All tools/AI Image Detection

AI Image DetectionImage provenance and risk

Review the model’s risk score, provenance clues and file metadata alongside the original image. Save the findings in a report.

How to read an image report

Read the score against the threshold shown in the report.

Does this image warrant a closer look?

When a contributor sends an image of uncertain origin, check where the score falls relative to the model threshold. Review the detail indicators alongside the image. A high score can be a reason to ask for the original file.

Screenshots, compression and local edits can affect the score. The score alone cannot identify who made the image.

Reviewing submissions and handing off assets

Before publishing a submission

If a photo from an event arrives as a chat screenshot, ask the photographer for the original. Analyze that file and note the editing records or model indicators that need a follow-up before publication.

Hand a review to a colleague

Send the corresponding image with the report. Note which claims the file contains and which questions still need an answer from the source, so your colleague can pick up the review.

Explore a sample report

AI Image DetectionReport preview
Generated illustration of a cat in a paper garden
garden-illustration.pngGenerated illustration for this sample report
Read the score alongside its threshold.

Compare the score with the threshold, then review the image itself.

The score indicates risk. It cannot establish authenticity.

Explore the sections of a sample report. This demo does not analyze the image or generate a risk score.

FAQ

Which images can I upload?

Select or drop a single-frame JPG, PNG or WebP image up to 10 MiB. The file is uploaded to Yunyulai’s private service for analysis. For a provenance check, use the original supplied by the photographer or contributor when available.

Is the score the probability that AI generated the image?

The model reports a risk score. It is neither a calibrated probability across all uses nor a measure of overall accuracy. A low score cannot establish that a camera took the image. Check high scores against the original file and its source.

Will my image be used for training?

Images, extracted features and results are excluded from model training. Cleanup of the original starts after processing. In exceptional cases it may be retained for up to 24 hours. Reports are kept for up to 90 days and can be deleted sooner.

Share something useful

Share

Send a direct link so your friends can open this page.

Visit statistics

We count page views, channel sources, product-link clicks and sharing actions to understand how our products are discovered. These statistics use no visitor identifiers, cookies or cross-site tracking, and do not read uploaded files or form contents.