Model outputs explained
View the content score, final risk score, detail indicator and propagation confidence, together with the model’s decision threshold.
Yunyulai AI Image Detection · Web version available
AI image detection.Results you can explore.
Upload an original image and explore AI-generation risk, model indicators and image metadata in a report you can understand and save.
The detector and its reports currently use a Chinese interface.
Explore the information behind a score
File and propagation details
See image dimensions, format, color mode and available camera, software and time metadata, alongside the model’s propagation classification.
Reports you can keep
Download a text report, revisit records in the same browser, delete detection data or submit feedback about a result.
Three steps from image to report
- 01
Choose an original file
Select, drop or paste an image. Review its dimensions and size, then select the source you know about.
- 02
Submit for model analysis
Read the service information and start detection. After file validation, Yunyulai’s self-hosted models analyze the image.
- 03
Read and save the report
Interpret scores, thresholds, propagation features and metadata together. Save the report and investigate the source when needed.
Before you begin
- Formats
- Single-frame JPG / JPEG, PNG and WebP
- File limit
- Currently 10 MiB; see the tool for the active limit
- Requirements
- A modern browser; no app installation
- Pricing
- The web version is currently free
- Processing
- Original files are uploaded to Yunyulai’s private servers
- Records
- Reports retained for up to 90 days; deletable by you
Powered by Yunyulai’s existing models
The web service uses self-hosted content, propagation and fusion models, followed by score calibration. Reports include model and file identifiers. Metadata is parsed from the original file for reference. Images and results are not used for model training.
Detection assesses AI-generation risk in the main image content and supports human review. Screenshots, compression, local edits and recaptured images can affect results. Scores do not establish that a depicted event is real.
Service and privacyAbout the web detector
Can this detector prove an image is real or fake?
It provides AI-generation risk indicators, not verification of depicted facts or authorship. Consider the original file, its source and creation records.
Are images compressed before upload?
The web app submits the selected original file without compressing or replacing it for previews. The inference service still applies the model’s own preprocessing.
What information is included in a report?
Reports include model risk scores, decision thresholds, content and detail indicators, propagation features and available image metadata. Missing information is explicitly identified.
How long are images retained?
Image cleanup starts when processing ends, with a 24-hour fallback limit. Structured reports are retained for up to 90 days and can be deleted. Images, features and results are not used for training.
Do I need an account or payment?
The web version uses an essential browser cookie session, without account registration. Detection and report downloads are currently free. Clearing cookies or changing browsers may make earlier records inaccessible.