Argus runs an uploaded image through a neural ensemble plus deterministic forensic checks — the statistical fingerprints diffusion models leave behind, provenance signals like C2PA credentials and SynthID watermarks, and EXIF metadata anomalies. It returns a synthetic-probability score and a Grad-CAM heatmap highlighting the regions that looked artificial, so you see why an image was flagged rather than trusting a single number.

Yes. Video is split into frames and analyzed with temporal-vector scanning that catches the frame-to-frame inconsistencies and warping artifacts deepfakes and face-swaps tend to leave — signals single-frame detectors miss. Argus supports MP4, MOV, WEBM, and AVI.

Argus targets the artifacts common to modern generative models — diffusion-based image tools and GAN face generators alike — rather than fingerprinting one product. Because new generators appear constantly, the detection ensemble is benchmarked against the latest models on an ongoing basis.

Accuracy depends heavily on the generator and how the media has been compressed or edited, so Argus reports a probability with an explainable heatmap rather than a yes/no guarantee. Results are best treated as strong forensic evidence to weigh alongside context — they are probabilistic estimates, not absolute proof.

Your first scan is free with no account required. Creating a free account unlocks additional scans, heatmaps, and downloadable PDF reports, and credits cover higher volumes.

Images in JPG, PNG, and WEBP, and video in MP4, MOV, WEBM, and AVI. You can upload a file, drop a whole folder for batch scanning, or paste a direct media URL.

No detector is infallible — heavy compression, screenshots, and adversarial editing can degrade any model's signal, which is why Argus combines multiple independent models with deterministic provenance checks and shows its reasoning. Conclusive evidence like a verified C2PA credential can override a probabilistic guess, but borderline results should always be read as estimates.