Combined risk
Image features, distribution state and model evidence contribute to an AI-generation risk probability and basic conclusion.
WeChat Mini ProgramAI image detection
AI Image Risk & Evidence
A risk assessment,with reasons.
Yunyulai AI Detection is for ordinary users who need to assess whether an image may have been AI-generated. It turns a risk probability, key evidence and practical boundaries into a result that is easier to understand.
Yunyulai AI DetectionRisk · Evidence · Report
What You Get
Turn one conclusion into information you can review
The result is more than a probability. It explains key evidence, distribution state and where the current assessment remains uncertain.
Supporting evidence
Key signals are organised so users can see what influenced the result and what remains uncertain.
Optional report
The basic assessment is free. A structured full report can be generated, saved or shared when needed.
Detection Flow
From selecting an image to understanding the result
Each step stays focused on the image at hand instead of presenting a single score as an absolute verdict.
- 01Select an image
JPG, JPEG, PNG and WebP are supported; large mobile originals are compressed when needed.
- 02Pass content safety checks
The image first uses WeChat's content-safety route. Content that fails does not enter subsequent analysis.
- 03Combine evidence
Image features, distribution distortion and model evidence are combined into a basic risk assessment.
- 04Review and retain
Read the basic conclusion and key evidence, then generate, save or share a full report if needed.
Result Boundary
Decision support, not a replacement for source verification
AI-image detection deals with probabilities and changing generation methods. Use the result alongside provenance, context and further checks.
Not an absolute verdictA risk probability is not a final ruling on authenticity, nor is it a legal or forensic conclusion.
No model covers every caseNew generators, compression, screenshots, recapture and editing can change visible evidence, so uncertainty remains.
Images are not training dataUser images, image-level features and results are not used for model training; originals are retained temporarily and deleted under product rules.