SANDS Lab launches FakeCheck 2.0, evolving to detect generative AI deepfake images and videos
AI models trained primarily on Korean facial characteristics establish a Korea-focused deepfake detection service Lengthy verification undermines victim protection; users can check directly without waiting AI cuts verification time and work requiring hundreds of responders to one-tenth, in a field that needs AI-versus-AI defenses
SANDS Lab launches FakeCheck 2.0, evolving to detect generative AI deepfake images and videos
- AI models trained primarily on Korean facial characteristics establish a Korea-focused deepfake detection service
- Lengthy verification undermines victim protection; users can check directly without waiting
- AI cuts verification time and work requiring hundreds of responders to one-tenth, in a field that needs AI-versus-AI defenses
Photo 1. The FakeCheck 2.0 deepfake detection service
AI security technology specialist SANDS Lab (CEO Kim Ki-hong) announced that it has completed the 2.0 update of FakeCheck, the deepfake detection service it opened to the public last year.
FakeCheck is an online service that allows users to upload a suspected deepfake image directly to its website. AI analyzes the data and determines whether it is manipulated within seconds. Anyone can use the service free of charge from a PC or mobile device without cumbersome registration or verification procedures.
The 2.0 update incorporates deepfake trends and improves features for more precise detection. In addition to increasing the number of AI detection models, SANDS Lab developed specialized models trained on public datasets and facial images collected with explicit consent. This significantly improves detection accuracy for deepfake images in specific domains and those targeting middle-aged and older Koreans, enabling balanced analysis across all age groups.
The company also enhanced separate models to distinguish not only images in which existing faces have been altered using face swapping, but also fictional people created by generative AI such as ChatGPT, improving detection rates. In internal tests using 100 genuine images from the web and 200 deepfake images generated with GPT-4o, SANDS Lab’s AI technology development team demonstrated strong performance with **99.33% accuracy** and a 99.5% F1 score.
Photo 2. FakeCheck 2.0 analysis results. Detection rates have improved by tailoring detection models to image characteristics.
Deepfake video detection is also scheduled to be introduced with the 2.0 update. A SANDS Lab representative said technical development has already been completed, but optimization is underway to preserve the service’s free public availability. Details such as video length and file size are being adjusted to avoid introducing fees, and video detection could be available as early as this month.
FakeCheck 2.0 is thus a Korea-focused deepfake service trained on Korean facial characteristics and specialized in analyzing the rapidly growing volume of generative AI imagery. It can immediately detect deepfakes that are difficult to distinguish with the human eye, potentially reducing to one-tenth the work that previously required hundreds of responders and second and third rounds of verification. This reflects the need to counter AI with AI rather than people alone, alongside the development of appropriate AI security technologies.
CEO Kim Ki-hong said, “When verification is required to protect victims of malware attacks or deepfakes, the critical response window can be missed and protection can lose its effectiveness. AI must therefore solve the problem quickly.” He added, “We understand better than anyone how victims feel when they can do nothing but wait for verification to finish. With FakeCheck, we have provided a fast, practical way for people in Korea to respond to digital sex crimes themselves. We will continue contributing through AI safety research and technology development to address the harmful effects of AI in society.”