Early-Signal Detector

task · first seen 5 Oct 2026

deepfake detection

Also: AIGI detection, image forensics, AI image detection, deepfake detectors, fake image detection, digital image forensics, synthetic image detection

Automatically determining whether an image or video has been manipulated or synthetically generated.

Deepfake detection is the task of deciding whether an image, video or audio track was manipulated or produced by a generative model. The sources show steady academic activity rather than a new idea. New benchmarks cover multilingual audio-visual fakes and fabricated surveillance footage. Several papers audit evaluation practice, finding that detector accuracy falls on newer generators and that reported fusion gains are often evaluation artifacts. Others propose physiological, lip-motion, physics-guided and test-time-adaptation cues. On the tooling side, there is an on-device browser extension, a Chinese multimodal fusion repo, and a vendor launch of Pangram Image.

Generalization to unseen generators and adversarial robustness remain unsolved, so the field's shift toward rigorous cross-domain evaluation and provenance-based authentication could change how detectors are trusted and deployed.

Analyst note (established), 17 Nov 2026.

Mentions per week

0919 May11 Aug10 NovPapersCodeDiscussion

PapersCodeDiscussion

Rank 41 on 14 Nov 2026 · 29 mentions in 28 days · 29 independent groups · papers → code → discussion · first seen 6 weeks ago · acceleration +0.50

Earliest sources

Latest sources

Mentioned together with

low-rank adaptationcontinual learningparameter-efficient fine-tuningC2PA content credentialscontrastive language-image pretrainingplausible deniability of provenancelip synchronizationself-hosted LLM deploymentlip sync forgery detection3D morphable model