Check any photo for AI and see exactly which part was generated. Free, no account, and it works long after the watermark is gone.
DOVER, Del., Nov. 03, 2026 (GLOBE NEWSWIRE) — Phrasly, an AI writing platform used by more than 5 million people, has released a free AI image detector that identifies AI generation in artwork and photos.
Generated images have reached the point where most people cannot identify them simply by looking at them. In a 2025 experiment by Microsoft’s AI for Good Lab, more than 12,500 participants were shown a mix of real photographs and AI-generated ones, and identified them correctly 62 percent of the time. With AI improving by the minute, expect those percentages to drop significantly in the coming months.
The industry’s answer has been to mark AI images at the source, with invisible watermarks and signed records written into the file. That holds up until someone takes a screenshot. The images that do the most damage are usually the ones that arrive carrying nothing at all, and that is exactly what Phrasly’s detector is for.
What Is the Phrasly AI Image Detector?
The detector returns three things on any image: a verdict on whether AI was involved, a confidence score, and a heatmap marking the areas it is reacting to. It is free to use, requires no account, and checks each image in memory rather than storing it.
It accepts JPG, PNG, WebP, HEIC, AVIF, and GIF files up to 20 MB, by upload, drag and drop, paste, or image link. It is the first detection tool Phrasly has built for images, alongside its AI text detector.
Why a Watermark Was Never Going to Be Enough
Watermarks and metadata come off by accident as often as on purpose. A screenshot removes them, as does a re-upload or a copy saved from a social feed. Many generators never add them in the first place.
So Phrasly trained its own model instead, on millions of real photographs and AI images from every major generator. The model reads the picture itself, so it does not depend on anything the file happens to carry.
In Phrasly’s own testing it is over 99 percent accurate with or without a watermark, on images from ChatGPT and DALL-E, Gemini, Midjourney, Stable Diffusion, FLUX, Adobe Firefly, and others.
“A year ago most people could still spot an AI image if they looked closely enough,” said Victor Rijo, CEO at Phrasly. “That is no longer true, and by the time an image reaches the person who has to make a decision about it, it has been screenshotted and reposted enough times that whatever proof was in the file is gone. We built the detector to read the picture itself, because that is the part that survives.”
What Every Scan Checks
Every scan runs three checks.
The first is Phrasly’s own model, which reads the image directly. This is the check that still works once everything else has been stripped away.
The second is SynthID, an invisible watermark developed by Google DeepMind and applied by Google and OpenAI to images from their tools. When it is present, the verdict is conclusive.
The third is AI signatures. ChatGPT, Gemini, Adobe Firefly, Copilot, and FLUX sign the images they produce, and Phrasly verifies each signature against the official C2PA trust list.
Where the file carries nothing, the result rests on the model alone and the confidence score reflects how certain it is. Every check appears beside the verdict, so a reader can see which kind of answer they are looking at.
Catching the Photos That Are Only Partly AI
The harder case is the image that is only partly generated. Most pictures in circulation were not made from nothing. They are real photographs with one element changed: a swapped background, an added object, a retouched face. A verdict on the whole image is of little use there, because the honest answer is both.
Phrasly returns a heatmap instead, marking the area that was generated so a reader can look straight at it. The same model handles screenshots and re-uploads, which have lost their metadata entirely and were never going to be identified any other way.
Who Phrasly Built It For
The detector is built for anyone who needs to know an image is real before acting on it. That includes journalists checking a photograph before it runs, teachers reviewing artwork handed in as original work, moderators clearing listings and profile pictures, and buyers deciding whether the product in a photo actually exists.
What they share is the same risk. A wrong call means publishing something false, grading work that was never done, or paying for something that was never there. At the far end it means a fabricated image spreading as evidence during breaking news, standing behind an accusation against someone who was never there, or driving a scam that works because a photograph used to settle the question.
The check runs both ways, exposing an image that was generated and confirming one that was not.
Availability
The Phrasly AI Image Detector is available now at no cost, with no sign-up and no limit on the number of images. It accepts JPG, PNG, WebP, HEIC, AVIF, and GIF files up to 20 MB, by upload, drag and drop, paste, or image link, and each image is checked in memory rather than stored. The product page also presents four photographs and invites readers to identify the AI ones before showing what Phrasly found in each.
About Phrasly
Phrasly is an AI writing and detection platform used by more than 5 million people in over 180 countries. Its tools include the AI Humanizer, AI Detector, AI Image Detector, Plagiarism Checker, Content Generator, and the Phrasly Docs editor. Phrasly builds its detection and humanization models in house rather than running on another provider’s API. Phrasly, LLC is based in Dover, Delaware. Learn more at https://phrasly.ai
CONTACT: Contact Information: Phone: 302-203-9836 Email: support@phrasly.ai
