Facial forgery and manipulation¶
Facial forgery encompasses techniques for creating or modifying facial images and videos to produce convincing synthetic faces. This includes face swapping (transferring one person's face onto another's body), face reenactment (using one person's expressions to animate another's face), and entire face synthesis (generating completely artificial faces from scratch). These techniques are driven by advances in generative models (GANs, diffusion models) and are central to both the deepfake threat landscape and synthetic media forensics research.
Key manipulation types¶
Face Swapping: Replacing one person's face with another in image or video. Examples include FaceSwap and Deepfakes using encoder-decoder architectures and adversarial learning.
Face Reenactment: Transferring facial expressions and head pose from a source to a target face, keeping identity intact. Used in Face2Face and neural rendering approaches.
Entire Face Synthesis: Generating realistic facial images from scratch using GANs or diffusion models. StyleGAN and other architectures produce high-fidelity synthetic faces.
Face Editing: Selective manipulation of facial attributes (age, expression, lighting) without full synthesis. Enables targeted modifications while preserving identity.
Key papers¶
- DeepfakeGenome: Toward Next-Generation Deepfake Attribution — Large-scale benchmark (100 algorithms, 2M images) for deepfake detection and attribution; proposes retrieval-based evaluation protocols for real-world active defense scenarios.
Related topics¶
- Deepfake Detection (identifying manipulated faces)
- Deepfakes (synthetic facial media)
- Synthetic media (broader category of AI-generated content)
- Generative Models (GANs, diffusion models, transformers for generation)