A free tool now checks whether a photo was made by AI

Phrasly, a company known for AI writing tools, has released a free tool that checks whether a photo or piece of artwork was made by AI. According to its November 3, 2026 announcement, it needs no account and has no limit on how many images you can check.

What it does

You can upload an image, drag and drop it, paste it, or give the tool a link. It accepts JPG, PNG, WebP, HEIC, AVIF and GIF files up to 20 MB. It gives you a verdict on whether AI was involved, a confidence score, and a heatmap that marks the areas it is reacting to. Phrasly says it checks each image in memory and does not store it.

Why it exists

People are not good at spotting fakes. In a 2025 test by Microsoft's AI for Good Lab, more than 12,500 participants saw a mix of real and AI-made photos. They got it right 62 percent of the time.

Many AI tools mark their images with invisible watermarks or signed records inside the file. Phrasly points out that a screenshot, a re-upload or a copy saved from a social feed can wipe those marks out. Some generators never add them at all.

How it checks

Phrasly says the tool runs three kinds of checks and shows which one produced the answer:

• Its own model, trained to read the picture itself rather than the file's hidden data.
• SynthID, an invisible watermark from Google DeepMind used on images from Google and OpenAI tools. If found, the verdict is conclusive.
• Signatures from ChatGPT, Gemini, Adobe Firefly, Copilot and FLUX, checked against the official C2PA trust list.

If the file carries nothing, the result rests on the model alone. The confidence score shows how sure it is.

Many images are real photos with one part changed, such as a new background or a retouched face. For those, the heatmap points to the area that was generated instead of giving a single yes or no.

How far to trust it

Phrasly says the tool is over 99 percent accurate in its own testing, with or without a watermark. It lists images from ChatGPT and DALL-E, Gemini, Midjourney, Stable Diffusion, FLUX and Adobe Firefly among those tested. Outsiders have not verified that figure, and the source is the company's own announcement. A verdict is best treated as one clue, not proof.

Who it is for

Phrasly names journalists, teachers, moderators and buyers as the main users. The risks it describes include publishing something false, grading work that was never done, or paying for a product that does not exist. The company says the check works both ways. It can flag a generated image, and it can also back up one that is real.