DEEPCHECK - v1.0
Preliminary forensic visual screening for visible image-generation and manipulation indicators.
KEEP THE RESEARCH MOVING
Independent checks, transparent limits, and future research updates. Support channels will be announced publicly.
Choose or drop a JPEG, PNG, or WebP image — maximum 8 MB / 20 MP.
Three free checks per browser each day. No account required.
Choose or drop a JPEG, PNG, or WebP image (maximum 8 MB / 20 MP)
HOW IT WORKS
The result is organized as localized visual evidence, not a final authenticity decision.
- 01VISUAL REVIEW
Multimodal reasoning reviews visible consistency across anatomy, texture, lighting, reflections, geometry, typography, and object boundaries.
- 02LOCALIZED TARGETS
Relevant observations are mapped to normalized image coordinates and displayed as Target boxes.
- 03EVIDENCE SUMMARY
Each Target describes the visible evidence, why it matters, and the limits of the interpretation.
- 04BETA BOUNDARY
The screen does not verify metadata or provenance and does not establish authenticity.
PRELIMINARY SCREENING REPORT DEEPCHECK v4.0
Reviews visible consistency in structure, lighting, geometry, text, and object boundaries. It does not inspect provenance records or establish authenticity.
PUBLIC BETA ROADMAP
We are building this in public: calibrate image checks first, then publish non-sensitive research updates.
- NOW · PUBLIC BETAThree no-account checks, visible limits, and abuse safeguards.
- NEXT · TRANSPARENCYPublish methodology and evaluation summaries without exposing private systems.
- LATER · COLLABORATIONConsider research and commercial pilots only after calibration and review.