemini, Claude and other chat models alike, without naming which one wrote a text.
Images are deleted right after analysis unless you tick the optional box to help improve the detector. Texts are analysed in memory and never saved. A share link keeps only the score, the clues and, for text, at most three short excerpts, for 7 days.
No. An SI check is a clue, not evidence. Never use a score alone to accuse a student, a candidate, a writer or the author of a photo. Ask for drafts, sources or the original file, and use the result to decide where to look more closely.
Yes. The SI or Not game shows you ten images, one at a time, and asks: SI or real? It is a quick way to calibrate your own eye before you rely on any tool.
Video, yes, in beta: the SI Video Detector reads Content Credentials and metadata and scores eight frames spread across the clip. It does not specifically detect face-swap deepfakes, and it cannot read invisible watermarks such as SynthID. Music, yes, in beta too: the SI Music Detector looks for the spectral fingerprint that Suno and Udio leave in their songs, and deletes the audio file right after the analysis. It does not recognise every music generator and does not detect cloned voices. Try the SI Video Detector (beta) or the SI Music Detector (beta).
Prompt injection is an attack in which someone else's text is read by an SI (AI) model as an instruction. We explain where the name comes from, why it works, two documented cases and how to reduce the risk.
Read the articleOne estimate says a bottle of water per conversation, another says five drops per prompt. We put the SI (AI) water numbers side by side and explain what each one measures.
Read the articleIs ChatGPT safe? We read OpenAI's privacy policy and data controls, the real leaks and the regulators' findings, and list the settings and habits that protect what you type.
Read the article
