Kakao Brain Unveils GAN-Based 'Nickface' Technology: Improved Face Shape Transformation Performance Compared to FaceShifter

Kakao Brain announced that it has developed a GAN-based Face Swapping model for 'Nickface' technology, which replaces faces with non-existent virtual person IDs. This model enables swapping with various IDs without additional training and meets three essential conditions: unique ID generation and processing of virtual person IDs. In comparison with the existing state-of-the-art model FaceShifter, Kakao Brain's model demonstrated superiority in face shape transformation performance by improving the Reference shape difference by 19.3% and reducing the ID shape difference by 23.0%. Additionally, by applying a method that directly copies and pastes the background from the Reference image rather than generating it directly via the Generator, the model enhanced background fidelity and image quality. Future plans include improving generation performance in profile views and occlusion scenarios, as well as developing ID generation features that reflect user intent.