AI image detection
The Image Detector reads an image through complementary model, forensic and source-signature checks. The combined report keeps the main probability and the strongest evidence together.
Asynchronous job lifecycle
- Upload
The signed-in web route validates a JPG, JPEG, PNG or WebP file up to 15 MB.
- Queue
A private owner-scoped job is created and claimed by the image worker.
- Analyze
The worker normalizes the image, runs the configured scan profile and records typed scanner results.
- Aggregate
Completed evidence is combined into one result while optional scanner failures remain isolated.
- Return
The web app polls the job and renders the report when processing is complete.
Image signal families
| Family | Examples | Purpose |
|---|---|---|
| Source signature | Metadata, EXIF, C2PA | Read available creation and editing context |
| Visible labels | Tesseract OCR | Find explicit AI labels and provider marks |
| Pixel forensics | FFT, DCT, residual noise, blur, edges | Measure file-level patterns |
| Specialist models | SSP and CO-SPY | Compare texture, semantic and pixel representations |
| Primary visual model | Deployment-configured visual bundle | Add a global scene-level reading |
File handling and retention
The source image is held for the short processing window, with a default retention of about one hour, then removed. The derived typed result can remain for reliability and report access, while the original image is not copied into text history.
Owner scoping applies to job creation, status and results. Browser code cannot request another account's job or select a deeper profile than the server permits.