AI Image Detector.info

AI Image Detector.info is an AI checker tool. A multi-model AI image detector that runs four independent engines under a published weighted-vote rule, checks C2PA credentials first, and returns graded conclusions with every vote named rather than a single percentage.

AI Image Detector.info is a web based screening service built around a simple premise: a single model outputting one confidence number is not evidence. The tool runs four independent detection models over an uploaded image and combines their outputs with a published weighted-vote rule, so the reader can see which engine said what rather than trusting a blended figure. Each engine is credited by name in the report, and when the engines disagree, the disagreement is stated plainly instead of being smoothed away. The detection stack mixes commercial and open source components. Sightengine GenAI handles per-generator classification across models such as Midjourney, Flux and Imagen. Winston AI and Was It AI supply independent second and third opinions, while the open source SAFE model (KDD 2025) runs on the service's own hardware as an auditable fourth vote. Above all of these sits a deterministic layer: C2PA credentials and generator self-tags are checked first and, when present and unambiguous, overrule the models outright. The output is a graded conclusion on a five point scale, from AI-generated through inconclusive to real, backed by the visible exhibit and a visual inspection note. The product fits people who need a defensible answer rather than a quick hunch. Trust and safety reviewers, marketplace moderators, dating app users, educators handling integrity disputes, artist communities policing submissions and editorial teams documenting provenance all appear in the tool's intended audience. Batch screening handles up to 50 images at once with a triage summary and CSV export, and reports can be shared by link or downloaded as a forensic PDF that includes the file's SHA-256 hash for chain-of-custody records. Against conventional detectors, the distinguishing choice is architectural. Instead of one model and one percentage, this one spreads risk across multiple engines and reports conflict openly, and a failed engine refunds its credit automatically rather than billing for silence. JPEG, PNG and WebP files up to 12MB are accepted, and stripped files still get screened, though the service itself concedes that screenshots, memes and metadata-poor images yield weaker evidence and are hedged accordingly.