One of the hardest hurdles for AI is handling "occlusions"—when an object (like a hand, a microphone, or hair) passes in front of the face. Earlier builds often resulted in the hand disappearing or the face "popping" over the object. The 1.2.0 model includes a refined mask predictor that can distinguish between the face and foreground objects, allowing hands to pass naturally in front of the swapped face without artifacts.
Previous iterations of the software often capped output resolution at 480p or 720p, limiting their use for professional projects like YouTube videos or short films. AI Video FaceSwap 1.2.0 integrates an internal super-resolution model. This means the AI performs the swap at a standard resolution and then upscales it to 1080p or even 4K, sharpening facial details and restoring skin texture that might otherwise look plastic or blurred. AI Video FaceSwap 1.2.0