This paper proposes the Deep Implied Volatility Factor Model for estimating the daily IV surface of individual stock options. The model combines neural networks to learn latent functions with linear regression for daily factor loadings, balancing flexibility and interpretability. Incorporating time-to-earnings-announcement information improves performance around such announcements, capturing key surface dynamics and shifts in the risk-neutral density. The framework enables consistent derivative pricing and computation of stock-level VIX-style indices, closely tracking official benchmarks and extending volatility estimates to periods with sparse data. Results for Apple and five additional stocks demonstrate the model’s robustness and scalability.