About · Product and principles
Product scope and publishing principles
GenDetect is a free tool and publishing brand for screening AI-content signals. The current version is local by default; optional remote image modes disclose upload and deletion. All results emphasize explainability and human review.
01 · Product scope
The site places a working detector, methodology, and verification guidance together so a user can see both a score and what that score can and cannot establish.
The current version requires no login, sign-up, or stored history. Text, video, and the default image heuristic run in the browser; optional remote image modes return provenance or metadata clues. It is not an identity, discipline, hiring, or legal decision system.
Open the GenDetect detectorText screening examines repetition, sentence length, punctuation, and character structure. Image screening combines known generator metadata declarations with pixel rules. Video applies similar pixel rules to three sampled frames and displays file metadata. The three indexes are not interchangeable and cannot reliably attribute a specific generator.
Screening signals, file facts, and limitations identify what to review next. A stronger conclusion also needs the original file, content credentials, version history, an author explanation, publishing provenance, and contextual evidence—not a single detector result treated as a final verdict.
Read the AI content verification guide02 · Method and limits
The current implementation is local-first with limited remote image provenance or metadata clues. It is not a source classifier trained, calibrated, and continuously monitored on a representative dataset.
Text, image, and sampled-video indexes combine explicit observable features, and the interface exposes supporting signals. If video frames cannot be decoded, the flow reports metadata facts without a generation score. Language, genre, editing, compression, sample length, and device processing can all alter the available evidence.
Read the methodology and limitationsHuman writing can be repetitive or highly regular, and real photos can be heavily processed. AI content can also be rewritten, edited, or compressed to weaken signals. A high text or image index is not proof, a low index does not rule generation out, and video metadata cannot determine origin on its own.
Review accuracy and false positivesGenDetect currently compares features inside the submitted material. It does not search the web, paper repositories, or publishing databases for matching sources. AI detection, plagiarism checking, fact-checking, and provenance verification address different questions and may need to be combined.
Compare AI detection and plagiarism checking03 · Publishing and upgrade principles
GenDetect uses production code and public pages as the reference. A change to the algorithm, data flow, supplier, or account capability requires corresponding methodology, privacy, and interface updates.
Product pages and guides do not present a planned deep model, frame-by-frame video analysis, content-credential verification, or generator attribution as a current feature. A page date identifies the public disclosure version; it is not independent certification or an accuracy endorsement.
A future algorithm upgrade should document input scope, feature or model version, threshold rationale, representative test sets, false positives and false negatives, language or media differences, and uncovered cases before methodology and score explanations change.
Optional remote image modes now disclose upload, purpose, and deletion before submission. Any new supplier, account, or cloud-history feature must additionally name submitted fields, retention, training use, and user controls in the synchronized privacy notice.
Read the current privacy notice