Abstract
Facial aging is a complex process, highly dependent on multiple factors like gender, ethnicity, lifestyle, etc., making it extremely challenging to learn a global aging prior to predict aging for any individual accurately. Existing techniques often produce realistic and plausible aging results, but the re-aged images often do not resemble the person's appearance at the target age and thus need personalization. In many practical applications of virtual aging, e.g. VFX in movies and TV shows, access to a personal photo collection of the user depicting aging in a small time interval (20-40 years) is often available. However, naive attempts to personalize global aging techniques on personal photo collections often fail. Thus, we propose MyTimeMachine (MyTM), which combines a global aging prior with a personal photo collection (using as few as 50 images) to learn a personalized age transformation. We introduce a novel Adapter Network that combines personalized aging features with global aging features and generates a re-aged image with StyleGAN2. We also introduce three loss functions to personalize the Adapter Network with personalized aging loss, extrapolation regularization, and adaptive w-norm regularization. Our approach can also be extended to videos, achieving high-quality, identity-preserving, and temporally consistent aging effects that resemble actual appearances at target ages, demonstrating its superiority over state-of-the-art approaches.
Introduction
Facial aging is a highly individualized process influenced by genetics, ethnicity, and lifestyle. While existing "global" aging models like SAM or CUSP produce realistic older or younger faces, they often fail to capture the specific way a particular individual ages, tending instead toward an "average" aged face. The primary challenge in personalizing these models is data scarcity; high-quality longitudinal photos spanning decades are rare. Naive fine-tuning on small, biased personal collections often leads to catastrophic overfitting, where the model loses its ability to generalize to new poses or ages outside the training set.
MyTimeMachine (MyTM) is a novel framework that moves beyond generic aging by augmenting a global aging prior with a small collection of an individual's personal photos (as few as 10–50). This system generates re-aged images that maintain high identity fidelity and temporal consistency, outperforming state-of-the-art global and diffusion-based models.
Methodology
MyTM builds upon the SAM (Style-based Age Manipulation) architecture, utilizing a pre-trained StyleGAN2 backbone. The core innovation is a lightweight Adapter Network positioned between the global age encoder and the GAN decoder.
Instead of replacing global knowledge, the Adapter Network learns a "personalized offset" ($\Delta W^+_{tgt}$) in the StyleGAN latent space, nudging the global aging trajectory to match unique individual characteristics. To train this effectively on limited data, three loss functions are introduced:
- Personalized Aging Loss: Uses a facial recognition network (ArcFace) to ensure identity features match reference images at similar target ages.
- Extrapolation Regularization: Anchors the model to the reliable global SAM output for ages not seen in personal photos, preventing nonsensical results at extreme ages.
- Adaptive w-norm Regularization: Manages the "inversion-editability trade-off" by allowing the latent code to move toward the distribution center as target ages diverge, facilitating necessary structural changes like skin sagging or bone growth.
Datasets and Evaluation
The research utilized a new longitudinal facial aging dataset featuring 12 celebrities with approximately 50 images each, spanning 20–40 years. This includes well-known figures such as Queen Elizabeth II and Al Pacino, whose aging processes are well-documented.
MyTM demonstrated superior performance in both age regression and progression. In identity preservation tasks, it achieved an 11.7% improvement in identity similarity ($ID_{sim}$) over the previous best model, FADING. It also achieved a significantly lower Mean Absolute Error ($Age_{MAE}$) compared to personalized diffusion baselines (e.g., 7.7 vs. 25.9 for FADING + Dreambooth). User preference studies consistently favored MyTM for its resemblance to the actual person at the target age. Furthermore, the video pipeline is highly efficient, completing in under 5 minutes on a single A6000 GPU, compared to hours for competing methods.
Conclusion
MyTimeMachine represents a significant step forward in personalized facial transformation, offering high-quality, identity-preserving results suitable for VFX and other creative industries. While limitations exist regarding accessories like glasses and hair color transitions at extreme ages, the "nudge" strategy via an Adapter Network proves highly effective for personalizing global priors with minimal data. Future work will focus on improving robustness for diverse accessories and hair editing modules.
Summary updated