This release marks the first stage of an iterative process. Future updates will continue to expand the dataset with outputs from additional generative models, ensuring broader coverage, higher variability, and improved robustness. This multi-model design strengthens its value for age verification, deepfake detection and prevention, and fairness research

This data is relevant because: 

• Model coverage. We independently research and integrate a wide range of face-generation/deepfake models, and we also propose our own pool of models to maximize coverage of approaches and styles

• Prompt strategies. For each model, we use diverse prompts and modes and systematically sweep parameters to extract maximum variability and quality

• Freshness. We operate on the latest stable model versions (as of August 2025) and update the pipeline as new releases appear, keeping the dataset current

The dataset is composed of multiple complementary subsets, each created with a different generative model. While all subsets span the three age bins (5–9, 9–12, 12–16), they contribute distinct strengths: some emphasize broad stylistic and demographic variety, while others focus on photorealism, consistency of facial proportions, and fine-grained details such as lighting and texture. By integrating outputs from several models, the dataset achieves both diversity and realism, reducing the risk of single-model bias